SLEEP

Fox, Henrik; remedē System Pivotal Trial Study Group; Oldenburg, Olaf; remedē System Pivotal Trial Study Group; Javaheri, Shahrokh; remedē System Pivotal Trial Study Group; Ponikowski, Piotr; remedē System Pivotal Trial Study Group; Augostini, Ralph; remedē System Pivotal Trial Study Group; Goldberg, Lee R; remedē System Pivotal Trial Study Group; Stellbrink, Christoph; remedē System Pivotal Trial Study Group; Mckane, Scott; remedē System Pivotal Trial Study Group; Meyer, Timothy E; remedē System Pivotal Trial Study Group; Abraham, William T; remedē System Pivotal Trial Study Group; Costanzo, Maria Rosa; remedē System Pivotal Trial Study Group

doi: 10.1093/sleep/zsz158pmid: 31634407

Abstract Study Objective To evaluate long-term efficacy and safety of phrenic nerve stimulation (PNS) in patients with moderate-to-severe central sleep apnea (CSA) through 3 years of therapy. Methods Patients in the remedē System Pivotal Trial were observed every 3 months after implant until US Food and Drug Administration approval. At the time of approval and study closure, all patients completed 24 months of follow-up; 33 patients had not reached the 36-month visit. Sleep metrics (polysomnography) and echocardiographic parameters are reported at baseline, 12, 18, and 24 months, in addition to available 36-month sleep results from polygraphy. Safety was assessed through 36 months; however, analysis focused through 24 months and available 36-month results are provided. Results Patients were assessed at 24 (n = 109) and 36 (n = 60) months. Baseline characteristics included mean age 64 years, 91% male, and mean apnea–hypopnea index 47 events per hour. Sleep metrics (apnea–hypopnea index (AHI), central apnea index, arousal index, oxygen desaturation index, rapid eye movement sleep) remained improved through 24 and 36 months with continuous use of PNS therapy. At least 60% of patients in the treatment group achieved at least 50% reduction in AHI through 24 months. Serious adverse events (SAEs) related to the remedē System implant procedure, device, or therapy through 24 months were reported by 10% of patients, no unanticipated adverse device effects or deaths, and all events resolved. No additional related SAEs were reported between 24 and 36 months. Conclusion These data suggest beneficial effects of long-term PNS in patients with CSA appear to sustain through 36 months with no new safety concerns. Trial Registration NCT01816776. central sleep apnea, phrenic nerve stimulation, transvenous stimulation Statement of Significance Central sleep apnea (CSA) is a highly common comorbidity in heart failure (HF) patients, but also affects sleep and quality of life in non-HF patients. CSA also results in alterations in autonomic, chemical, rhythmic, and inflammatory processes affecting overall individual health and is associated with increased morbidity and mortality. Thus, therapeutic options to treat CSA are needed. Phrenic nerve stimulation (PNS) is a recent US Food and Drug Administration-approved technology shown to be safe and effective in adult patients for the treatment of moderate-to-severe CSA through 12 months. The importance of the current analysis is that safety and effectiveness of PNS therapy is extended out through 36 months with no unanticipated adverse device effects related to the device or delivered therapy. Introduction Epidemiological and interventional studies have shown that adverse health outcomes are associated with sleep-disordered breathing (SDB) [1]. SDB is typically classified into two predominant types: obstructive sleep apnea (OSA) and central sleep apnea (CSA). Regardless of concomitant pathology, patients may have mixed OSA and CSA, but one or the other type of SDB is generally predominant [2]. CSA is frequent in patients with underlying cardiovascular diseases and especially with heart failure (HF) [3, 4]. Neurophysiologically, CSA is a manifestation of central breathing instability with transient inhibition of ventilatory motor [5]. Depending on the underlying cause of the CSA, treatment options vary. These include positive pressure devices, pharmacological therapy, and phrenic nerve stimulation (PNS) [5]. Patients with CSA have limited treatment options because mask-based therapies are associated with poor patient adherence, mask and pressure intolerance and, more recently, a contraindication for some devices (adaptive servo-ventilation) in patients with decreased cardiac systolic function. Independent of what therapeutic option is used, the main goal of the treatment is to eliminate abnormal respiratory events and stabilize sleep infrastructure with the hope to eliminate CSA-associated symptoms and improve quality of life [5]. Given various causes and pathophysiological conditions associated with CSA, long-term randomized controlled trials are needed to ascertain the long-term effectiveness of individual treatment modalities [5]. So far such studies are scarce. The recent randomized Treatment of Sleep-Disordered Breathing with Predominant Central Sleep Apnea by Adaptive Servo-Ventilation in Patients with Heart Failure (SERVE-HF) study demonstrated an increase in mortality in the adaptive servo-ventilation treatment group compared to control in patients with reduced ejection fraction and predominant CSA [6]; thus, novel therapeutic approaches are critically needed [7, 8]. Transvenous PNS is a promising therapy shown to be safe and to effectively treat events and symptoms of CSA [9]. Improvements in sleep indices, sleep quality, and quality of life observed in the remedē System Pivotal Trial were sustained through 12 months of active therapy in patients with and without HF [10]. The remedē System (Respicardia, Inc., Minnetonka, MN) is a fully implantable neurostimulation system that delivers PNS continuously throughout the night to treat CSA, mitigating therapy compliance issues that often occur with mask-based therapies [11]. Despite the sustained benefit and safety of PNS through 12 months, longer-term data are essential because the disorders underlying CSA are typically chronic and progressive. Therefore, the objective of this paper was to show both therapeutic effectiveness measures as well as safety profile indicators through 24 and 36 months of active remedē System therapy. Methods Design, methods, oversight, primary results as well as 12-month and HF results of the remedē System Pivotal Trial (NCT01816776) have previously been reported [9, 10]. Briefly, the remedē System Pivotal Trial was a prospective, multicenter, randomized, open-label controlled trial in patients with predominant CSA of different etiologies to assess transvenous unilateral PNS versus no stimulation [9]. Patients in the control group had the device implanted at time of randomization but not activated until 6-month effectiveness endpoints were assessed, ending the randomized portion of the trial; this cohort is referred to as former control once therapy was activated. Patients were observed every 3 months post implant until US Food and Drug Administration (FDA) approval of the remedē System (October 2017). As a result, the patients analyzed here have different durations of active therapy. All patients remaining in the trial at study closure had been followed for a minimum of 24 months after implant, but 33 patients had not yet reached the 36-month visit when the study was closed. In agreement with FDA, ongoing patients were asked to enroll into the remedē System Post Approval Study (NCT03425188). Effectiveness of PNS was assessed by full overnight, attended polysomnography (PSG) completed at baseline and at 6-month intervals through 24 months, adding a final home sleep test of cardiorespiratory polygraphy (PG) using a NOX-T3 apparatus (ResMed Corp., San Diego, CA) at 36 months. Typical sleep measurements were analyzed including the apnea–hypopnea index (AHI) and its components (components central apnea index [CAI], obstructive apnea index [OAI], mixed apnea index [MAI], hypopnea index [HI]), oxygen desaturation ≥4% index (ODI4), arousal index (ArI), and sleep stages in PSG. A PSG/PG core laboratory (Registered Sleepers, Leicester, NC) performed all scoring and was blinded to randomization assignment. Standard criteria were used to score sleep studies as detailed previously Costanzo Lancet [9]. However, hypopneas were not scored as central versus obstructive as part of this trial. Echocardiograms were performed at the time of sleep studies through 24 months and were interpreted by a core laboratory blinded to randomization assignment (United Heart and Vascular Center, St. Paul, MN). The former control patients, due to their 6-month delay of active therapy, had their last PSG and echocardiogram at 18 months of active therapy. The protocol was approved by local ethics or institutional review boards; all patients provided written informed consent. The study was conducted in accordance with the Declaration of Helsinki, Good Clinical Practice, and ISO 14155:2011 and registered at ClinicalTrials.gov (Identifier: NCT01816776). Statistical analysis Observed effectiveness results are presented separately for each group (treatment and former control) based on months since therapy activation using the predefined per protocol population. Owing to the exploratory nature of this analysis, statistical test results for change from baseline are not included and imputation was not performed for missing data. Multivariate analysis to find subgroups that may respond better or worse included identification of covariates that may have a univariate association with achieving at least 50% reduction in AHI from baseline to 24 months. Covariates with univariate p value less than 0.2 from logistic regression were included in a backward selection multivariate logistic regression model to determine a final model (p value of 0.1 used to retain variables in the model). The analysis was repeated using pooled data from the treatment and former control patients (after having therapy turned on) following 18 months of active therapy in order to have a larger sample size. Safety, reported as freedom from related serious adverse events (SAEs) through 24 months, was summarized as a binomial proportion. Overall survival in the pooled group of treatment and former control patients who had therapy initiated was assessed using the Kaplan–Meier method, censoring patients who did not die at the date of study exit. Deaths were also summarized by modality (pump failure, sudden cardiac death, other cardiovascular and noncardiovascular), as adjudicated by an independent Clinical Events Committee (CEC). The Kaplan–Meier method was used to estimate median time to battery depletion. SAS, version 9.4 (Cary, NC) was used for all analyses. Results At the time of the Pivotal Trial closure, the original 151 patients had been followed for 32 ± 13 months (median = 35, maximum = 52 months) and 94 patients were ongoing at the time of trial closure. All subjects remaining in the trial at the time of closure had completed a minimum of 24 months of follow-up; however, 33 patients had not yet reached the 36-month visit. Reasons for patient exclusion or withdrawal from analysis (e.g. device, therapy, or implant related) at each visit are displayed in the CONSORT diagram Figure 1. Patient demographics for treatment and control patients at baseline as well as for pooled patients at both 24 months (n = 109) and 36 months (n = 60) are reported in Table 1. Patients with HF were on optimal guideline recommended therapy before enrollment and throughout the duration of follow-up [9]. Patient scheduled site follow-up visits occurred every 3 months per the trial design and additionally as needed to titrate therapy. Table 1. Baseline characteristics (per protocol population) Baseline characteristics . Treatment (n = 58) . Control (n = 73) . Pooled 24 months* (n = 109) . Pooled 36 months† (n = 60) . Age (years) 63 ± 12 64 ± 14 64 ± 13 64 ± 13 Men 51 (88%) 68 (93%) 99 (91%) 52 (87%) White 57 (98%) 69 (95%) 106 (97%) 58 (97%) Body mass index (kg/m2) 30 ± 5 31 ± 7 31 ± 6 32 ± 6 Neck circumference (cm) 42 ± 5 43 ± 5 43 ± 5 42 ± 5 Heart rate (beats per minute) 76 ± 13 73 ± 14 75 ± 14 76 ± 15 Systolic blood pressure (mm Hg) 124 ± 18 124 ± 18 126 ± 18 126 ± 18 Diastolic blood pressure (mm Hg) 74 ± 11 75 ± 12 76 ± 11 76 ± 10 Respiration rate (breaths per minute) 18 ± 3 17 ± 3 17 ± 3 17 ± 3 AHI (events/hour) 50 ± 19 44 ± 17 47 ± 19 48 ± 19 CAI (events/hour) 32 ± 19 26 ± 16 29 ± 17 29 ± 17 OAI (events/hour) 2 ± 2 2 ± 3 2 ± 3 3 ± 3 MAI (events/hour) 3 ± 4 2 ± 3 3 ± 4 3 ± 4 HI (events/hour) 13 ± 11 13 ± 12 13 ± 12 15 ± 13 Percent of sleep with oxygen saturation <90% (%) 17 ± 18 11 ± 12 14 ± 16 16 ± 18 Oxygen desaturation ≥4% index (events/hour) 44 ± 22 37 ± 18 41 ± 20 42 ± 20 Epworth Sleepiness Scale (points) 11 ± 5 9 ± 6 10 ± 6 10 ± 5 Atrial fibrillation 22 (38%) 29 (40%) 39 (36%) 19 (32%) LVEF ≤ 45% 32/57 (56%) 42/70 (60%) 59/106 (56%) 34/60 (57%) HF 35 (60%) 45 (62%) 62 (57%) 32 (53%) New York Heart Association class  I 5 (9%) 12 (16%) 15 (14%) 11 (18%)  II 14 (24%) 20 (27%) 27 (25%) 16 (27%)  III 16 (28%) 13 (18%) 20 (18%) 5 (8%)  IV 0% 0% 0% 0% Coronary artery disease 33 (57%) 42 (58%) 61 (56%) 33 (55%) Hypertension 42 (72%) 55 (75%) 79 (72%) 41 (68%) Diabetes mellitus 20 (34%) 17 (23%) 31 (28%) 18 (30%) Prior stroke 4 (7%) 5 (7%) 7 (6%) 4 (7%) Renal impairment 11 (19%) 20 (27%) 23 (21%) 9 (15%) Concomitant cardiac devices 24 (41%) 30 (41%) 40 (37%) 20 (33%) Implantable cardioverter defibrillator 14 (24%) 13 (18%) 20 (18%) 12 (20%) Cardiac resynchronization therapy defibrillator 8 (14%) 9 (12%) 11 (10%) 4 (6%) Noncardiac resynchronization therapy pacemaker 2 (3%) 8 (11%) 9 (8%) 4 (6%) Medications Angiotensin-converting enzyme inhibitor 28 (48%) 35 (48%) 52 (48%) 29 (48%) Angiotensin receptor blocker 9 (16%) 13 (18%) 17 (16%) 8 (13%) Aldosterone-blocking agent 25 (43%) 17 (23%) 30 (28%) 12 (20%) Beta-blocker 36 (62%) 47 (64%) 66 (61%) 36 (60%) Loop diuretic 26 (45%) 26 (36%) 40 (37%) 18 (30%) Thiazide diuretic 15 (26%) 16 (22%) 23 (21%) 8 (13%) Thiazide-like diuretic 5 (9%) 2 (3%) 6% (6) 4 (7%) Antiarrhythmic 4 (7%) 7 (10%) 8 (7%) 2 (3%) Digoxin 10 (17%) 12 (16%) 17 (16%) 8 (13%) Baseline characteristics . Treatment (n = 58) . Control (n = 73) . Pooled 24 months* (n = 109) . Pooled 36 months† (n = 60) . Age (years) 63 ± 12 64 ± 14 64 ± 13 64 ± 13 Men 51 (88%) 68 (93%) 99 (91%) 52 (87%) White 57 (98%) 69 (95%) 106 (97%) 58 (97%) Body mass index (kg/m2) 30 ± 5 31 ± 7 31 ± 6 32 ± 6 Neck circumference (cm) 42 ± 5 43 ± 5 43 ± 5 42 ± 5 Heart rate (beats per minute) 76 ± 13 73 ± 14 75 ± 14 76 ± 15 Systolic blood pressure (mm Hg) 124 ± 18 124 ± 18 126 ± 18 126 ± 18 Diastolic blood pressure (mm Hg) 74 ± 11 75 ± 12 76 ± 11 76 ± 10 Respiration rate (breaths per minute) 18 ± 3 17 ± 3 17 ± 3 17 ± 3 AHI (events/hour) 50 ± 19 44 ± 17 47 ± 19 48 ± 19 CAI (events/hour) 32 ± 19 26 ± 16 29 ± 17 29 ± 17 OAI (events/hour) 2 ± 2 2 ± 3 2 ± 3 3 ± 3 MAI (events/hour) 3 ± 4 2 ± 3 3 ± 4 3 ± 4 HI (events/hour) 13 ± 11 13 ± 12 13 ± 12 15 ± 13 Percent of sleep with oxygen saturation <90% (%) 17 ± 18 11 ± 12 14 ± 16 16 ± 18 Oxygen desaturation ≥4% index (events/hour) 44 ± 22 37 ± 18 41 ± 20 42 ± 20 Epworth Sleepiness Scale (points) 11 ± 5 9 ± 6 10 ± 6 10 ± 5 Atrial fibrillation 22 (38%) 29 (40%) 39 (36%) 19 (32%) LVEF ≤ 45% 32/57 (56%) 42/70 (60%) 59/106 (56%) 34/60 (57%) HF 35 (60%) 45 (62%) 62 (57%) 32 (53%) New York Heart Association class  I 5 (9%) 12 (16%) 15 (14%) 11 (18%)  II 14 (24%) 20 (27%) 27 (25%) 16 (27%)  III 16 (28%) 13 (18%) 20 (18%) 5 (8%)  IV 0% 0% 0% 0% Coronary artery disease 33 (57%) 42 (58%) 61 (56%) 33 (55%) Hypertension 42 (72%) 55 (75%) 79 (72%) 41 (68%) Diabetes mellitus 20 (34%) 17 (23%) 31 (28%) 18 (30%) Prior stroke 4 (7%) 5 (7%) 7 (6%) 4 (7%) Renal impairment 11 (19%) 20 (27%) 23 (21%) 9 (15%) Concomitant cardiac devices 24 (41%) 30 (41%) 40 (37%) 20 (33%) Implantable cardioverter defibrillator 14 (24%) 13 (18%) 20 (18%) 12 (20%) Cardiac resynchronization therapy defibrillator 8 (14%) 9 (12%) 11 (10%) 4 (6%) Noncardiac resynchronization therapy pacemaker 2 (3%) 8 (11%) 9 (8%) 4 (6%) Medications Angiotensin-converting enzyme inhibitor 28 (48%) 35 (48%) 52 (48%) 29 (48%) Angiotensin receptor blocker 9 (16%) 13 (18%) 17 (16%) 8 (13%) Aldosterone-blocking agent 25 (43%) 17 (23%) 30 (28%) 12 (20%) Beta-blocker 36 (62%) 47 (64%) 66 (61%) 36 (60%) Loop diuretic 26 (45%) 26 (36%) 40 (37%) 18 (30%) Thiazide diuretic 15 (26%) 16 (22%) 23 (21%) 8 (13%) Thiazide-like diuretic 5 (9%) 2 (3%) 6% (6) 4 (7%) Antiarrhythmic 4 (7%) 7 (10%) 8 (7%) 2 (3%) Digoxin 10 (17%) 12 (16%) 17 (16%) 8 (13%) Continuous variables reported as mean ± SD and categorical display n (percent). *24-month visit represents 24 months of therapy for treatment and 18 for former control. †The 36-month visit represents 36 months of therapy for treatment and 30 months for former control; 33 subjects remaining in the trial at time of FDA approval and study closure had not reached the 36-month visit. Open in new tab Table 1. Baseline characteristics (per protocol population) Baseline characteristics . Treatment (n = 58) . Control (n = 73) . Pooled 24 months* (n = 109) . Pooled 36 months† (n = 60) . Age (years) 63 ± 12 64 ± 14 64 ± 13 64 ± 13 Men 51 (88%) 68 (93%) 99 (91%) 52 (87%) White 57 (98%) 69 (95%) 106 (97%) 58 (97%) Body mass index (kg/m2) 30 ± 5 31 ± 7 31 ± 6 32 ± 6 Neck circumference (cm) 42 ± 5 43 ± 5 43 ± 5 42 ± 5 Heart rate (beats per minute) 76 ± 13 73 ± 14 75 ± 14 76 ± 15 Systolic blood pressure (mm Hg) 124 ± 18 124 ± 18 126 ± 18 126 ± 18 Diastolic blood pressure (mm Hg) 74 ± 11 75 ± 12 76 ± 11 76 ± 10 Respiration rate (breaths per minute) 18 ± 3 17 ± 3 17 ± 3 17 ± 3 AHI (events/hour) 50 ± 19 44 ± 17 47 ± 19 48 ± 19 CAI (events/hour) 32 ± 19 26 ± 16 29 ± 17 29 ± 17 OAI (events/hour) 2 ± 2 2 ± 3 2 ± 3 3 ± 3 MAI (events/hour) 3 ± 4 2 ± 3 3 ± 4 3 ± 4 HI (events/hour) 13 ± 11 13 ± 12 13 ± 12 15 ± 13 Percent of sleep with oxygen saturation <90% (%) 17 ± 18 11 ± 12 14 ± 16 16 ± 18 Oxygen desaturation ≥4% index (events/hour) 44 ± 22 37 ± 18 41 ± 20 42 ± 20 Epworth Sleepiness Scale (points) 11 ± 5 9 ± 6 10 ± 6 10 ± 5 Atrial fibrillation 22 (38%) 29 (40%) 39 (36%) 19 (32%) LVEF ≤ 45% 32/57 (56%) 42/70 (60%) 59/106 (56%) 34/60 (57%) HF 35 (60%) 45 (62%) 62 (57%) 32 (53%) New York Heart Association class  I 5 (9%) 12 (16%) 15 (14%) 11 (18%)  II 14 (24%) 20 (27%) 27 (25%) 16 (27%)  III 16 (28%) 13 (18%) 20 (18%) 5 (8%)  IV 0% 0% 0% 0% Coronary artery disease 33 (57%) 42 (58%) 61 (56%) 33 (55%) Hypertension 42 (72%) 55 (75%) 79 (72%) 41 (68%) Diabetes mellitus 20 (34%) 17 (23%) 31 (28%) 18 (30%) Prior stroke 4 (7%) 5 (7%) 7 (6%) 4 (7%) Renal impairment 11 (19%) 20 (27%) 23 (21%) 9 (15%) Concomitant cardiac devices 24 (41%) 30 (41%) 40 (37%) 20 (33%) Implantable cardioverter defibrillator 14 (24%) 13 (18%) 20 (18%) 12 (20%) Cardiac resynchronization therapy defibrillator 8 (14%) 9 (12%) 11 (10%) 4 (6%) Noncardiac resynchronization therapy pacemaker 2 (3%) 8 (11%) 9 (8%) 4 (6%) Medications Angiotensin-converting enzyme inhibitor 28 (48%) 35 (48%) 52 (48%) 29 (48%) Angiotensin receptor blocker 9 (16%) 13 (18%) 17 (16%) 8 (13%) Aldosterone-blocking agent 25 (43%) 17 (23%) 30 (28%) 12 (20%) Beta-blocker 36 (62%) 47 (64%) 66 (61%) 36 (60%) Loop diuretic 26 (45%) 26 (36%) 40 (37%) 18 (30%) Thiazide diuretic 15 (26%) 16 (22%) 23 (21%) 8 (13%) Thiazide-like diuretic 5 (9%) 2 (3%) 6% (6) 4 (7%) Antiarrhythmic 4 (7%) 7 (10%) 8 (7%) 2 (3%) Digoxin 10 (17%) 12 (16%) 17 (16%) 8 (13%) Baseline characteristics . Treatment (n = 58) . Control (n = 73) . Pooled 24 months* (n = 109) . Pooled 36 months† (n = 60) . Age (years) 63 ± 12 64 ± 14 64 ± 13 64 ± 13 Men 51 (88%) 68 (93%) 99 (91%) 52 (87%) White 57 (98%) 69 (95%) 106 (97%) 58 (97%) Body mass index (kg/m2) 30 ± 5 31 ± 7 31 ± 6 32 ± 6 Neck circumference (cm) 42 ± 5 43 ± 5 43 ± 5 42 ± 5 Heart rate (beats per minute) 76 ± 13 73 ± 14 75 ± 14 76 ± 15 Systolic blood pressure (mm Hg) 124 ± 18 124 ± 18 126 ± 18 126 ± 18 Diastolic blood pressure (mm Hg) 74 ± 11 75 ± 12 76 ± 11 76 ± 10 Respiration rate (breaths per minute) 18 ± 3 17 ± 3 17 ± 3 17 ± 3 AHI (events/hour) 50 ± 19 44 ± 17 47 ± 19 48 ± 19 CAI (events/hour) 32 ± 19 26 ± 16 29 ± 17 29 ± 17 OAI (events/hour) 2 ± 2 2 ± 3 2 ± 3 3 ± 3 MAI (events/hour) 3 ± 4 2 ± 3 3 ± 4 3 ± 4 HI (events/hour) 13 ± 11 13 ± 12 13 ± 12 15 ± 13 Percent of sleep with oxygen saturation <90% (%) 17 ± 18 11 ± 12 14 ± 16 16 ± 18 Oxygen desaturation ≥4% index (events/hour) 44 ± 22 37 ± 18 41 ± 20 42 ± 20 Epworth Sleepiness Scale (points) 11 ± 5 9 ± 6 10 ± 6 10 ± 5 Atrial fibrillation 22 (38%) 29 (40%) 39 (36%) 19 (32%) LVEF ≤ 45% 32/57 (56%) 42/70 (60%) 59/106 (56%) 34/60 (57%) HF 35 (60%) 45 (62%) 62 (57%) 32 (53%) New York Heart Association class  I 5 (9%) 12 (16%) 15 (14%) 11 (18%)  II 14 (24%) 20 (27%) 27 (25%) 16 (27%)  III 16 (28%) 13 (18%) 20 (18%) 5 (8%)  IV 0% 0% 0% 0% Coronary artery disease 33 (57%) 42 (58%) 61 (56%) 33 (55%) Hypertension 42 (72%) 55 (75%) 79 (72%) 41 (68%) Diabetes mellitus 20 (34%) 17 (23%) 31 (28%) 18 (30%) Prior stroke 4 (7%) 5 (7%) 7 (6%) 4 (7%) Renal impairment 11 (19%) 20 (27%) 23 (21%) 9 (15%) Concomitant cardiac devices 24 (41%) 30 (41%) 40 (37%) 20 (33%) Implantable cardioverter defibrillator 14 (24%) 13 (18%) 20 (18%) 12 (20%) Cardiac resynchronization therapy defibrillator 8 (14%) 9 (12%) 11 (10%) 4 (6%) Noncardiac resynchronization therapy pacemaker 2 (3%) 8 (11%) 9 (8%) 4 (6%) Medications Angiotensin-converting enzyme inhibitor 28 (48%) 35 (48%) 52 (48%) 29 (48%) Angiotensin receptor blocker 9 (16%) 13 (18%) 17 (16%) 8 (13%) Aldosterone-blocking agent 25 (43%) 17 (23%) 30 (28%) 12 (20%) Beta-blocker 36 (62%) 47 (64%) 66 (61%) 36 (60%) Loop diuretic 26 (45%) 26 (36%) 40 (37%) 18 (30%) Thiazide diuretic 15 (26%) 16 (22%) 23 (21%) 8 (13%) Thiazide-like diuretic 5 (9%) 2 (3%) 6% (6) 4 (7%) Antiarrhythmic 4 (7%) 7 (10%) 8 (7%) 2 (3%) Digoxin 10 (17%) 12 (16%) 17 (16%) 8 (13%) Continuous variables reported as mean ± SD and categorical display n (percent). *24-month visit represents 24 months of therapy for treatment and 18 for former control. †The 36-month visit represents 36 months of therapy for treatment and 30 months for former control; 33 subjects remaining in the trial at time of FDA approval and study closure had not reached the 36-month visit. Open in new tab Figure 1. Open in new tabDownload slide CONSORT diagram. Composition of the per protocol population with sleep study data through the 36-month post-therapy initiation visit. Figure 1. Open in new tabDownload slide CONSORT diagram. Composition of the per protocol population with sleep study data through the 36-month post-therapy initiation visit. Sleep metrics The primary effectiveness endpoint in the Pivotal Trial was a 50% or greater reduction in AHI from baseline. In the per protocol population, 60% of treatment group patients achieved this target at 6 months (11% in untreated control) and effectiveness remained high through 24 months of therapy, with 67%, 64%, and 60% at 12 (n = 54), 18 (n = 45), and 24 (n = 42) months (Figure 2). Furthermore, 93% of treatment group patients (39/42) had a reduction in AHI at the 24-month visit (Figure 3, panel A). Notably, these results were maintained with minimal additional titration visits in which programming changes were made (mean <1 programming change, median = 0) for each 6-month interval after the 6-month visit. Figure 2. Open in new tabDownload slide Proportion of treatment group patients that achieved a ≥50% reduction in AHI at 6, 12, 18, and 24-months post-therapy initiation. Figure 2. Open in new tabDownload slide Proportion of treatment group patients that achieved a ≥50% reduction in AHI at 6, 12, 18, and 24-months post-therapy initiation. Figure 3. Open in new tabDownload slide Percentage change in AHI after 24 months of therapy for each treatment group patient (panel A) and 18 months of active therapy for each former control group patient (panel B). Panel A: The improvements represent the percentage change in AHI from baseline to 24 months of active therapy for each patient in the treatment group. Patients with any decrease in AHI from baseline are shown in green bars and patients with any increase in AHI from baseline are shown in red bars. Panel B: The improvements represent the percentage change in AHI from baseline (prior to therapy activation) to 18 months of active therapy for each patient in the former control group. Patients with any decrease in AHI from baseline are shown in green bars and patients with any increase in AHI from baseline are shown in red bars. Figure 3. Open in new tabDownload slide Percentage change in AHI after 24 months of therapy for each treatment group patient (panel A) and 18 months of active therapy for each former control group patient (panel B). Panel A: The improvements represent the percentage change in AHI from baseline to 24 months of active therapy for each patient in the treatment group. Patients with any decrease in AHI from baseline are shown in green bars and patients with any increase in AHI from baseline are shown in red bars. Panel B: The improvements represent the percentage change in AHI from baseline (prior to therapy activation) to 18 months of active therapy for each patient in the former control group. Patients with any decrease in AHI from baseline are shown in green bars and patients with any increase in AHI from baseline are shown in red bars. Multivariate analysis modeling of 24-month treatment group results retained three variables in the model to determine what characteristics may lead to therapy response (50% reduction in AHI): body mass index (BMI), patient height, and HI. The analysis suggests patients may be more likely to respond to PNS with lower BMI and higher HI. Repeating the multivariate analysis using the pooled treatment and former control patients following 18 months of active therapy also retained BMI as the most significant predictor. Other variables retained were sex (female more likely to respond) and oxygen desaturation ≥4% index (ODI4; higher ODI4 more likely to respond), however, only 10 females had data of which 9 experienced 50% AHI reduction compared to 53% (51/96) of males. Analyses to determine subgroups that may be more or less likely to respond to therapy did not yield definitive results. However, the data demonstrate that most subgroups seem to improve to varying degree following PNS. The forest plot displaying response rates at 24 months of therapy for various subgroups (i.e. subjects <65 and ≥65) indicates a majority of subgroups appear to have reasonable chance to have at least half of the patients achieve a 50% or greater reduction in AHI at 24 months (Supplementary Figure S1). Sleep metrics show consistent improvement throughout the follow-up period. Median events per hour AHI [interquartile range] at 12, 18, and 24 months were 15 [9, 31], 17 [10, 25], and 16 [7, 37], respectively (Figure 4 and Table 2). Median central apnea index (CAI) was less than or equal to 1.0 event per hour and median OAI was 2–3 events per hour at 12 months and beyond. ODI4 improved similarly to the AHI improvement [median 41 [30, 56] events per hour at baseline and 14 [7, 26], 17 [8, 23], and 13 [5, 34] at 12, 18, and 24 months, respectively]. One of the unique findings of the Pivotal Trial was that PNS additionally improved the ArI at 6 months and long-term data through 24 months suggest persistent improvements in ArI (median 41 [34, 64] events per hour at baseline and 20 [14, 32], 16 [12, 30], and 17 [9, 29] at 12, 18, and 24 months, respectively). Moreover, median percentage of rapid eye movement (REM) sleep improved from 10% at baseline to 14%, 15%, and 19% at each follow-up visit. In addition, the percentage of sleep with oxygen saturation less than 90% was approximately half compared to baseline levels (median 9%) at each subsequent visit (medians ranged from 3% to 5%). Table 2. Treatment group PSG results by visit (per protocol population) . Baseline (n = 58) . 6 months (n = 58) . 12 months (n = 54) . 18 months (n = 45) . 24 months (n = 42) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 60% (47% to 72%) 67% (53% to 78%) 64% (50% to 77%) 60% (44% to 73%) AHI (events/hour) 49 [35, 60] 21 [11, 35] 15 [9, 31] 17 [10, 25] 16 [7, 37] CAI (events/hour) 30 [16, 43] 1 [0, 7] 1 [0, 3] 1 [0, 3] 0 [0, 3] OAI (events/hour) 2 [1, 3] 4 [2, 9] 2 [1, 7] 2 [1, 6] 3 [0, 8] MAI (events/hour) 1 [0, 4] 0 [0, 0] 0 [0, 0] 0 [0, 1] 0 [0, 0] HI (events/hour) 13 [3, 20] 9 [3, 19] 8 [5, 16] 10 [5, 16] 8 [3, 19] ODI4 (events/hour) 41 [30, 56] 19 [8, 37] 14 [7, 26] 17 [8, 23] 13 [5, 34] ArI (events/hour) 41 [34, 64] 22 [16, 33] 20 [14, 32] 16 [12, 30] 17 [9, 29] Sleep with oxygen saturation <90% (%) 9 [5, 27] 5 [1, 20] 3 [1, 14] 4 [2, 12] 5 [1, 21] Sleep with oxygen saturation <90% (min) 33 [16, 87] 14 [2, 44] 8 [2, 26] 11 [5, 33] 15 [3, 53] Percent of sleep in REM (%) 10 [6, 15] 14 [5, 18] 14 [6, 21] 15 [9, 22] 19 [8, 23] . Baseline (n = 58) . 6 months (n = 58) . 12 months (n = 54) . 18 months (n = 45) . 24 months (n = 42) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 60% (47% to 72%) 67% (53% to 78%) 64% (50% to 77%) 60% (44% to 73%) AHI (events/hour) 49 [35, 60] 21 [11, 35] 15 [9, 31] 17 [10, 25] 16 [7, 37] CAI (events/hour) 30 [16, 43] 1 [0, 7] 1 [0, 3] 1 [0, 3] 0 [0, 3] OAI (events/hour) 2 [1, 3] 4 [2, 9] 2 [1, 7] 2 [1, 6] 3 [0, 8] MAI (events/hour) 1 [0, 4] 0 [0, 0] 0 [0, 0] 0 [0, 1] 0 [0, 0] HI (events/hour) 13 [3, 20] 9 [3, 19] 8 [5, 16] 10 [5, 16] 8 [3, 19] ODI4 (events/hour) 41 [30, 56] 19 [8, 37] 14 [7, 26] 17 [8, 23] 13 [5, 34] ArI (events/hour) 41 [34, 64] 22 [16, 33] 20 [14, 32] 16 [12, 30] 17 [9, 29] Sleep with oxygen saturation <90% (%) 9 [5, 27] 5 [1, 20] 3 [1, 14] 4 [2, 12] 5 [1, 21] Sleep with oxygen saturation <90% (min) 33 [16, 87] 14 [2, 44] 8 [2, 26] 11 [5, 33] 15 [3, 53] Percent of sleep in REM (%) 10 [6, 15] 14 [5, 18] 14 [6, 21] 15 [9, 22] 19 [8, 23] Continuous variables reported as median [interquartile range]. CI = confidence interval; N/A = not applicable; ODI4 = oxygen desaturation ≥4% index. Open in new tab Table 2. Treatment group PSG results by visit (per protocol population) . Baseline (n = 58) . 6 months (n = 58) . 12 months (n = 54) . 18 months (n = 45) . 24 months (n = 42) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 60% (47% to 72%) 67% (53% to 78%) 64% (50% to 77%) 60% (44% to 73%) AHI (events/hour) 49 [35, 60] 21 [11, 35] 15 [9, 31] 17 [10, 25] 16 [7, 37] CAI (events/hour) 30 [16, 43] 1 [0, 7] 1 [0, 3] 1 [0, 3] 0 [0, 3] OAI (events/hour) 2 [1, 3] 4 [2, 9] 2 [1, 7] 2 [1, 6] 3 [0, 8] MAI (events/hour) 1 [0, 4] 0 [0, 0] 0 [0, 0] 0 [0, 1] 0 [0, 0] HI (events/hour) 13 [3, 20] 9 [3, 19] 8 [5, 16] 10 [5, 16] 8 [3, 19] ODI4 (events/hour) 41 [30, 56] 19 [8, 37] 14 [7, 26] 17 [8, 23] 13 [5, 34] ArI (events/hour) 41 [34, 64] 22 [16, 33] 20 [14, 32] 16 [12, 30] 17 [9, 29] Sleep with oxygen saturation <90% (%) 9 [5, 27] 5 [1, 20] 3 [1, 14] 4 [2, 12] 5 [1, 21] Sleep with oxygen saturation <90% (min) 33 [16, 87] 14 [2, 44] 8 [2, 26] 11 [5, 33] 15 [3, 53] Percent of sleep in REM (%) 10 [6, 15] 14 [5, 18] 14 [6, 21] 15 [9, 22] 19 [8, 23] . Baseline (n = 58) . 6 months (n = 58) . 12 months (n = 54) . 18 months (n = 45) . 24 months (n = 42) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 60% (47% to 72%) 67% (53% to 78%) 64% (50% to 77%) 60% (44% to 73%) AHI (events/hour) 49 [35, 60] 21 [11, 35] 15 [9, 31] 17 [10, 25] 16 [7, 37] CAI (events/hour) 30 [16, 43] 1 [0, 7] 1 [0, 3] 1 [0, 3] 0 [0, 3] OAI (events/hour) 2 [1, 3] 4 [2, 9] 2 [1, 7] 2 [1, 6] 3 [0, 8] MAI (events/hour) 1 [0, 4] 0 [0, 0] 0 [0, 0] 0 [0, 1] 0 [0, 0] HI (events/hour) 13 [3, 20] 9 [3, 19] 8 [5, 16] 10 [5, 16] 8 [3, 19] ODI4 (events/hour) 41 [30, 56] 19 [8, 37] 14 [7, 26] 17 [8, 23] 13 [5, 34] ArI (events/hour) 41 [34, 64] 22 [16, 33] 20 [14, 32] 16 [12, 30] 17 [9, 29] Sleep with oxygen saturation <90% (%) 9 [5, 27] 5 [1, 20] 3 [1, 14] 4 [2, 12] 5 [1, 21] Sleep with oxygen saturation <90% (min) 33 [16, 87] 14 [2, 44] 8 [2, 26] 11 [5, 33] 15 [3, 53] Percent of sleep in REM (%) 10 [6, 15] 14 [5, 18] 14 [6, 21] 15 [9, 22] 19 [8, 23] Continuous variables reported as median [interquartile range]. CI = confidence interval; N/A = not applicable; ODI4 = oxygen desaturation ≥4% index. Open in new tab Figure 4. Open in new tabDownload slide Treatment group sleep indices by visit. Median AHI, ArI, CAI, and ODI4 are displayed by visit for subjects in the treatment group. A PSG was performed at baseline, 6, 12, 18, and 24 months. A PG was performed at 36 months (33 subjects had not reached this visit at time of study closure). AHI = apnea-hyponea index; ArI = arousal index; CAI = central apnea index; Mo = month; ODI4 = oxygen desaturation ≥4% index. Figure 4. Open in new tabDownload slide Treatment group sleep indices by visit. Median AHI, ArI, CAI, and ODI4 are displayed by visit for subjects in the treatment group. A PSG was performed at baseline, 6, 12, 18, and 24 months. A PG was performed at 36 months (33 subjects had not reached this visit at time of study closure). AHI = apnea-hyponea index; ArI = arousal index; CAI = central apnea index; Mo = month; ODI4 = oxygen desaturation ≥4% index. In addition, 22 treatment group patients that reached the 36-month visit prior to the primary trial closure underwent PG examinations, with results showing enduring and similar effectiveness measurements. Although PG results are not fully comparable with PSG results, at the 36-month visit the median number of events per hour were AHI = 13, CAI = 1, OAI = 4, MAI = 0, and HI = 6 (Table 3). Table 3. PG results at 36-month visit (per protocol population) . Treatment 36-month active therapy (n = 22) . Former control 30-month active therapy (n = 28) . AHI (events/hour) 13 [8, 37] 14 [8, 18] CAI (events/hour) 1 [0, 3] 2 [0, 4] OAI (events/hour) 4 [1, 11] 5 [2, 7] MAI (events/hour) 0 [0, 0] 0 [0, 0] HI (events/hour) 6 [3, 10] 5 [3, 10] . Treatment 36-month active therapy (n = 22) . Former control 30-month active therapy (n = 28) . AHI (events/hour) 13 [8, 37] 14 [8, 18] CAI (events/hour) 1 [0, 3] 2 [0, 4] OAI (events/hour) 4 [1, 11] 5 [2, 7] MAI (events/hour) 0 [0, 0] 0 [0, 0] HI (events/hour) 6 [3, 10] 5 [3, 10] Reported as median [interquartile range]. Not all subjects had reached the 36-month visit at the time of study closure. AHI = apnea-hypopnea index; ArI = arousal index; CAI = central apnea index; HI = hypopnea index; MAI = mixed apnea index; N/A = not applicable; OAI = obstructive apnea index; ODI4 = oxygen desaturation ≥4% index; REM = rapid eye movement. Open in new tab Table 3. PG results at 36-month visit (per protocol population) . Treatment 36-month active therapy (n = 22) . Former control 30-month active therapy (n = 28) . AHI (events/hour) 13 [8, 37] 14 [8, 18] CAI (events/hour) 1 [0, 3] 2 [0, 4] OAI (events/hour) 4 [1, 11] 5 [2, 7] MAI (events/hour) 0 [0, 0] 0 [0, 0] HI (events/hour) 6 [3, 10] 5 [3, 10] . Treatment 36-month active therapy (n = 22) . Former control 30-month active therapy (n = 28) . AHI (events/hour) 13 [8, 37] 14 [8, 18] CAI (events/hour) 1 [0, 3] 2 [0, 4] OAI (events/hour) 4 [1, 11] 5 [2, 7] MAI (events/hour) 0 [0, 0] 0 [0, 0] HI (events/hour) 6 [3, 10] 5 [3, 10] Reported as median [interquartile range]. Not all subjects had reached the 36-month visit at the time of study closure. AHI = apnea-hypopnea index; ArI = arousal index; CAI = central apnea index; HI = hypopnea index; MAI = mixed apnea index; N/A = not applicable; OAI = obstructive apnea index; ODI4 = oxygen desaturation ≥4% index; REM = rapid eye movement. Open in new tab Former control group patients had PSG results available through 18 months of active therapy and a PG at 30 months of active therapy. PSG analysis demonstrated results comparable to the treatment group, with 55%, 52%, and 53% of patients experiencing a 50% or greater improvement in AHI at 6 (n = 65), 12 (n = 61), and 18 (n = 59) months of active therapy, respectively (Table 4). Eighty-six percent (51/59) of patients had a reduction in AHI at 18 months of active therapy (Figure 3, panel B). Moreover, additional sleep metrics and parameters revealed comparable results for this group through 18 months of active therapy (Table 4). The available results from patients completing a PG prior to trial closure at 30 months of active therapy (n = 28) demonstrated similar improvement to that observed in the treatment group (median AHI = 14 events per hour, CAI = 2, OAI = 5, MAI = 0, and HI = 5) (Table 3). Table 4. Former control group PSG results by months of active therapy (per protocol population) . Baseline (n = 73) . 6-month active therapy (n = 65) . 12-months active therapy (n = 61) . 18-month active therapy (n = 59) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 55% (43% to 67%) 52% (40% to 64%) 53% (40% to 65%) AHI (events/hour) 42 [32, 60] 18 [8, 32] 22 [10, 34] 17 [6, 30] CAI (events/hour) 22 [10, 35] 2 [0, 5] 1 [0, 5] 1 [0, 3] OAI (events/hour) 2 [1, 4] 2 [1, 8] 3 [1, 10] 2 [1, 7] MAI (events/hour) 1 [0, 5] 0 [0, 1] 0 [0, 1] 0 [0, 1] HI (events/hour) 11 [4, 18] 7 [3, 15] 9 [4, 19] 8 [3, 18] ODI4 (events/hour) 39 [26, 57] 17 [7, 30] 18 [9, 32] 15 [5, 27] ArI (events/hour) 37 [25, 56] 19 [13, 30] 19 [14, 34] 17 [12, 35] Sleep with oxygen saturation <90% (%) 8 [2, 19] 4 [1, 16] 4 [1, 11] 4 [0, 15] Sleep with oxygen saturation <90% (min) 28 [8, 63] 13 [2, 36] 11 [3, 27] 13 [2, 33] Percent of sleep in REM 11 [6, 16] 16 [10, 20] 16 [7, 22] 16 [9, 21] . Baseline (n = 73) . 6-month active therapy (n = 65) . 12-months active therapy (n = 61) . 18-month active therapy (n = 59) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 55% (43% to 67%) 52% (40% to 64%) 53% (40% to 65%) AHI (events/hour) 42 [32, 60] 18 [8, 32] 22 [10, 34] 17 [6, 30] CAI (events/hour) 22 [10, 35] 2 [0, 5] 1 [0, 5] 1 [0, 3] OAI (events/hour) 2 [1, 4] 2 [1, 8] 3 [1, 10] 2 [1, 7] MAI (events/hour) 1 [0, 5] 0 [0, 1] 0 [0, 1] 0 [0, 1] HI (events/hour) 11 [4, 18] 7 [3, 15] 9 [4, 19] 8 [3, 18] ODI4 (events/hour) 39 [26, 57] 17 [7, 30] 18 [9, 32] 15 [5, 27] ArI (events/hour) 37 [25, 56] 19 [13, 30] 19 [14, 34] 17 [12, 35] Sleep with oxygen saturation <90% (%) 8 [2, 19] 4 [1, 16] 4 [1, 11] 4 [0, 15] Sleep with oxygen saturation <90% (min) 28 [8, 63] 13 [2, 36] 11 [3, 27] 13 [2, 33] Percent of sleep in REM 11 [6, 16] 16 [10, 20] 16 [7, 22] 16 [9, 21] Continuous variables reported as median [interquartile range]. CI = confidence interval; N/A = not applicable; ODI4 = oxygen desaturation ≥4% index. Open in new tab Table 4. Former control group PSG results by months of active therapy (per protocol population) . Baseline (n = 73) . 6-month active therapy (n = 65) . 12-months active therapy (n = 61) . 18-month active therapy (n = 59) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 55% (43% to 67%) 52% (40% to 64%) 53% (40% to 65%) AHI (events/hour) 42 [32, 60] 18 [8, 32] 22 [10, 34] 17 [6, 30] CAI (events/hour) 22 [10, 35] 2 [0, 5] 1 [0, 5] 1 [0, 3] OAI (events/hour) 2 [1, 4] 2 [1, 8] 3 [1, 10] 2 [1, 7] MAI (events/hour) 1 [0, 5] 0 [0, 1] 0 [0, 1] 0 [0, 1] HI (events/hour) 11 [4, 18] 7 [3, 15] 9 [4, 19] 8 [3, 18] ODI4 (events/hour) 39 [26, 57] 17 [7, 30] 18 [9, 32] 15 [5, 27] ArI (events/hour) 37 [25, 56] 19 [13, 30] 19 [14, 34] 17 [12, 35] Sleep with oxygen saturation <90% (%) 8 [2, 19] 4 [1, 16] 4 [1, 11] 4 [0, 15] Sleep with oxygen saturation <90% (min) 28 [8, 63] 13 [2, 36] 11 [3, 27] 13 [2, 33] Percent of sleep in REM 11 [6, 16] 16 [10, 20] 16 [7, 22] 16 [9, 21] . Baseline (n = 73) . 6-month active therapy (n = 65) . 12-months active therapy (n = 61) . 18-month active therapy (n = 59) . Proportion with 50% reduction in AHI from baseline [% (95% CI)] N/A 55% (43% to 67%) 52% (40% to 64%) 53% (40% to 65%) AHI (events/hour) 42 [32, 60] 18 [8, 32] 22 [10, 34] 17 [6, 30] CAI (events/hour) 22 [10, 35] 2 [0, 5] 1 [0, 5] 1 [0, 3] OAI (events/hour) 2 [1, 4] 2 [1, 8] 3 [1, 10] 2 [1, 7] MAI (events/hour) 1 [0, 5] 0 [0, 1] 0 [0, 1] 0 [0, 1] HI (events/hour) 11 [4, 18] 7 [3, 15] 9 [4, 19] 8 [3, 18] ODI4 (events/hour) 39 [26, 57] 17 [7, 30] 18 [9, 32] 15 [5, 27] ArI (events/hour) 37 [25, 56] 19 [13, 30] 19 [14, 34] 17 [12, 35] Sleep with oxygen saturation <90% (%) 8 [2, 19] 4 [1, 16] 4 [1, 11] 4 [0, 15] Sleep with oxygen saturation <90% (min) 28 [8, 63] 13 [2, 36] 11 [3, 27] 13 [2, 33] Percent of sleep in REM 11 [6, 16] 16 [10, 20] 16 [7, 22] 16 [9, 21] Continuous variables reported as median [interquartile range]. CI = confidence interval; N/A = not applicable; ODI4 = oxygen desaturation ≥4% index. Open in new tab Echocardiographic metrics Left ventricular ejection fraction (LVEF) showed small but significant improvements for the treatment group throughout follow-up in the subset of patients with HF, baseline ejection fraction less than or equal to 45% and no permanent atrial fibrillation. Although median LVEF at baseline was 27% (n = 19), paired changes from baseline to each visit reveal slight improvements in absolute percentages of 4%, 8%, and 6% at 12 (n = 17), 18 (n = 14), and 24 (n = 12) months after therapy activation, respectively. Former control group experienced smaller levels of improvement, with a median LVEF at baseline of 34% (n = 26) and paired changes to each visit of 2% at both 12 (n = 22) and 18 (n = 21) months of active therapy. Safety This study provides safety data of the remedē System therapy in a long-term setting. SAEs adjudicated as related to the implant procedure, device, or delivered therapy occurred in 10% of patients (15/151) through the 24-month visit, which is similar to the 9% rate reported through 12 months. The types of events experienced are expected for an implantable neurostimulation device, including concomitant device interactions (three patients), lead component failure (two), lead dislodgment (two), implant site infection (two), impending pocket erosion (two), inadequate lead position (one), lead displacement (one), feeling a sensation in an area remote from the diaphragm (one), implant site hematoma (one), noncardiac chest pain (one), and elevated transaminase (one) (Table 5). All procedure-related events resolved with routine care and all device- or therapy-related events resolved with remedē System revisions or programming. Further safety data beyond the 24 months follow-up will be collected as part of the ongoing remedē System Post Approval Study (NCT03425188), which is following patients from the Pivotal Trial throughout 5 years post implant. However, no additional related SAEs were reported between 24 and 36 months in the Pivotal Trial. Table 5. Implant procedure-, device-, or delivered-therapy-related serious adverse events through 24 months of follow-up Event . Pooled groups (N = 151) . Any event 15 (10%) Concomitant device interaction 3 (2%) Lead component failure 2 (1%) Lead dislodgment 2 (1%) Implant site infection 2 (1%) Impending pocket erosion 2 (1%) Inadequate lead position 1 (1%) Lead displacement 1 (1%) Feeling sensation in an area remote from the diaphragm 1 (1%) Implant site hematoma 1 (1%) Noncardiac chest pain 1 (1%) Elevated transaminase 1 (1%) Event . Pooled groups (N = 151) . Any event 15 (10%) Concomitant device interaction 3 (2%) Lead component failure 2 (1%) Lead dislodgment 2 (1%) Implant site infection 2 (1%) Impending pocket erosion 2 (1%) Inadequate lead position 1 (1%) Lead displacement 1 (1%) Feeling sensation in an area remote from the diaphragm 1 (1%) Implant site hematoma 1 (1%) Noncardiac chest pain 1 (1%) Elevated transaminase 1 (1%) Reported as number with event (percent). All events resolved with routine care, system revisions, or programming changes. Open in new tab Table 5. Implant procedure-, device-, or delivered-therapy-related serious adverse events through 24 months of follow-up Event . Pooled groups (N = 151) . Any event 15 (10%) Concomitant device interaction 3 (2%) Lead component failure 2 (1%) Lead dislodgment 2 (1%) Implant site infection 2 (1%) Impending pocket erosion 2 (1%) Inadequate lead position 1 (1%) Lead displacement 1 (1%) Feeling sensation in an area remote from the diaphragm 1 (1%) Implant site hematoma 1 (1%) Noncardiac chest pain 1 (1%) Elevated transaminase 1 (1%) Event . Pooled groups (N = 151) . Any event 15 (10%) Concomitant device interaction 3 (2%) Lead component failure 2 (1%) Lead dislodgment 2 (1%) Implant site infection 2 (1%) Impending pocket erosion 2 (1%) Inadequate lead position 1 (1%) Lead displacement 1 (1%) Feeling sensation in an area remote from the diaphragm 1 (1%) Implant site hematoma 1 (1%) Noncardiac chest pain 1 (1%) Elevated transaminase 1 (1%) Reported as number with event (percent). All events resolved with routine care, system revisions, or programming changes. Open in new tab The median months to battery depletion estimated from Kaplan–Meier analysis was 45.1 months for a left stimulation lead and 34.6 for a right stimulation lead, with a combined estimate of 39.4 months. If a device was replaced prior to full battery depletion, the date of the replacement was used in the calculation so the actual duration may be underestimated. There were three implant site infections following the initial implant procedure, however, no implant site infections have been reported following any remedē System implantable pulse generator replacement or lead revision procedure. Survival Overall survival through 3 years is illustrated in a Kaplan–Meier curve (Figure 5) for all patients who had therapy activated (both treatment and former control groups pooled). Through 24 months, the independent CEC adjudicated deaths to the following modalities: 8 (5% of 151 patients) noncardiovascular, 6 (4%) pump failure including 2 control patients prior to therapy activation, 4 (3%) experienced sudden cardiac death (2 out of hospital cardiac arrests, one due to ventricular arrhythmia [per device interrogation], and one other cardiovascular death), and 1 (1%) intracranial bleed subsequent to a fall. Beyond 24 months, five additional pump failure deaths, three noncardiovascular deaths, one sudden cardiac death (asystole in the setting of end stage HF [per device interrogation]), and one other cardiovascular death had been reported prior to study closure. The survival rates remain constant over time, suggesting no additional risk based on duration of therapy. Critically, no deaths have been associated with the device, therapy, or implant procedure. Figure 5. Open in new tabDownload slide Kaplan–Meier curve of mortality. Kaplan–Meier curve showing estimated mortality through 36 months of active therapy using the pooled treatment and former control groups. Patients who did not die were censored at last contact if they did not reach 36 months of active therapy. Figure 5. Open in new tabDownload slide Kaplan–Meier curve of mortality. Kaplan–Meier curve showing estimated mortality through 36 months of active therapy using the pooled treatment and former control groups. Patients who did not die were censored at last contact if they did not reach 36 months of active therapy. Discussion This study provides the first data on long-term safety and effectiveness of the novel PNS for the treatment of moderate-to-severe CSA through 36 months of patient follow-up. This study shows therapy effectiveness remains consistent and reliable for 36 months of active therapy. Ninety-three percent of patients in the Pivotal Trial treatment group and 86% of the former control group had a sustained reduction in AHI at the 24-month PSG. CAI was less than or equal to 1 event per hour at 12 months and beyond and more importantly the PNS system improved the ArI in this long-term study [median 41 [34, 64] events per hour at baseline and 20 [14, 32], 16 [12, 30], and 17 [9, 29] at 12, 18, and 24 months]. Moreover, PNS therapy improved REM sleep and the percentage and minutes of sleep with oxygen saturation less than 90%, which is an independent predictor of all-cause mortality in chronic HF as recently demonstrated [12]. LVEF showed small but measurable improvements with this therapy. Above all, this study demonstrates evidence of enduring safety using PNS as a long-term CSA therapy, with SAEs adjudicated to the implant procedure, device, or delivered therapy in 10% of patients through 24 months. This is only one percentage point higher than at 12 months, indicating most complications occur in the first year. All of the events resolved with routine care, remedē System revisions or programming changes. Importantly, although the patient populations are quite different, the rate of subjects experiencing sudden cardiac death is lower than the control group in the SERVE-HF trial. However, this finding should be confirmed in larger trials with particular attention paid to adjudicating any cardiac death. Overall survival demonstrates constant survival rates during the trial follow-up, suggesting no additional risk with increasing duration of PNS therapy, as depicted in the Kaplan–Meier survival curve in Figure 5. This is very important as clinicians consider appropriate therapy options for their patients with moderate-to-severe CSA. As the first description of the remedē System by Augostini et al. [13], interest was high in this novel approach of treating CSA, but interventionalists and sleep specialists wanted to see evidence of longer-term effectiveness and safety. The randomized controlled Pivotal Trial achieved the primary effectiveness endpoint with an AHI reduction from baseline of 50% or greater at 6 months of follow-up that was significantly higher in the treatment group (51%) than in the control group (11%) in the intent-to-treat population [9]. Also, 91% of patients were free from SAEs related to the implant procedure, device, or delivered therapy through 12 months [9]. Subsequent analysis of 12 months follow-up demonstrated sustained at least 50% reduction in AHI in 67% of the treatment group and 55% of the former control group experienced at least 50% reduction in AHI following 6 months of active therapy [10]. In addition, patient global assessment was markedly or moderately improved at 6 and 12 months in 60% of treatment patients [10]. The data in this manuscript show long-term performance of PNS remains consistent with previously reported data. REM sleep is essential for any human. Patients with CSA and cardiovascular disease, especially those with HF, are often deprived of REM sleep and experience poor sleep quality [14]. Although the debate on this topic continues, REM sleep deprivation is associated with emergence of anxiety, irritability, hallucinations, impairment in concentration, and an increase in appetite, which are aspects that can be found in patients with HF. Reduced sleep quality is not only associated with impaired quality of life but also with increased mortality in patients with HF [15]. Less restorative sleep, changes in sympathovagal balance and impaired resetting of important reflexes may contribute to worse cardiovascular outcomes in HF patients with CSA [14], but CSA treatment has been shown to improve the proportion of sleep in REM [15]. This analysis from the remedē System Pivotal Trial shows similar improvements for REM sleep through the use of PNS in a long-term setting, whereas a recent publication demonstrated PNS to be associated with benefits on HF quality of life [16]. This is of particular importance, as more than 50% of patients with CSA have either overt or yet undiagnosed underlying HF [4]. The forest plot (Supplementary Figure S1) showing the proportion of subjects achieving at least 50% reduction in AHI for subgroups of interest suggests that all subgroups displayed have a reasonable probability of subjects reaching 50% improvement in AHI. The few subgroups that have an estimated proportion below 50% are associated with wide confidence intervals that have the upper limit extend beyond 50%, meaning the true proportion may exceed 50%. Our study results show that PNS using the remedē System can be performed with high procedural success and low rates of SAEs, close to what is reported in CRT trials [17]. Moreover, the battery longevity of the remedē System is close to 4 years for a stimulation lead implanted on the left side where the majority of the study implants occurred and within the expected range of 17–55 months based on high and low-energy use conditions; the replacement procedure is comparable to other cardiac devices [11]. The 24-month multivariate analysis showed a lower BMI and higher HI to be predictors of better response to PNS. Higher HI may indicate patients that are easier to treat with the assumption that a majority of the hypopneas are central. However, as hypopneas were not scored as central versus obstructive in this trial, we cannot determine whether that is true. In general, known risk factors for OSA include obesity, large neck circumference, male gender, increasing age, alcohol use, smoking, menopausal status, and black race so it may not be surprising that a higher BMI could limit PNS [18, 19]. We repeated the analysis using pooled (treatment and former control) 18-month results and found BMI, sex, and ODI4 to be predictors of response. The interpretation for the sex finding is limited by small sample size (9/10 females had 50% AHI reduction). The ODI4 may simply suggest that patients with more severe disease more easily improve AHI by 50%. The common finding of BMI in both analyses may suggest that patients with high BMI are less likely to attain 50% reduction in AHI following PNS, although a particular BMI cutoff where that happens was not identified. However, future investigations are needed to confirm and clarify these findings. LVEF slightly improved. Similar ranges of LVEF improvements have been reported in earlier trials when CSA was effectively controlled [20–22], but whether this finding is of any clinical relevance will be a cornerstone of future clinical trials [8]. This is the first trial to provide long-term data through 36 months on the treatment of CSA with PNS and its results are consistent with previously reported data. Importantly, the enduring improvement in sleep indices and patient safety may indicate that this novel technology can be an option for long-term treatment of CSA. Therapy effectiveness was shown to be reproducible in the former control group with comparable sustained therapy affect supporting enduring and reliable treatment. This study adds confirmation to available data investigating PNS for CSA treatment and supports PNS to be an effective and safe treatment improving sleep and quality of life in CSA patients. One of our limitations in the study design is that the control group was followed only for 6 months prior to activating therapy so control data are somewhat limited. However, at the time of the study design, it was felt that depriving patients with symptomatic CSA of any treatment for longer than 6 months was unethical. Importantly, in assessing long-term safety and effectiveness, the ability to pool the former control patients with the therapy group provided more patients in which to assess the safety profile as well as effectiveness of the therapy. Another limitation is that not all patients completed 36 months at the time the Pivotal Trial was closed following FDA approval. However, the data collected through 24 and 36 months indicate continued effectiveness and safety based on the available PG results and reported adverse events (AEs). Moreover, additional AEs after 24 months may be reported in the ongoing remedē System Post Approval Study (NCT03425188) that is following patients from the Pivotal Trial through 5 years post implant. In conclusion, this 36-month long-term analysis establishes that unilateral transvenous PNS with the remedē System in patients with CSA is associated with a high response to therapy as indicated by sustained improvement in key sleep indices through 36-months follow-up as well as echocardiography parameters through 24 months. Also, the remedē System demonstrates a strong safety profile through 36 months. The beneficial effects of long-term PNS in patients with CSA appear to be sustained through 36 months with no new safety concerns. Acknowledgments The authors wish to acknowledge the technical expertise of United Heart and Vascular Clinic’s Echocardiography Core Laboratory, supervised by Dr. Alan Bank (St. Paul, MN) as well as Registered Sleepers Sleep, Inc. Core Laboratory, supervised by Tim Winchester (Leicester, NC). Funding Conflict of interest statement. This trial was funded by Respicardia. The authors have following conflicts of interest to declare: Fox: no conflict of interest to declare regarding this manuscript. Oldenburg: no conflict of interest to declare regarding this manuscript. Javaheri: personal fees from Respicardia. Ponikowski: research grants from Respicardia and Coridea; personal fees from Respicardia, Coridea. Augostini: Consultant/Speaker Bureau, Respicardia; Advisory Board, Respicardia and Philips. Goldberg: research grants and consulting fees from Respircardia. Stellbrink: research grants from Respicardia, St. Jude Medical, Biotronik, Medtronic, and Sorin/LivaNova; advisory board for Sorin/LivaNova. McKane and Meyer: Employees of Respicardia. Abraham: research grants to institution from Respicardia; personal fees from Respicardia (consulting, Advisory Board). Costanzo: personal fees from Respicardia (consulting and study principal investigator for remedē System Pivotal Trial). References 1. Linz D , et al. The importance of sleep-disordered breathing in cardiovascular disease . Clin Res Cardiol. 2015 ; 104 ( 9 ): 705 – 718 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Anker SD , et al. Sleep-disordered breathing and cardiovascular disease . Indian Heart J. 2016 ; 68 ( Suppl 1 ): S69 – S76 . Google Scholar Crossref Search ADS PubMed WorldCat 3. Fox H , et al. Prevalence of sleep-disordered breathing and patient characteristics in a coronary artery disease cohort undergoing cardiovascular rehabilitation . J Cardiopulm Rehabil Prev. 2016 ; 36 ( 6 ): 421 – 429 . Google Scholar Crossref Search ADS PubMed WorldCat 4. Arzt M , et al. Phenotyping of sleep-disordered breathing in patients with chronic heart failure with reduced ejection fraction-the SchlaHF registry . J Am Heart Assoc. 2017 ; 6 ( 12 ): 1 – 10 . Google Scholar Crossref Search ADS WorldCat 5. Javaheri S , et al. Sleep apnea: types, mechanisms, and clinical cardiovascular consequences . J Am Coll Cardiol. 2017 ; 69 ( 7 ): 841 – 858 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Cowie MR , et al. Adaptive servo-ventilation for central sleep apnea in systolic heart failure . N Engl J Med. 2015 ; 373 ( 12 ): 1095 – 1105 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Oldenburg O , et al. Addendum zum Positionspapier „Schlafmedizin in der Kardiologie. Update 2014” . Somnologie . 2017 ; 21 ( 1 ): 51 – 52 . Google Scholar Crossref Search ADS WorldCat 8. Oldenburg O , et al. Adaptive servoventilation to treat sleep-disordered breathing in cardiac patients . Somnologie . 2017 ; 21 ( 1 ): 82 – 83 . Google Scholar Crossref Search ADS WorldCat 9. Costanzo MR , et al. ; remedé System Pivotal Trial Study Group . Transvenous neurostimulation for central sleep apnoea: a randomised controlled trial . Lancet. 2016 ; 388 ( 10048 ): 974 – 982 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Costanzo MR , et al. ; remedē System Pivotal Trial Study Group . Sustained 12 month benefit of phrenic nerve stimulation for central sleep apnea . Am J Cardiol. 2018 ; 121 ( 11 ): 1400 – 1408 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Fox H , et al. Long-term experience with first-generation implantable neurostimulation device in central sleep apnea treatment . Pacing Clin Electrophysiol. 2017 ; 40 ( 5 ): 498 – 503 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Oldenburg O , et al. Nocturnal hypoxaemia is associated with increased mortality in stable heart failure patients . Eur Heart J. 2016 ; 37 ( 21 ): 1695 – 1703 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Augostini R . A novel approach to the treatment of central sleep apnea in patients with heart failure . Herzschrittmacherther Elektrophysiol. 2012 ; 23 ( 1 ): 9 – 13 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Türoff A , et al. Sleep duration and quality in heart failure patients . Sleep Breath. 2017 ; 21 ( 4 ): 919 – 927 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Heider K , et al. Adaptive servo-ventilation and sleep quality in treatment emergent central sleep apnea and central sleep apnea in patients with heart disease and preserved ejection fraction . Clin Res Cardiol. 2018 ; 107 ( 5 ): 421 – 429 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Costanzo MR , et al. ; remedē® System Pivotal Trial Study Group . Phrenic nerve stimulation to treat patients with central sleep apnoea and heart failure . Eur J Heart Fail. 2018 ; 20 ( 12 ): 1746 – 1754 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Al-Majed NS , et al. Meta-analysis: cardiac resynchronization therapy for patients with less symptomatic heart failure . Ann Intern Med. 2011 ; 154 ( 6 ): 401 – 412 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Malhotra A , et al. Obstructive sleep apnoea . Lancet. 2002 ; 360 ( 9328 ): 237 – 245 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Davies RJ , et al. Neck circumference and other clinical features in the diagnosis of the obstructive sleep apnoea syndrome . Thorax. 1992 ; 47 ( 2 ): 101 – 105 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Aurora RN , et al. The treatment of central sleep apnea syndromes in adults: practice parameters with an evidence-based literature review and meta-analyses . Sleep. 2012 ; 35 ( 1 ): 17 – 40 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Fietze I , et al. Bi-level positive pressure ventilation and adaptive servo ventilation in patients with heart failure and Cheyne-Stokes respiration . Sleep Med. 2008 ; 9 ( 6 ): 652 – 659 . Google Scholar Crossref Search ADS PubMed WorldCat 22. Pepperell JC , et al. A randomized controlled trial of adaptive ventilation for Cheyne-Stokes breathing in heart failure . Am J Respir Crit Care Med. 2003 ; 168 ( 9 ): 1109 – 1114 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press [on behalf of the Sleep Research Society]. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact [email protected] © Sleep Research Society 2019. Published by Oxford University Press [on behalf of the Sleep Research Society].

Zhang,, Linda;Fabbri,, Daniel;Upender,, Raghu;Kent,, David

doi: 10.1093/sleep/zsz159pmid: 31289828

Abstract Study Objectives Polysomnography (PSG) scoring is labor intensive and suffers from variability in inter- and intra-rater reliability. Automated PSG scoring has the potential to reduce the human labor costs and the variability inherent to this task. Deep learning is a form of machine learning that uses neural networks to recognize data patterns by inspecting many examples rather than by following explicit programming. Methods A sleep staging classifier trained using deep learning methods scored PSG data from the Sleep Heart Health Study (SHHS). The training set was composed of 42 560 hours of PSG data from 5213 patients. To capture higher-order data, spectrograms were generated from electroencephalography, electrooculography, and electromyography data and then passed to the neural network. A holdout set of 580 PSGs not included in the training set was used to assess model accuracy and discrimination via weighted F1-score, per-stage accuracy, and Cohen’s kappa (K). Results The optimal neural network model was composed of spectrograms in the input layer feeding into convolutional neural network layers and a long short-term memory layer to achieve a weighted F1-score of 0.87 and K = 0.82. Conclusions The deep learning sleep stage classifier demonstrates excellent accuracy and agreement with expert sleep stage scoring, outperforming human agreement on sleep staging. It achieves comparable or better F1-scores, accuracy, and Cohen’s kappa compared to literature for automated sleep stage scoring of PSG epochs. Accurate automated scoring of other PSG events may eventually allow for fully automated PSG scoring. polysomnography, sleep staging, deep learning, machine learning Statement of Significance Sleep staging is an important part of evaluating overnight polysomnograms. Sleep stages are scored by technicians and physicians based on visual examination of neurophysiologic signal patterns. This process is labor intensive and suffers from variability between scorers. In this study, large amounts of publicly available polysomnography data were used to train a sleep staging classifier. Sleep staging classification by the model achieved better agreement than human agreement in literature. Generalizability of the model to other unseen datasets from different public projects is also demonstrated. Introduction Overnight polysomnography (PSG) is central to the diagnosis and management of many sleep disorders. The clinical standard for PSG sleep staging requires visual inspection of the data by trained sleep technicians and physicians. Staging historically followed the Rechtschaffen and Kales (RK) criteria until the American Academy of Sleep Medicine (AASM) published updated criteria in 2007 [1, 2]. The AASM rules divide sleep into five stages: Wake, Non-Rapid Eye Movement stages 1, 2, and 3 (N1, N2, and N3), and Rapid Eye Movement (REM). PSG scoring is a labor-intensive process that requires up to 2 hours for a sleep technologist to complete [3]. Inter-rater and intra-rater reliability of PSG staging and event scoring is also known to suffer from considerable variability [4–13]. Significant effort has been invested in developing computer-assistive or automated staging technologies, but they have struggled to achieve human-level performance [3, 14–25]. In order for a staging system to have clinical utility it should be at least as accurate and reliable as a trained human scorer. Therefore, a practical non-inferiority threshold for staging algorithms is an overall agreement of 82.0% (Cohen’s kappa = 0.76), which is the overall inter-rater agreement between trained scorers at eight European centers using the 2007 AASM PSG scoring rules [5]. Machine learning is a field of computer science where classifiers discover novel patterns within a dataset without the traditional explicit encoding of all rules. Because PSG data are complex, different machine learning methods for detecting sleep stages have been trialed over the last 20 years. Published models have used hand-tuned feature extraction techniques such as spectral power, time domain analysis, and time–frequency domain (wavelet) analysis [26–29]. Other systems use fuzzy logic, support vector machines, hidden Markov models, or artificial neural networks [30–38]. Most of these systems do not achieve human-level inter-rater agreement or are tested against a small set of preselected, high-quality PSGs that do not reflect realistic testing environments. Few have been validated against large clinical datasets. In recent years, deep neural networks have rapidly found favor for signal analysis. They have proven to be remarkably robust in developing classifier systems for noisy, “real-world” datasets: the type of data represented by PSGs. A standard neural network consists of a number of simple connected processors called neurons that mathematically transform an input signal into an output. The relative strength, or weight, of each neuron is iteratively adjusted during model training to maximize the accuracy between the network output and the expected value. Deep neural networks have many layers of neurons, where the output of one layer provides the input to the next layer, enabling discovery of nonlinear and hierarchical relationships within the data. Convolutional neural networks (CNNs) emphasize patterns in close spatial proximity and are well suited to problems in the image classification and recognition space [39, 40]. Recurrent neural networks function well with information contained in sequences such as natural language, where the next word or character depends on the immediately preceding data [41]. PSGs are well suited for convolutional and recurrent processing methodologies as they consist of spatially and temporally related signal data. For example, a k-complex may signal onset of N2 sleep, even though subsequent electromyogram (EMG) data may be low-amplitude mixed-frequency data visually identical to N1. The increase in available computing power and publicly available PSG datasets over the last several years has brought the era of Big Data and machine learning to sleep medicine and made deep neural network processing of PSGs feasible [42, 43]. Successful development of a reliable and accurate automated scoring system using machine learning will ease the burden of PSG scoring and will reduce sleep staging inter-rater variability that affects Sleep Medicine research and clinical practice. Methods This study was designed as a retrospective analysis of PSG data collected through several multicenter cohort studies made available through the National Sleep Research Resource (NSRR) [43–45]. The study design was approved by the Vanderbilt University Medical Center Institutional Review Board (#171186) and data access was approved by the NSRR. Study datasets A deep neural network model was trained and tested on 5804 Type II PSGs from multiple centers containing patients with and without sleep-disordered breathing collected for the Sleep Heart Health Study (SHHS; Table 1) [43–45]. Two additional unrelated datasets available through the NSRR were used to test the generalizability of the model: the Study of Osteoporotic Fractures (SOF) and the Osteoporotic Fractures in Men study (MrOS; Table 2). Table 1. Sleep Heart Health Study summary statistics Category Mean Median Min, max Age 63.1 63 [39, 90] Body mass index 28.2 27.5 [18, 50] Apnea–hypopnea index 17.9 13.2 [0, 161.8] Sleep time (minutes) 359.8 367.0 [34.5, 519] Category Mean Median Min, max Age 63.1 63 [39, 90] Body mass index 28.2 27.5 [18, 50] Apnea–hypopnea index 17.9 13.2 [0, 161.8] Sleep time (minutes) 359.8 367.0 [34.5, 519] Open in new tab Table 1. Sleep Heart Health Study summary statistics Category Mean Median Min, max Age 63.1 63 [39, 90] Body mass index 28.2 27.5 [18, 50] Apnea–hypopnea index 17.9 13.2 [0, 161.8] Sleep time (minutes) 359.8 367.0 [34.5, 519] Category Mean Median Min, max Age 63.1 63 [39, 90] Body mass index 28.2 27.5 [18, 50] Apnea–hypopnea index 17.9 13.2 [0, 161.8] Sleep time (minutes) 359.8 367.0 [34.5, 519] Open in new tab Table 2. Summary of datasets used in study Dataset Polysomnography studies (n) Study population W (%) N1 (%) N2 (%) N3 (%) R (%) SHHS 5793 Adults aged 40 years and older 28.8 3.7 40.9 12.6 13.9 MrOS 2907 Men 65 years or older 46.1 3.7 33.9 5.8 10.6 SOF 461 Women ages 65–89 years 41.9 2.9 32.5 11.9 10.7 Dataset Polysomnography studies (n) Study population W (%) N1 (%) N2 (%) N3 (%) R (%) SHHS 5793 Adults aged 40 years and older 28.8 3.7 40.9 12.6 13.9 MrOS 2907 Men 65 years or older 46.1 3.7 33.9 5.8 10.6 SOF 461 Women ages 65–89 years 41.9 2.9 32.5 11.9 10.7 Open in new tab Table 2. Summary of datasets used in study Dataset Polysomnography studies (n) Study population W (%) N1 (%) N2 (%) N3 (%) R (%) SHHS 5793 Adults aged 40 years and older 28.8 3.7 40.9 12.6 13.9 MrOS 2907 Men 65 years or older 46.1 3.7 33.9 5.8 10.6 SOF 461 Women ages 65–89 years 41.9 2.9 32.5 11.9 10.7 Dataset Polysomnography studies (n) Study population W (%) N1 (%) N2 (%) N3 (%) R (%) SHHS 5793 Adults aged 40 years and older 28.8 3.7 40.9 12.6 13.9 MrOS 2907 Men 65 years or older 46.1 3.7 33.9 5.8 10.6 SOF 461 Women ages 65–89 years 41.9 2.9 32.5 11.9 10.7 Open in new tab PSG data All PSG files were downloaded in the European Data Format which contained the raw time series data of physiologic signals from each PSG as well as human scored sleep stages and apneic events. For the training phase, 5213 PSGs were randomly selected from the SHHS dataset, providing 42 560 hours of sleep data in 5 107 200 30-second epochs. PSGs in all three datasets were recorded as Type II unattended home studies previously scored using modified RK criteria [43–45]. PSG signal data and sleep stage labeling (Wake, N1, N2, N3, N4, or REM) were extracted from each study cohort. RK stages 3 and 4 were combined into a single stage N3 label to more closely align with modern AASM scoring conventions and to aid comparison with previously published literature. The model was trained and tuned using 90% of the SHHS visit 1 data (5213 patients). A 10% holdout set (580 patients) was taken and set aside to validate the model. Input data and feature selection Signal data from the electroencephalogram (EEG), EMG, and electrooculogram (EOG) PSG channels were extracted for model analysis. The Type II PSGs across all three cohorts were recorded using a single central (C3) EEG channel. Sampling rates across data channels from SOF and MrOS were down- or up-sampled as indicated to match corresponding baseline data sampling rates from SHHS. Two different methodologies for feature representation were tested. In the first method, raw PSG signal data were provided directly as input to the network in per-epoch units and tested under various model architectures. In the second method, short-time Fourier transforms were used to generate a spectrogram for each epoch and then provided to the model as the input. Spectrograms were generated using 2-second sub-epochs formed by a Tukey window with 25% of the window inside the tapered cosine region (Figure 1). Signal normalization and filter signal preprocessing methods (median, finite impulse response, and infinite impulse response filters) were tested to evaluate the impact of noise and artifact reduction. Figure 1. Open in new tabDownload slide Representative raw data sample from each sleep stage with associated spectrogram. Figure 1. Open in new tabDownload slide Representative raw data sample from each sleep stage with associated spectrogram. All data preprocessing was performed using the signal module in the python packages SciPy and scikit-learn. Model development was performed using Keras on a TensorFlow backend. Model architecture Convolutional and recurrent network layers were used to take advantage of the temporally linked, sequential construction of PSG data. Convolutional layers were generated to evaluate the co-occurrence of signal patterns within one-dimensional PSG data channels or co-occurrence of frequencies within single spectrograms. Recurrent layers were designed to take advantage of the temporal relationships in the data such as epochs of equivalent stage occurring in sequence. The deep neural network combined recurrent and convolutional structures to evaluate input spectrograms generated from the raw data (Figure 2). Multiple combinations of dense, convolutional, and recurrent layers were tested against the training set in the network architecture (Supplementary Appendix A). Figure 2. Open in new tabDownload slide Simplified example model architecture for one data channel. LSTM = long short-term memory layer. Figure 2. Open in new tabDownload slide Simplified example model architecture for one data channel. LSTM = long short-term memory layer. Model tuning Deep neural networks contain tunable hyperparameters (i.e. number of layers, number of units in each layer, number of filters in convolutional layers). A set of parameter search spaces were defined for each hyperparameter, and the best combination of hyperparameters were found using the python package hyperopt with a random search algorithm for parameter tuning [46]. Multiple hyperparameter configurations were evaluated using the training set. Model evaluation Model performance was evaluated with accuracy, F1-score, and Cohen’s kappa. Weighted and unweighted accuracy and F1-score were calculated to assess the effect of sleep stage class imbalances in the data. Weighted accuracy was calculated as the average of the per-class stage accuracies. Because the “ground truth” comparators are human-tagged PSG events with their own level of inter-rater reliability, model agreement was also assessed using inter-rater agreement statistics (Cohen’s kappa). Transition epoch F1-scores were calculated as scoring agreement is known to degrade during transition from one stage of sleep to another. Transition stages account for approximately 0.5% of the data, but were evaluated as they potentially convey physiologically relevant information. Transfer learning Generalizability was assessed using the SOF and MrOS datasets. These studies were conducted in different environments with various types of acquisition hardware and on different patient populations than SHHS. Model performance was additionally evaluated on subsets of the SHHS population with mild, moderate, and severe obstructive sleep apnea (OSA) to demonstrate model transferability between patients with different degrees of sleep-disordered breathing. A separate model was also trained and tested on only severe patients to demonstrate validity even when restricted to a subset of studied patients. Results The optimal sleep staging model’s architecture consisted of a combination of separate networks for each signal channel. Spectrograms of each channel were fed into convolutional layers that examined the proximal relationships of the frequencies in time as well as recurrent layers that examined the sequential relationships of epochs (Table 3). The subnetworks for each signal channel were combined into two dense layers feeding into a final softmax output layer used to generate discrete stage predictions for each epoch. Table 3. Base model architecture per data channel Layer Layer type Size Output size Input (2, 1, 129, 16) C1 Convolutional (32, 64, 3) (2, 32, 66, 14) C2 Convolutional (32, 64, 3) (2, 32, 2, 12) P1 Max pooling (2, 2) (2, 32, 1, 6) R1 Reshape (2, 192) L1 Long short-term memory 256 256 D1 Dense 512 512 Layer Layer type Size Output size Input (2, 1, 129, 16) C1 Convolutional (32, 64, 3) (2, 32, 66, 14) C2 Convolutional (32, 64, 3) (2, 32, 2, 12) P1 Max pooling (2, 2) (2, 32, 1, 6) R1 Reshape (2, 192) L1 Long short-term memory 256 256 D1 Dense 512 512 Open in new tab Table 3. Base model architecture per data channel Layer Layer type Size Output size Input (2, 1, 129, 16) C1 Convolutional (32, 64, 3) (2, 32, 66, 14) C2 Convolutional (32, 64, 3) (2, 32, 2, 12) P1 Max pooling (2, 2) (2, 32, 1, 6) R1 Reshape (2, 192) L1 Long short-term memory 256 256 D1 Dense 512 512 Layer Layer type Size Output size Input (2, 1, 129, 16) C1 Convolutional (32, 64, 3) (2, 32, 66, 14) C2 Convolutional (32, 64, 3) (2, 32, 2, 12) P1 Max pooling (2, 2) (2, 32, 1, 6) R1 Reshape (2, 192) L1 Long short-term memory 256 256 D1 Dense 512 512 Open in new tab Model testing The SHHS dataset was split into a 90% training and 10% holdout set. The training set was further split into training and validation sets, which were used to train the model, select the optimal deep learning architecture (Supplementary Appendix A), and tune the model hyperparameters (Supplementary Appendix B). Model training required approximately 48 hours on an Nvidia GTX Titan X GPU. A learning curve plateauing around 1 000 000 training epochs demonstrated that the dataset was sufficiently large (Figure 3). Testing on the holdout set required approximately 30 minutes. Figure 3. Open in new tabDownload slide The deep neural network model learning curve begins to plateau after training on approximately 1 000 000 epochs. Figure 3. Open in new tabDownload slide The deep neural network model learning curve begins to plateau after training on approximately 1 000 000 epochs. Model evaluation Signal preprocessing methods were tested on the raw input signal. No significant improvement in accuracy or F1-scores were found using normalization or filters, so signal preprocessing was not used in the final pipeline (data not shown). Multiple model architectures were tested on the SHHS dataset. The first model was a simple baseline Markov chain that predicted the next stage based on overall stage transition probabilities measured directly from SHHS. Because stages commonly occur in long chains with relatively rare transitions, this model has a high F1-score, but low transition F1-score. Following this baseline model, a CNN was tested against raw PSG data, followed by separate CNN and long short-term memory (LSTM) models on the spectrogram data, and finally a combination of CNN + LSTM, which yielded the best performance (Figure 4). Figure 4. Open in new tabDownload slide Model performance under various architectures against the SHHS dataset. CNN = convolutional neural network; LSTM = long short-term memory Figure 4. Open in new tabDownload slide Model performance under various architectures against the SHHS dataset. CNN = convolutional neural network; LSTM = long short-term memory The optimal neural network model was composed of spectrograms in the input layer feeding into CNN layers and an LSTM layer to achieve a weighted F1-score of 0.87 and Cohen’s Unweighted kappa of K = 0.82, higher than that of human agreement found in literature (K = 0.76). A confusion matrix was generated for model performance against all tested epochs (Figure 5) as well as transition epochs (Figure 6). When considering all epochs, the model scored Wake, N1, N2, N3, and REM stages correctly 92%, 37%, 91%, 77%, and 88% of the time, respectively. During transition epochs correct staging was scored for Wake, N1, N2, N3, and REM 75%, 44%, 79%, 54%, and 88% of the time, respectively. Table 4 compares staging accuracy of this model to others published in the literature using the class imbalances present in the underlying dataset. Table 5 permits comparison to other models in the literature that used methods to balance the classes such that all classes contribute equally in model training. Figure 7 demonstrates agreement between a trained scorer and the automated scoring model in one example PSG hypnogram. Table 4. Performance of class imbalanced model compared to other studies Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy Cohen’s kappa Biswal et al. [47] 10 000 Train–validation–test 84.5% 56.2% 88.4% 85.4% 92% 85.8% 81.3% 0.795 Sors et al. [48] 5793 Training–validation– test 91% 35% 89% 85% 86% 87% 77.2% 0.81 Sharma et al. [49] 100 10-fold-CV 95% 17% 76% 57% 36% 91.7% 56.5% N/A Proposed model 5793 Train–validation–test 92% 37% 91% 77% 88% 87% 77% 0.82 Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy Cohen’s kappa Biswal et al. [47] 10 000 Train–validation–test 84.5% 56.2% 88.4% 85.4% 92% 85.8% 81.3% 0.795 Sors et al. [48] 5793 Training–validation– test 91% 35% 89% 85% 86% 87% 77.2% 0.81 Sharma et al. [49] 100 10-fold-CV 95% 17% 76% 57% 36% 91.7% 56.5% N/A Proposed model 5793 Train–validation–test 92% 37% 91% 77% 88% 87% 77% 0.82 Open in new tab Table 4. Performance of class imbalanced model compared to other studies Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy Cohen’s kappa Biswal et al. [47] 10 000 Train–validation–test 84.5% 56.2% 88.4% 85.4% 92% 85.8% 81.3% 0.795 Sors et al. [48] 5793 Training–validation– test 91% 35% 89% 85% 86% 87% 77.2% 0.81 Sharma et al. [49] 100 10-fold-CV 95% 17% 76% 57% 36% 91.7% 56.5% N/A Proposed model 5793 Train–validation–test 92% 37% 91% 77% 88% 87% 77% 0.82 Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy Cohen’s kappa Biswal et al. [47] 10 000 Train–validation–test 84.5% 56.2% 88.4% 85.4% 92% 85.8% 81.3% 0.795 Sors et al. [48] 5793 Training–validation– test 91% 35% 89% 85% 86% 87% 77.2% 0.81 Sharma et al. [49] 100 10-fold-CV 95% 17% 76% 57% 36% 91.7% 56.5% N/A Proposed model 5793 Train–validation–test 92% 37% 91% 77% 88% 87% 77% 0.82 Open in new tab Table 5. Performance of class balanced model compared to other studies Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy F1-Score Cohen’s kappa Supratak et al. [50] 62 31-fold cross validation 87.3% 43.5% 90.5% 77.1% 80.9% 86.2% 75.9% 0.817 0.8 Tsinalis et al. [51] 40 20-fold cross validation 70% 60% 73% 91% 74% 82% 74% 0.81 N/A Chambon et al. [52] 62 5-fold cross validation 85% 52% 77% 91% 83% 79% 77.6% 0.72 N/A Proposed model 5793 Train–validation–test 91% 46% 89% 77% 88% 86% 78% 0.81 0.82 Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy F1-Score Cohen’s kappa Supratak et al. [50] 62 31-fold cross validation 87.3% 43.5% 90.5% 77.1% 80.9% 86.2% 75.9% 0.817 0.8 Tsinalis et al. [51] 40 20-fold cross validation 70% 60% 73% 91% 74% 82% 74% 0.81 N/A Chambon et al. [52] 62 5-fold cross validation 85% 52% 77% 91% 83% 79% 77.6% 0.72 N/A Proposed model 5793 Train–validation–test 91% 46% 89% 77% 88% 86% 78% 0.81 0.82 Open in new tab Table 5. Performance of class balanced model compared to other studies Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy F1-Score Cohen’s kappa Supratak et al. [50] 62 31-fold cross validation 87.3% 43.5% 90.5% 77.1% 80.9% 86.2% 75.9% 0.817 0.8 Tsinalis et al. [51] 40 20-fold cross validation 70% 60% 73% 91% 74% 82% 74% 0.81 N/A Chambon et al. [52] 62 5-fold cross validation 85% 52% 77% 91% 83% 79% 77.6% 0.72 N/A Proposed model 5793 Train–validation–test 91% 46% 89% 77% 88% 86% 78% 0.81 0.82 Study Sample size (studies) Evaluation split W Accuracy N1 Accuracy N2 Accuracy N3 Accuracy REM Accuracy Overall Accuracy Balanced Accuracy F1-Score Cohen’s kappa Supratak et al. [50] 62 31-fold cross validation 87.3% 43.5% 90.5% 77.1% 80.9% 86.2% 75.9% 0.817 0.8 Tsinalis et al. [51] 40 20-fold cross validation 70% 60% 73% 91% 74% 82% 74% 0.81 N/A Chambon et al. [52] 62 5-fold cross validation 85% 52% 77% 91% 83% 79% 77.6% 0.72 N/A Proposed model 5793 Train–validation–test 91% 46% 89% 77% 88% 86% 78% 0.81 0.82 Open in new tab Figure 5. Open in new tabDownload slide Confusion matrix for all epochs. Figure 5. Open in new tabDownload slide Confusion matrix for all epochs. Figure 6. Open in new tabDownload slide Transition epoch confusion matrix. Figure 6. Open in new tabDownload slide Transition epoch confusion matrix. Figure 7. Open in new tabDownload slide Example output hypnogram of a PSG scored by the model overlaid on the human manual scoring. Figure 7. Open in new tabDownload slide Example output hypnogram of a PSG scored by the model overlaid on the human manual scoring. Performance on cohorts with and without sleep-disordered breathing The model performs similarly on subsets of the holdout set with different apnea severity (Table 6). A model trained and tested on severe OSA patients only achieved an unweighted F1-score of 0.846, similar to the model trained on heterogeneous data. Table 6. SHHS model performance on patient subgroups of varying obstructive sleep apnea severity Testing cohort F1 Epochs (N) All 0.872 621 794 Normal (AHI < 5) 0.871 132 742 Mild (5 < AHI < 15) 0.864 262 426 Moderate (15 < AHI < 30) 0.853 168 074 Severe (AHI > 30) 0.841 58 552 Testing cohort F1 Epochs (N) All 0.872 621 794 Normal (AHI < 5) 0.871 132 742 Mild (5 < AHI < 15) 0.864 262 426 Moderate (15 < AHI < 30) 0.853 168 074 Severe (AHI > 30) 0.841 58 552 AHI = apnea–hypopnea index. Open in new tab Table 6. SHHS model performance on patient subgroups of varying obstructive sleep apnea severity Testing cohort F1 Epochs (N) All 0.872 621 794 Normal (AHI < 5) 0.871 132 742 Mild (5 < AHI < 15) 0.864 262 426 Moderate (15 < AHI < 30) 0.853 168 074 Severe (AHI > 30) 0.841 58 552 Testing cohort F1 Epochs (N) All 0.872 621 794 Normal (AHI < 5) 0.871 132 742 Mild (5 < AHI < 15) 0.864 262 426 Moderate (15 < AHI < 30) 0.853 168 074 Severe (AHI > 30) 0.841 58 552 AHI = apnea–hypopnea index. Open in new tab Transfer learning After training on SHHS data, model generalizability was tested against two additional NSRR datasets. The microvolt mean and SD of each included data channel was significantly different between studies, suggesting different signal architectures between datasets (Table 7). Table 7. Mean and SD of the channels for each dataset Channel SHHS MrOS SOF EEG (uV) −0.39 ± 30.31 2.5 ± 38.08* −8.87 ± 43.02* EMG (uV) 0.54 ± 9.68 −1.06 ± 58.49* 10.05 ± 34.47* EOG(L) (uV) −3.57 ± 30.60 −12.5 ± 49.28* −9.81 ± 35.60* EOG(R) (uV) −4.19 ± 31.36 3.33 ± 50.81* 5.32 ± 41.37* Channel SHHS MrOS SOF EEG (uV) −0.39 ± 30.31 2.5 ± 38.08* −8.87 ± 43.02* EMG (uV) 0.54 ± 9.68 −1.06 ± 58.49* 10.05 ± 34.47* EOG(L) (uV) −3.57 ± 30.60 −12.5 ± 49.28* −9.81 ± 35.60* EOG(R) (uV) −4.19 ± 31.36 3.33 ± 50.81* 5.32 ± 41.37* *indicates significant difference from SHHS data at p < 0.05. Open in new tab Table 7. Mean and SD of the channels for each dataset Channel SHHS MrOS SOF EEG (uV) −0.39 ± 30.31 2.5 ± 38.08* −8.87 ± 43.02* EMG (uV) 0.54 ± 9.68 −1.06 ± 58.49* 10.05 ± 34.47* EOG(L) (uV) −3.57 ± 30.60 −12.5 ± 49.28* −9.81 ± 35.60* EOG(R) (uV) −4.19 ± 31.36 3.33 ± 50.81* 5.32 ± 41.37* Channel SHHS MrOS SOF EEG (uV) −0.39 ± 30.31 2.5 ± 38.08* −8.87 ± 43.02* EMG (uV) 0.54 ± 9.68 −1.06 ± 58.49* 10.05 ± 34.47* EOG(L) (uV) −3.57 ± 30.60 −12.5 ± 49.28* −9.81 ± 35.60* EOG(R) (uV) −4.19 ± 31.36 3.33 ± 50.81* 5.32 ± 41.37* *indicates significant difference from SHHS data at p < 0.05. Open in new tab F1-score and Cohen’s kappa scores on the MrOS and SOF datasets demonstrated moderate-to-strong inter-rater agreement between the model and trained scorers depending on the selected testing data and achieved high performance in the balance of precision and recall on sleep staging (Table 8). Table 8. Generalizability of the SHHS model to novel datasets Model F1-score (weighted) Cohen’s kappa Training data: SHHS Testing data: SHHS 0.87 0.82 Training data: SHHS Testing data: MrOS 0.79 0.70 Training data: SHHS Testing data: SOF 0.77 0.68 Training data: MrOS Testing data: SHHS 0.69 0.56 Training data: SOF Testing data: SHHS 0.66 0.53 Model F1-score (weighted) Cohen’s kappa Training data: SHHS Testing data: SHHS 0.87 0.82 Training data: SHHS Testing data: MrOS 0.79 0.70 Training data: SHHS Testing data: SOF 0.77 0.68 Training data: MrOS Testing data: SHHS 0.69 0.56 Training data: SOF Testing data: SHHS 0.66 0.53 Open in new tab Table 8. Generalizability of the SHHS model to novel datasets Model F1-score (weighted) Cohen’s kappa Training data: SHHS Testing data: SHHS 0.87 0.82 Training data: SHHS Testing data: MrOS 0.79 0.70 Training data: SHHS Testing data: SOF 0.77 0.68 Training data: MrOS Testing data: SHHS 0.69 0.56 Training data: SOF Testing data: SHHS 0.66 0.53 Model F1-score (weighted) Cohen’s kappa Training data: SHHS Testing data: SHHS 0.87 0.82 Training data: SHHS Testing data: MrOS 0.79 0.70 Training data: SHHS Testing data: SOF 0.77 0.68 Training data: MrOS Testing data: SHHS 0.69 0.56 Training data: SOF Testing data: SHHS 0.66 0.53 Open in new tab Discussion The deep learning model presented here automatically predicts sleep stage with moderate-to-strong agreement compared with expert human scorers across multiple datasets. The optimal model used input consisting of spectrograms derived from the EEG, EMG, and EOG channels passed to a deep learning architecture with convolutional and recurrent layers. A learning curve demonstrated that sufficient data was available to train the model well. The model performs comparably or better than other models reported in literature and, when tested against studies with structure similar to the underlying training dataset, meets or exceeds the accepted benchmark of K = 0.76 between trained human scorers. Spectrograms are used to represent the data provided to the model in the form of dimensionally reduced input that retains important information for sleep stage classification. The Fourier transforms used to generate spectrograms organized PSG data into component frequencies more easily compared across different platforms than raw signal data, which contains baseline signal noise and variation due to different recording environments and hardware. Spectrogram construction also aided network throughput as the volume of input data were reduced without significant loss of key signal information. Preprocessing raw signal data for noise and artifact reduction did not significantly affect classification results in preliminary testing. Prior performance analyses have demonstrated that deep learning models become more robust when trained on noisy data [53], and we suspect that training on raw, unprocessed data may be advantageous for accuracy and transferability when testing across clinical datasets as well. Noisy input data are hypothesized to improve the robustness of deep learning models by stabilizing against distortions in the input [54]. Networks trained on unprocessed data are better able to handle noise arising in unseen data. By training neural networks on unprocessed data, the need for preprocessing in new data is reduced and a greater proportion of relevant signal data can be preserved for analysis. PSGs have significant class imbalances between stage types due to the natural asymmetric distribution of sleep stages. The SHHS dataset is no exception, with large differences in representation between several of the stages. Accounting for class imbalances by overrepresenting minority classes (such as N1) can improve single class accuracy, but often at the expense of larger classes. For instance, in SHHS N1 is only 3.7% of the dataset, whereas N2 is 40.9%. The model presented here scored 31% of N1 and 91% of N2 epochs correctly with an overall accuracy of 87% when the native class imbalances are not adjusted. When N1 was oversampled to balance class representation, accuracy of N1 increased to 45% at the expense of other stages, such as N2, which decreased to an accuracy of 88%. Class balancing decreased overall model scoring accuracy to 86%. Class imbalances also complicate comparison of performance metrics between published models. We believe that preserving native class imbalances best represents how the model would perform in a production setting. However, performance metrics for models trained on natural as well as balanced class distributions are provided in order to facilitate comparison with previously published models (Tables 4 and 5). Accuracy in N1 scoring is worse than other sleep stages for this model, consistent with other published models [48–52]. This may be an artifact of PSG scoring rules, which allow for low-amplitude mixed pattern EEG signals identical to N1 to be scored as N2 if the preceding stage was also scored as N2. These rules, along with the large class imbalances between N1 and N2, likely compromise N1 accuracy. Other issues may complicate scoring accuracy, such as patient movement artifacts contaminating W and N1 stages. Unlike many other published works, this model was not trained on a curated set of high-quality PSGs and contains studies partially contaminated by signal and motion artifacts. Contaminated epochs scored by humans theoretically contain enough signal information that they should be of value in training a machine learning algorithm that will be exposed to similar data in a production environment. The inclusion of this more ambiguous data may create systemic difficulties in scoring W and N1 in the same way that it would degrade inter-rater agreement between human scorers. To this point, Younes et al. [13] recently found an intra-class correlation coefficient of 0.69 (range: 0.30–0.86) in N1 scoring, suggesting only poor to moderate agreement between trained human scorers. This model presented in this work has several strengths. It meets or exceeds performance of other published works. A large and diverse training dataset increased transferability, demonstrated across several other large datasets. Significant differences existed in mean microvolt channel levels across the tested datasets (Table 6), suggesting significant underlying differences in dataset structure due to differences in recording hardware, environment, study populations, or other variables. Despite these differences, the model presented here could be trained on one dataset and still perform with moderate-to-strong agreement on other datasets (MrOS F1 = 0.78, K = 0.68 and SOF F1 = 0.68, K = 0.55). The model also performed similarly on cohorts composed of subjects with varying degrees of sleep disordered breathing, with F1-scores ranging from 0.841 to 0.872, suggesting that sleep-disordered breathing does not significantly affect sleep stage classification patterns for the model. In comparison, a model trained only on patients with severe sleep apnea and tested on the same cohort performs only slightly better than one trained on all patients, demonstrating model transferability between different disease populations. Taken together, the transferability properties illustrated here suggest that automated deep learning classifiers have the potential for use in different clinical sleep laboratory environments without complete retraining on local data. Few other studies test models on PSGs collected from a variety of recording environments and hardware platforms. Patanaik et al. [55] did so, demonstrating generalizability by testing against two novel datasets with inter-rater agreement of K = 0.740 and K = 0.597. However, their reported outcomes (accuracy) were obtained from model training data instead of separate holdout data, limiting inner-dataset comparability to the work presented here. The kappa values are also not directly comparable to our inter-rater agreement of K = 0.70 and K = 0.56. The datasets in Patanaik et al. were acquired using the same framework and pipeline, whereas the external test datasets presented here were acquired on a variety of different hardware platforms that were then down- or up-sampled to match SHHS dataset frequencies. Both studies demonstrate comparable performance on external datasets that the models were not trained on, demonstrating transferability. This work is not without limitations. The datasets examined here are composed of Type II PSGs recorded in subject home environments with a limited, single EEG channel montage. Generalizability to more common Type I or Type III PSGs could not be evaluated; however, we suspect that training the model with additional EEG signals available in Type I PSGs would likely yield performance improvements from additional channel data. Retraining the model with additional channels while maintaining input from previously evaluated channels would be expected to improve performance, as deep neural networks generally perform better as more data is available [56]. Comparison with more limited montage datasets, such as consumer wearables using actigraphy and heart rate monitoring, is limited by the lack of large, publicly available datasets. In addition, accuracy outcomes may differ between AASM sleep staging criteria and RK staging criteria. In conclusion, this work suggests that automated PSG scoring systems can rapidly annotate PSG files with inter-rater agreement rivaling that of trained human scorers. Future work will require institutions and interested stakeholders to make available large libraries of high-quality datasets using modern scoring criteria in order for data scientists to develop robust, generalizable scoring models. Funding This work was supported by the NLM training grant 4T15LM 7450-15 from the National Institutes of Health. Conflict of interest statement. None declared. References 1. Rechtschaffen A , et al. A Manual of Standardized Terminology, Techniques and Scoring System for Sleep Stages of Human Subjects . Los Angeles, CA : National Institutes of Health Publication no. 204 ; 1968 . Google Preview WorldCat COPAC 2. Iber C. et al. The AASM manual for the scoring of sleep and associated events: Rules, terminology and technical specifications . Vol. 1 Westchester, IL : American Academy of Sleep Medicine ; 2007 . Google Preview WorldCat COPAC 3. Malhotra A , et al. Performance of an automated polysomnography scoring system versus computer-assisted manual scoring . Sleep. 2013 ; 36 ( 4 ): 573 – 582 . Google Scholar Crossref Search ADS PubMed WorldCat 4. Silber MH , et al. The visual scoring of sleep in adults . J Clin Sleep Med. 2007 ; 3 ( 2 ): 121 – 131 . Google Scholar PubMed WorldCat 5. Danker-Hopfe H , et al. Interrater reliability for sleep scoring according to the Rechtschaffen & Kales and the new AASM standard . J Sleep Res. 2009 ; 18 ( 1 ): 74 – 84 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Collop NA . Scoring variability between polysomnography technologists in different sleep laboratories . Sleep Med. 2002 ; 3 ( 1 ): 43 – 47 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Loredo JS , et al. Night-to-night arousal variability and interscorer reliability of arousal measurements . Sleep. 1999 ; 22 ( 7 ): 916 – 920 . Google Scholar Crossref Search ADS PubMed WorldCat 8. Norman RG , et al. Interobserver agreement among sleep scorers from different centers in a large dataset . Sleep. 2000 ; 23 ( 7 ): 901 – 908 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Bliwise D , et al. Measurement error in visually scored electrophysiological data: respiration during sleep . J Neurosci Methods. 1984 ; 12 ( 1 ): 49 – 56 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Lord S , et al. Interrater reliability of computer-assisted scoring of breathing during sleep . Sleep. 1989 ; 12 ( 6 ): 550 – 558 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Whitney CW , et al. Reliability of scoring respiratory disturbance indices and sleep staging . Sleep. 1998 ; 21 ( 7 ): 749 – 757 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Drinnan MJ , et al. Interobserver variability in recognizing arousal in respiratory sleep disorders . Am J Respir Crit Care Med. 1998 ; 158 ( 2 ): 358 – 362 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Younes M , et al. Reliability of the American Academy of Sleep Medicine rules for assessing sleep depth in clinical practice . J Clin Sleep Med. 2018 ; 14 ( 2 ): 205 – 213 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Schaltenbrand N , et al. Sleep stage scoring using the neural network model: comparison between visual and automatic analysis in normal subjects and patients . Sleep. 1996 ; 19 ( 1 ): 26 – 35 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Anderer P , et al. An E-health solution for automatic sleep classification according to Rechtschaffen and Kales: validation study of the Somnolyzer 24 × 7 utilizing the Siesta database . Neuropsychobiology. 2005 ; 51 ( 3 ): 115 – 133 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Berthomier C , et al. Automatic analysis of single-channel sleep EEG: validation in healthy individuals . Sleep. 2007 ; 30 ( 11 ): 1587 – 1595 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Anderer P , et al. Computer-assisted sleep classification according to the standard of the American Academy of Sleep Medicine: validation study of the AASM version of the Somnolyzer 24 × 7 . Neuropsychobiology. 2010 ; 62 ( 4 ): 250 – 264 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Fraiwan L , et al. Classification of sleep stages using multi-wavelet time frequency entropy and LDA . Methods Inf Med. 2010 ; 49 ( 3 ): 230 – 237 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Liang SF , et al. A rule-based automatic sleep staging method . J Neurosci Methods. 2012 ; 205 ( 1 ): 169 – 176 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Lajnef T , et al. Learning machines and sleeping brains: automatic sleep stage classification using decision-tree multi-class support vector machines . J Neurosci Methods. 2015 ; 250 : 94 – 105 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Wang Y , et al. Evaluation of an automated single-channel sleep staging algorithm . Nat Sci Sleep. 2015 ; 7 : 101 – 111 . Google Scholar PubMed WorldCat 22. Punjabi NM , et al. Computer-assisted automated scoring of polysomnograms using the somnolyzer system . Sleep. 2015 ; 38 ( 10 ): 1555 – 1566 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Hassan AR , et al. A decision support system for automatic sleep staging from EEG signals using tunable Q-factor wavelet transform and spectral features . J Neurosci Methods. 2016 ; 271 : 107 – 118 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Younes M , et al. Accuracy of automatic polysomnography scoring using frontal electrodes . J Clin Sleep Med. 2016 ; 12 ( 5 ): 735 – 746 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Younes M , et al. Performance of a new portable wireless sleep monitor . J Clin Sleep Med. 2017 ; 13 ( 2 ): 245 – 258 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Agarwal R , et al. Computer-assisted sleep staging . IEEE Trans Biomed Eng. 2001 ; 48 ( 12 ): 1412 – 1423 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Zoubek L , et al. Feature selection for sleep/wake stages classification using data driven methods . Biomed Signal Process Control . 2007 ; 2 ( 3 ): 171 – 179 . Google Scholar Crossref Search ADS WorldCat 28. Fraiwan L , et al. Automatic sleep stage scoring with wavelet packets based on single EEG recording . World Acad Sci Eng Technol. 2009 ; 54 ( 3 ): 485 – 488 . WorldCat 29. Bajaj V , et al. Automatic classification of sleep stages based on the time-frequency image of EEG signals . Comput Methods Programs Biomed . 2013 ; 112 ( 3 ): 320 – 328 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Chapotot F , et al. Automated sleep–wake staging combining robust feature extraction, artificial neural network classification, and flexible decision rules . Int J Adapt Control Signal Process. 2010 ; 24 ( 5 ): 409 – 423 . WorldCat 31. Figueroa Helland VC , et al. Investigation of an automatic sleep stage classification by means of multiscorer hypnogram . Methods Inf Med . 2010 ; 49 ( 5 ): 467 – 472 . Google Scholar Crossref Search ADS PubMed WorldCat 32. Jo HG , et al. Genetic fuzzy classifier for sleep stage identification . Comput Biol Med . 2010 ; 40 ( 7 ): 629 – 634 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Güneş S , et al. Efficient sleep stage recognition system based on EEG signal using k-means clustering based feature weighting . Expert Syst Appl . 2010 ; 37 ( 12 ): 7922 – 7928 . Google Scholar Crossref Search ADS WorldCat 34. Doroshenkov LG , et al. Classification of human sleep stages based on EEG processing using hidden Markov models . Biomed Eng . 2007 ; 41 ( 1 ): 25 – 28 . Google Scholar Crossref Search ADS WorldCat 35. Dong J , et al. Automated sleep staging technique based on the empirical mode decomposition algorithm: a preliminary study . Advances in Adaptive Data Analysis . 2010 ; 2 ( 02 ): 267 – 276 . Google Scholar Crossref Search ADS WorldCat 36. Koley B , et al. An ensemble system for automatic sleep stage classification using single channel EEG signal . Comput Biol Med. 2012 ; 42 ( 12 ): 1186 – 1195 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Hsu Y-L , et al. Automatic sleep stage recurrent neural classifier using energy features of EEG signals . Neurocomputing 2013 ; 104 : 105 – 114 . Google Scholar Crossref Search ADS WorldCat 38. Krakovská A , et al. Automatic sleep scoring: a search for an optimal combination of measures . Artif Intell Med . 2011 ; 53 ( 1 ): 25 – 33 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Krizhevsky A , et al. ImageNet classification with deep convolutional neural networks . NIPS. In: Advances in neural information processing systems . Dec 3-8 , 2012 ; Curran Associates , Red Hook, NY . 1106 – 1114 . WorldCat 40. Cecotti H , et al. Convolutional neural networks for P300 detection with application to brain-computer interfaces . IEEE Trans Pattern Anal Mach Intell. 2011 ; 33 ( 3 ): 433 – 445 . Google Scholar Crossref Search ADS PubMed WorldCat 41. Graves A , et al. Speech recognition with deep recurrent neural networks . In: IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . May 26-31 , 2013 ; IEEE , Piscataway, NJ . 2013 : 6645 – 6649 . 42. Budhiraja R , et al. The role of big data in the management of sleep-disordered breathing . Sleep Med Clin. 2016 ; 11 ( 2 ): 241 – 255 . Google Scholar Crossref Search ADS PubMed WorldCat 43. Dean DA II , et al. Scaling up scientific discovery in sleep medicine: the National Sleep Research Resource . Sleep. 2016 ; 39 ( 5 ): 1151 – 1164 . Google Scholar Crossref Search ADS PubMed WorldCat 44. Quan SF , et al. The Sleep Heart Health Study: design, rationale, and methods . Sleep . 1997 ; 12 : 1077 – 1085 . WorldCat 45. Redline S , et al. Methods for obtaining and analyzing unattended polysomnography data for a multicenter study. Sleep Heart Health Research Group . Sleep. 1998 ; 21 ( 7 ): 759 – 767 . Google Scholar Crossref Search ADS PubMed WorldCat 46. Pan SJ , et al. A survey on transfer learning . IEEE Trans Knowl Data Eng . 2010 ; 22 ( 10 ): 1345 – 1359 . Google Scholar Crossref Search ADS WorldCat 47. Biswal S , et al. SLEEPNET: automated sleep staging system Via deep learning. arXiv . 2017 : 1 – 17 . https://arxiv.org/pdf/1707.08262.pdf. Accessed January 1 , 2018 . 48. Sors A , et al. A convolutional neural network for sleep stage scoring from raw single-channel EEG . Biomed Signal Process Control . 2018 ; 42 : 107 – 114 . Google Scholar Crossref Search ADS WorldCat 49. Sharma M , et al. An accurate sleep stages classification system using a new class of optimally time-frequency localized three-band wavelet filter bank . Comput Biol Med. 2018 ; 98 : 58 – 75 . Google Scholar Crossref Search ADS PubMed WorldCat 50. Supratak A , et al. DeepSleepNet: a model for automatic sleep stage scoring based on raw single-channel EEG . IEEE Trans Neural Syst Rehabil Eng. 2017 ; 25 ( 11 ): 1998 – 2008 . Google Scholar Crossref Search ADS PubMed WorldCat 51. Tsinalis O , et al. Automatic sleep stage scoring with single-channel EEG using convolutional neural networks. arXiv . 2016 ; 1 – 10 . https://arxiv. org/abs/1610.01683. Accessed January 1 , 2018 . 52. Chambon S , et al. A deep learning architecture for temporal sleep stage classification using multivariate and multimodal time series . IEEE Trans Neural Syst Rehabil Eng. 2018 ; 26 ( 4 ): 758 – 769 . Google Scholar Crossref Search ADS PubMed WorldCat 53. Nazaré TS , et al. Deep convolutional neural networks and noisy images . InIberoamerican Congress on Pattern Recognition ; Nov 7, 2017 (pp. 416 – 424 ). Springer , Cham . 54. Stephan Z , et al. Improving the robustness of deep neural networks via stability training. In: proceedings of the IEEE Conference on Computer Vision and Pattern Recognition . Jun 26, 2016 – Jul 1, 2016 ; IEEE , Piscataway, NJ . 4480 – 4488 . 55. Patanaik A , et al. An end-to-end framework for real-time automatic sleep stage classification . Sleep . 2018 ; 41 ( 5 ): 1 – 11 . Google Scholar Crossref Search ADS WorldCat 56. Sun C , et al. Revisiting Unreasonable Effectiveness of Data in Deep Learning Era . In: ICCV . Oct 22-29, 2017; IEEE, Piscataway, NJ. Google Preview WorldCat COPAC © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Pomares, Florence, B;Boucetta,, Soufiane;Lachapelle,, Francis;Steffener,, Jason;Montplaisir,, Jacques;Cha,, Jungho;Kim,, Hosung;Dang-Vu, Thien, Thanh

doi: 10.1093/sleep/zsz156pmid: 31328786

Abstract Idiopathic hypersomnia (IH) is characterized by excessive daytime sleepiness but, in contrast to narcolepsy, does not involve cataplexy, sleep-onset REM periods, or any consistent hypocretin-1 deficiency. The pathophysiological mechanisms of IH remain unclear. Because of the involvement of the default-mode network (DMN) in alertness and sleep, our aim was to investigate the structural and functional modifications of the DMN in IH. We conducted multimodal magnetic resonance imaging (MRI) in 12 participants with IH and 15 good sleeper controls (mean age ± SD: 32 ± 9.6 years, range 22–53 years, nine males). Self-reported as well as objective measures of daytime sleepiness were collected. Gray matter volume and cortical thickness were analyzed to investigate brain structural differences between good sleepers and IH. Structural covariance and resting-state functional connectivity were analyzed to investigate changes in the DMN. Participants with IH had greater volume and cortical thickness in the precuneus, a posterior hub of the DMN. Cortical thickness in the left medial prefrontal cortex was positively correlated with thickness of the precuneus, and the strength of this correlation was greater in IH. In contrast, functional connectivity at rest was lower within the anterior DMN (medial prefrontal cortex) in IH, and correlated with self-reported daytime sleepiness. The present results show that IH is associated with structural and functional differences in the DMN, in proportion to the severity of daytime sleepiness, suggesting that a disruption of the DMN contributes to the clinical features of IH. Larger volume and thickness in this network might reflect compensatory changes to lower functional connectivity in IH. neuroimaging, narcolepsy, sleep and the brain, brain imaging, cortical activation, functional brain imaging Statement of Significance Idiopathic hypersomnia is a disorder characterized by excessive daytime sleepiness, and its underlying mechanisms remain largely unknown. Here, we evaluated with neuroimaging whether a major brain network involved in alertness (default-mode network [DMN]) was affected in idiopathic hypersomniacs. We showed that not only the structure but also the functional connections within the regions belonging to this network were altered. These changes suggest the involvement of the DMN in the daytime sleepiness encountered by patients with idiopathic hypersomnia. More studies are needed to investigate whether these modifications are specific to idiopathic hypersomnia as compared to other conditions characterized by excessive sleepiness during the day. Introduction Idiopathic hypersomnia (IH) is a central disorder of hypersomnolence which is characterized by excessive daytime sleepiness and difficulties waking up (called “sleep drunkenness”) stemming from an unknown cause. In contrast to narcolepsy, people suffering from IH do not present cataplexy, rapid REM sleep onset (at the multiple sleep latency test [MSLT]), or any consistent hypocretin-1 deficiency. The prevalence of IH remains unclear, but one report suggests it to be around 0.3% of the population [1]. IH has been shown associated with poor quality of life, cognitive impairments, and emotional disturbances [2]. As the pathophysiological mechanisms of IH remain unclear and there is no objective biomarker specific to the condition, it is often only diagnosed after exclusion of other disorders. The neuroimaging studies of IH remain scarce. No obvious structural brain differences were detected on single-subject magnetic resonance imaging (MRI) images [3], and only two neuroimaging studies have investigated functional changes in IH. One study, using fluorodeoxyglucose-positron emission tomography (FDG-PET), found that patients with IH showed hypermetabolism in the insula, anterior and middle cingulate cortex, and caudate nucleus compared to good sleepers [4]. In the other study, our group used single photon emission computed tomography (SPECT) to demonstrate that regional cerebral blood flow was lower in the medial prefrontal cortex as well as in the posterior cingulate cortex, putamen, and cerebellum of IH [5]. This hypometabolism in the medial prefrontal cortex was correlated with objective and self-reported measures of sleepiness, pointing to the involvement of the default-mode network (DMN) in the daytime sleepiness in IH. The DMN is composed of the medial prefrontal cortex, bilateral posterior cingulate cortex, precuneus, inferior parietal lobule, parts of the hippocampal formation, and the lateral temporal cortex [6]. Generally, the DMN is deactivated during tasks and activated at rest, and is involved in multiple cognitive processes such as higher cognition, emotion, and interoception [7]. Therefore, changes in its structure, activity, or connectivity have been found in a wide variety of neurological disorders [8]. The DMN is a key network involved in sleep [9, 10], during which its overall activity level decreases. In addition, while the activities between the posterior (namely the posterior cingulate cortex and inferior parietal lobe) and anterior (medial prefrontal cortex/anterior cingulate cortex) regions of the DMN are positively correlated at wake, they have been shown disconnected from each other during sleep [11]. This disconnection becomes more pronounced as sleep deepens: functional connectivity between posterior cingulate cortex and anterior DMN regions progressively decreases with the descent into sleep and reaches its lowest level during slow-wave sleep [9]. Given the importance of the DMN for sleep physiology and the previously reported changes in blood flow or glucose metabolism in regions belonging to the DMN with IH, we therefore aimed at investigating the structural and functional differences (including connectivity) in the DMN using multimodal MRI in a group of participants with IH compared to good sleepers. We hypothesized that participants with IH would display changes in volume, thickness, and connectivity between nodes of the DMN compared to healthy controls. Methods Participants Eighteen participants with IH were recruited from several sleep clinics in the Montreal area, as well as through advertisements in local patients’ associations. Participants with IH underwent a MSLT to evaluate mean sleep latency during the day as well as the presence of sleep-onset REM periods (SOREMPs). Inclusion criteria for IH participants were the following: (1) excessive daytime sleepiness present for at least 3 months; (2) daytime mean sleep latency < 8 minutes based on MSLT OR self-reported 24-hour total sleep time was > 11 hours; (3) absence of cataplexy; (4) number of SOREMPs < 2; (5) absence of other causes of hypersomnia (e.g. other sleep or neurological disorders, use of drugs or medications). These criteria are for the most part in line with the ICSD-3 [12] diagnostic criteria of IH, except item 2) on 24-hour total sleep time > 11 hours, for which ICSD-3 requires confirmation by actigraphy or ad libitum polysomnography. This criterion only concerned two of our IH participants, as all the others had a mean sleep latency at the MSLT < 8 minutes. Psychotropic medications that could influence sleep and alertness (e.g. psychostimulants) were withdrawn 2 weeks prior to the start of the protocol and during the whole study procedure. Seventeen good sleepers were also recruited for the study as healthy control participants. Recruitment was conducted through local advertisements. The following exclusion criteria were applied for all the participants: (1) sleep disorders, other than IH, as assessed by a semi-structured interview and polysomnography (e.g. sleep apnea-hypopnea syndrome with apnea–hypopnea index > 5/hour); (2) systemic or neurological diseases such as diabetes, hypertension, dementia, stroke, and epilepsy; (3) shift or night work; (4) psychiatric disorders according to the DSM-5 [13]; (5) history of head injury, encephalopathy, or intracranial surgery; (6) history of alcoholism or drug abuse; and contraindication to the imaging. The study protocol was approved by ethics committees of the Hôpital du Sacré-Coeur de Montréal and the Regroupement Neuroimagerie Quebec, Centre de Recherche de l’Institut Universitaire de Gériatrie de Montréal (CRIUGM), and all participants provided written informed consent. Clinical variables Demographic variables (age, sex, and level of education) and the self-reported onset of symptoms were collected. Daytime mean sleep latency (as an objective measure of sleepiness) and the number of SOREMPs were extracted from the MSLT. Additionally, self-reported daytime sleepiness was measured with the Epworth sleepiness scale (ESS) [14]. Other clinical variables included self-reported measures of general sleep quality assessed with the Pittsburgh Sleep Quality Index (PSQI) [15], chronotype assessed with the Morningness-Evening Questionnaire (MEQ) [16], anxiety assessed with the Beck Anxiety Inventory (BAI) [17] and depression assessed with the Beck Depression Inventory (BDI) [18]. Clinical variables were compared between groups using independent sample t-tests and some of them were further used to inform the significance of imaging data in whole-brain regression analyses detailed in the following sections. Given that depression has been associated with hypersomnolence symptoms [19] with an increase in depression symptoms previously reported in hypersomnia disorders [20], and because the DMN has been shown affected with depression [21, 22], all the analyses were controlled for depression scores obtained with the BDI to disentangle the anatomical and functional effects related to depression symptoms from those of IH. The analyses uncontrolled for depression scores are reported in the Supplementary Material. Imaging acquisition Anatomical MRI was conducted on a 3 Tesla Siemens Trio scanner, with a 32-channel head coil. The scanning procedure took place at the Unité de Neuroimagerie Fonctionnelle located at the CRIUGM during the afternoon (between 1 pm and 5 pm), starting with the resting-state sequence followed by the anatomical sequence. The resting-state functional MRI (fMRI) sequence was acquired with multi-slice T2*-weighted images with the following parameters: repetition time (TR) 2.6 seconds, echo time (TE) 30 ms, flip angle 90°, field of view (FoV) 218 mm, matrix size 64 × 64, 150 volumes, 42 slices, resolution 3.4 mm isotropic, in order to compute connectivity in the DMN. During the resting-state fMRI sequence, which lasted 6 minutes, participants were asked to keep their eyes open, not fall asleep, and fixate a dot at the center of the screen. The maintenance of wakefulness was verified using a camera with a focus on participant’s eyes, under the constant monitoring of a research assistant (F.L.) who ensured that participants kept their eyes open during resting-state fMRI. If participants showed signs of sleepiness on the camera (e.g. eyes starting to close) during the resting-state session, the research assistant talked to them using the interphone to remind them to keep their eyes open. Such reminders were only required in a minority of participants, at one or two instances maximum per session. High-resolution anatomical T1-weighted imaging (multi-echo magnetization-prepared rapid acquisition with gradient echo [MP-RAGE], TR 2.53 seconds, TE 1.64 ms, flip angle 7°, FoV 256 mm, matrix size 256 × 256, 176 slices, resolution 1 mm isotropic) was used to perform voxel-based morphometry [23] as well as measure cortical thickness. Visual quality check of all the MRI data was conducted prior to any analysis. Some of these participants also underwent a SPECT scan, and SPECT results were previously reported in Boucetta et al. [5]. Gray matter volume analysis To investigate subcortical as well as cortical regional gray matter volume, voxel-based morphometry was performed using the Computational Anatomy Toolbox (CAT12) [24] in Statistical Parametric Mapping software (SPM12, Wellcome Trust Centre for Neuroimaging). The following steps were used for data processing: (1) images were normalized to standard space by linearly registering them to the International Consortium for Brain Mapping (ICBM152) template; (2) segmented into gray matter, white matter, and cerebrospinal fluid using the intensity distribution of the images; (3) gray matter probability maps were modulated, i.e. the value at each voxel was multiplied by the Jacobian determinant accounting for the amount of contraction or retraction led by the non-linear spatial normalization to obtain regional volumes; (4) the modulated normalized gray matter maps were spatially blurred with an 8 × 8 × 8 mm (full width at half maximum) Gaussian smoothing kernel. Whole-brain voxel-wise group differences in gray matter volume were assessed using a general linear model (GLM), controlling for total intracranial volume (gray matter + white matter + CSF), age, sex and depression score from the BDI. Separate linear regression analyses at the whole group level were conducted with self-reported sleepiness based on ESS scores and objective sleepiness score from mean sleep latency at the MSLT, controlling for age, sex, and BDI. Cortical thickness analysis Cortical thickness, corresponding to the distance between the inner cortical surface and the outer/pial surface, was computed using the CIVET pipeline version 2.1.0 [25–31] running on CBRAIN [32, 33]. The following steps were used for data processing: (1) anatomical images were corrected for image intensity inhomogeneities with N3 correction [26], linearly and nonlinearly registered to the Montreal Neurological Institute (MNI) standard space using the ICBM152 nonlinear 6th generation template [34] with linear transformations; (2) images in standard space were segmented into gray matter, white matter and cerebrospinal fluid with INSECT (Intensity Normalized Stereotaxic Environment for the Classification of Tissue) [27]; (3) surface extraction was computed using the “marching cubes” algorithm to produce surfaces without surface bridges from the segmented images, and a 30-mm surface-based smoothing was applied [35]; (4) each cortical surface was regularized to 40 962 vertices by icosahedron resampling and a spherical registration to the template surface [36]. The resampled surface was transformed back into the participant’s native space, then used to calculate cortical thickness [35–37]. Statistical analysis of cortical thickness was performed with SurfStat (http://www.math.mcgill.ca/keith/surfstat/, accessed July 11, 2019). Whole-brain group differences in cortical thickness were assessed using a GLM, controlling for age, sex, and depression score from the BDI. Separate linear regression analyses at the whole group level were conducted with ESS and mean sleep latency at the MSLT, controlling for age, sex, and BDI. Structural covariance analysis Structural covariance analysis allows estimating morphological similarity between cortical regions, based on their cortical thickness. Structural connectivity explains a substantial portion of inter-regional structural covariance within the brain [38]. This assumes that regions with related/similar function show greater correlation with each other, i.e. greater structural covariance [39]. The correlation between cortical thicknesses at a given seed vertex (Thickness@seed) and cortical thickness measured at each of all other vertices was computed in SurfStat for both the IH and the good sleeper participants. The seed was located in the posterior cingulate cortex/precuneus (MNI coordinates x = −14, y = −58, z = 32) and in the medial prefrontal cortex (MNI coordinates x = −2, y = 44, z = 16), based on the SPECT results from ref. [5]. Statistical analysis of cortical thickness was performed with SurfStat. By fitting a linear model and assessing the interaction effect of group (IH or good sleeper) × Thickness@seed [40], we compared the difference in the structural covariance between groups controlling for age, sex, and BDI. Separate linear regression analyses at the whole group level were conducted with ESS and mean sleep latency at the MSLT, controlling for age, sex, and depression score from the BDI. Resting-state functional connectivity analysis Resting-state functional connectivity measures the degree of correlation between the time course of fMRI signals. For this analysis, fMRI data was preprocessed using the Neuroimaging Analysis Kit (NIAK) version 1.0.1 [41], http://niak.simexp-lab.org, accessed July 11, 2019. Each fMRI dataset was corrected for inter-slice difference in acquisition time and the parameters of a rigid-body motion were estimated for each time frame. Rigid-body motion was estimated within as well as between runs. The median volume of one selected fMRI run for each participant was coregistered with a T1 individual scan using Minctracc [42], which was itself nonlinearly transformed to the MNI template [43] using the CIVET pipeline. The following steps were used: (1) registration to the MNI space was conducted using the ICBM152 nonlinear 6th generation symmetric template; the rigid-body, fMRI-to-T1 transform and T1-to-MNI transform were all combined, and the functional volumes were resampled in the MNI space at a 4-mm isotropic resolution; (2) “scrubbing” method was used to remove the volumes with excessive motion (frame displacement greater than 0.5) [44]; (3) the following nuisance parameters were regressed out from the time series at each voxel: slow time drifts (basis of discrete cosines with a 0.01 Hz high-pass cut-off), average signals in conservative masks of the white matter and the lateral ventricles as well as the first principal components (95% energy) of the six rigid-body motion parameters and their squares [45, 46]; (4) fMRI volumes were finally spatially smoothed with an 8-mm isotropic Gaussian blurring kernel. Based on the SPECT findings [5] the average timeseries used for the functional connectivity analysis were extracted separately (masks) for the anterior and posterior DMN. Timeseries from all the voxels within each DMN mask were averaged and the average timeseries for each mask was correlated with all the voxels in the brain (including voxels within the DMN), resulting in a voxel-wise functional connectivity map (representing the Pearson’s correlation corrected by the Fisher transform for each participant). The DMN masks were extracted from the 12 distributed networks using the Cambridge parcellation, the anterior DMN mask comprising the anterior cingulate cortex, medial prefrontal cortex and superior frontal cortex, and the posterior DMN mask comprising the posterior cingulate cortex and precuneus (Supplementary Figure 1). The Cambridge parcellation is a derivative from the Cambridge sample found in the “1000 functional connectome project” (http://fcon_1000.projects.nitrc.org/fcpClassic/FcpTable.html, accessed July 11, 2019) [48], in which brain parcellations were generated from 200 young healthy participants resting-state fMRI images, using a method called bootstrap analysis of stable clusters (BASC). Statistical analysis of functional connectivity was performed with SPM12. Group differences in functional connectivity were assessed using a GLM, controlling for age, sex, and depression score from the BDI. Separate linear regression analyses at the whole group level were conducted with ESS and mean sleep latency at the MSLT, controlling for age, sex, and BDI. Correction for multiple comparisons Voxel-wise and vertex-wise statistical analyses were corrected for multiple comparisons using the random field theory for non-isotropic images [47]. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, which is the minimum height threshold recommended [49]. An extent threshold of p < 0.05 corrected for multiple comparisons was then applied at the cluster level. Images were overlaid on the average anatomical MRI scan in MNI standard space from all the participants. Results Clinical variables Six out of 18 participants with IH were not included in the study, because five had comorbid clinical depression and one did not fulfill the ICSD-3 criteria. In addition, two good sleeper participants were not included in the analysis: one participant because of contraindication to the MRI scan but did complete the SPECT scan; another participant because they were identified as an outlier in the VBM analysis (large and asymmetric ventricles). Therefore, the final sample comprised 12 participants with IH and 15 good sleepers. Ten out of the 12 participants with IH had a history of daily long sleep time (i.e., 24-hour total sleep time > 11 hours). Demographic and clinical variables from all participants are presented in Table 1. Participants with IH had significantly higher scores on ESS and PSQI, indicating higher self-reported daytime sleepiness and more complaint about their sleep—especially its impact on daytime function. The difference in PSQI scores was indeed driven by the daytime dysfunction subscale (mean score IH: 2.3, GS:0.3, t-value = 9.9, p < 0.001). Although the clinical threshold for depression and anxiety was not reached, participants with IH had significantly higher scores in both BDI and BAI. No group difference was detected on chronotype, polysomnography measures, and demographic variables. Table 1. Participants’ characteristics including demographics, polysomnography measures, questionnaire scores, and brain measures Clinical variables Good sleepers Idiopathic hypersomnia Mean (SD) Mean (SD) P-value Demographics  N 15 12  Ratio male/female 6M/9F 3M/9F 0.083  Age 31.2 (9.8) 33.4 (10.1) 0.523  BMI (kg/m2) 23.52 (2.35) 23.77 (4.56) 0.862  Education (years) 16.53 (1.88) 15.75 (2.83) 0.397  Symptoms duration (years) 12.0 (8.7) Multiple Sleep Latency Test  Latency at MSLT (minutes) 7.28 (3.34)  Number of SOREMPs at MSLT 0.25 (0.45) Polysomnography  TST (minutes) 420.88 (59.11) 455.46 (40.49) 0.097  Mean sleep latency (minutes) 13.21 (7.65) 13.34 (8.79) 0.966  REM latency (minutes) 117.85 (47.02) 97.88 (33.55) 0.294  sleep efficiency (%) 91.01 (3.95) 91.52 (5.49) 0.784  WASO (minutes) 23.22 (17.04) 32.32 (20.59) 0.220  REM (minutes) 100.64 (38.61) 90.42 (38.26) 0.550 Questionnaire scores  ESS (/24) 4.9 (2.4) 17.5 (4.4) <0.0001***  PSQI (/21) 3.1 (1.2) 4.9 (1.0) <0.0001***  MEQ (/86) 54.7 (8.9) 49.2 (8.2) 0.111  BDI (/63) 2.7 (3.0) 10.4 (6.8) 0.001*  BAI (/63) 2.4 (3.3) 10.2 (9.9) 0.008* Brain measures  TIV (mm3) 1442 (123) 1486 (145) 0.402  GM (mm3) 674 (44) 681 (61) 0.743  WM (mm3) 491 (47) 519 (62) 0.205  CSF (mm3) 276 (50) 286 (38) 0.559 Clinical variables Good sleepers Idiopathic hypersomnia Mean (SD) Mean (SD) P-value Demographics  N 15 12  Ratio male/female 6M/9F 3M/9F 0.083  Age 31.2 (9.8) 33.4 (10.1) 0.523  BMI (kg/m2) 23.52 (2.35) 23.77 (4.56) 0.862  Education (years) 16.53 (1.88) 15.75 (2.83) 0.397  Symptoms duration (years) 12.0 (8.7) Multiple Sleep Latency Test  Latency at MSLT (minutes) 7.28 (3.34)  Number of SOREMPs at MSLT 0.25 (0.45) Polysomnography  TST (minutes) 420.88 (59.11) 455.46 (40.49) 0.097  Mean sleep latency (minutes) 13.21 (7.65) 13.34 (8.79) 0.966  REM latency (minutes) 117.85 (47.02) 97.88 (33.55) 0.294  sleep efficiency (%) 91.01 (3.95) 91.52 (5.49) 0.784  WASO (minutes) 23.22 (17.04) 32.32 (20.59) 0.220  REM (minutes) 100.64 (38.61) 90.42 (38.26) 0.550 Questionnaire scores  ESS (/24) 4.9 (2.4) 17.5 (4.4) <0.0001***  PSQI (/21) 3.1 (1.2) 4.9 (1.0) <0.0001***  MEQ (/86) 54.7 (8.9) 49.2 (8.2) 0.111  BDI (/63) 2.7 (3.0) 10.4 (6.8) 0.001*  BAI (/63) 2.4 (3.3) 10.2 (9.9) 0.008* Brain measures  TIV (mm3) 1442 (123) 1486 (145) 0.402  GM (mm3) 674 (44) 681 (61) 0.743  WM (mm3) 491 (47) 519 (62) 0.205  CSF (mm3) 276 (50) 286 (38) 0.559 BMI: body mass index, MSLT: multiple sleep latency test; TST: total sleep time; WASO: wake after sleep onset; ESS: Epworth Sleepiness Scale; PSQI: Pittsburgh Sleep Quality Index; MEQ: Morningness-Eveningness Questionnaire; BDI: Beck Depression Inventory; BAI: Beck Anxiety Inventory; TIV: total intracranial volume; GM: gray matter; WM: white matter; CSF: cerebrospinal fluid. P-values from Student’s t-test comparing groups are reported. ***p <0.001. *p <0.05. Open in new tab Table 1. Participants’ characteristics including demographics, polysomnography measures, questionnaire scores, and brain measures Clinical variables Good sleepers Idiopathic hypersomnia Mean (SD) Mean (SD) P-value Demographics  N 15 12  Ratio male/female 6M/9F 3M/9F 0.083  Age 31.2 (9.8) 33.4 (10.1) 0.523  BMI (kg/m2) 23.52 (2.35) 23.77 (4.56) 0.862  Education (years) 16.53 (1.88) 15.75 (2.83) 0.397  Symptoms duration (years) 12.0 (8.7) Multiple Sleep Latency Test  Latency at MSLT (minutes) 7.28 (3.34)  Number of SOREMPs at MSLT 0.25 (0.45) Polysomnography  TST (minutes) 420.88 (59.11) 455.46 (40.49) 0.097  Mean sleep latency (minutes) 13.21 (7.65) 13.34 (8.79) 0.966  REM latency (minutes) 117.85 (47.02) 97.88 (33.55) 0.294  sleep efficiency (%) 91.01 (3.95) 91.52 (5.49) 0.784  WASO (minutes) 23.22 (17.04) 32.32 (20.59) 0.220  REM (minutes) 100.64 (38.61) 90.42 (38.26) 0.550 Questionnaire scores  ESS (/24) 4.9 (2.4) 17.5 (4.4) <0.0001***  PSQI (/21) 3.1 (1.2) 4.9 (1.0) <0.0001***  MEQ (/86) 54.7 (8.9) 49.2 (8.2) 0.111  BDI (/63) 2.7 (3.0) 10.4 (6.8) 0.001*  BAI (/63) 2.4 (3.3) 10.2 (9.9) 0.008* Brain measures  TIV (mm3) 1442 (123) 1486 (145) 0.402  GM (mm3) 674 (44) 681 (61) 0.743  WM (mm3) 491 (47) 519 (62) 0.205  CSF (mm3) 276 (50) 286 (38) 0.559 Clinical variables Good sleepers Idiopathic hypersomnia Mean (SD) Mean (SD) P-value Demographics  N 15 12  Ratio male/female 6M/9F 3M/9F 0.083  Age 31.2 (9.8) 33.4 (10.1) 0.523  BMI (kg/m2) 23.52 (2.35) 23.77 (4.56) 0.862  Education (years) 16.53 (1.88) 15.75 (2.83) 0.397  Symptoms duration (years) 12.0 (8.7) Multiple Sleep Latency Test  Latency at MSLT (minutes) 7.28 (3.34)  Number of SOREMPs at MSLT 0.25 (0.45) Polysomnography  TST (minutes) 420.88 (59.11) 455.46 (40.49) 0.097  Mean sleep latency (minutes) 13.21 (7.65) 13.34 (8.79) 0.966  REM latency (minutes) 117.85 (47.02) 97.88 (33.55) 0.294  sleep efficiency (%) 91.01 (3.95) 91.52 (5.49) 0.784  WASO (minutes) 23.22 (17.04) 32.32 (20.59) 0.220  REM (minutes) 100.64 (38.61) 90.42 (38.26) 0.550 Questionnaire scores  ESS (/24) 4.9 (2.4) 17.5 (4.4) <0.0001***  PSQI (/21) 3.1 (1.2) 4.9 (1.0) <0.0001***  MEQ (/86) 54.7 (8.9) 49.2 (8.2) 0.111  BDI (/63) 2.7 (3.0) 10.4 (6.8) 0.001*  BAI (/63) 2.4 (3.3) 10.2 (9.9) 0.008* Brain measures  TIV (mm3) 1442 (123) 1486 (145) 0.402  GM (mm3) 674 (44) 681 (61) 0.743  WM (mm3) 491 (47) 519 (62) 0.205  CSF (mm3) 276 (50) 286 (38) 0.559 BMI: body mass index, MSLT: multiple sleep latency test; TST: total sleep time; WASO: wake after sleep onset; ESS: Epworth Sleepiness Scale; PSQI: Pittsburgh Sleep Quality Index; MEQ: Morningness-Eveningness Questionnaire; BDI: Beck Depression Inventory; BAI: Beck Anxiety Inventory; TIV: total intracranial volume; GM: gray matter; WM: white matter; CSF: cerebrospinal fluid. P-values from Student’s t-test comparing groups are reported. ***p <0.001. *p <0.05. Open in new tab Gray matter volume There was no significant difference in overall gray matter, white matter or CSF volume between IH and good sleepers (Table 1). Participants with IH had smaller inferior frontal gyrus compared to good sleepers (Figure 1, left panel; Table 2). Participants with IH had larger gray matter volume in the left middle occipital gyrus and right precuneus, both of which belong to the posterior DMN (Figure 1, right panel; Table 2). Table 2. Statistics table of whole-brain analyses Region Side Cluster size Cluster p-value T-value Z-value equivalent Voxel p-value MNI coordinates (x, y, z) GM: Good Sleepers > Idiopathic Hypersomnia  Inferior frontal gyrus Left 233 0.044 4.48 3.71 <0.001 −57 21 12 4.40 3.66 <0.001 −48 22 10 GM: Idiopathic Hypersomnia > Good Sleepers  Middle occipital gyrus Left 500 <0.001 6.74 4.86 <0.001 −6 −69 14 4.19 3.53 <0.001 −12 −80 0  Precuneus Right 264 0.025 5.09 4.06 <0.001 2 −54 52 GM: Regression with subjective daytime sleepiness score (positive)  Middle occipital gyrus Left 343 0.006 6.37 4.70 0.000 −9 −68 14 CT: Idiopathic Hypersomnia > Good Sleepers  Precuneus Left 2403 0.046 3.75 <0.001 −14 −64 33 SC from MPFC: Interaction  Precuneus Left 7145 0.005 3.75 <0.001 −14 −64 33 FC: Good Sleepers > Idiopathic Hypersomnia  Orbitofrontal cortex Right 62 0.079 6.41 4.77 <0.001 6 36 −28 FC: Regression with subjective daytime sleepiness score (negative)  Fusiform gyrus Left 73 0.045 5.21 4.16 <0.001 −50 −24 −24  Medial prefrontal cortex Left 74 0.043 4.86 3.97 <0.001 −10 52 4 4.41 3.69 <0.001 6 56 4  Orbitofrontal cortex Right 95 0.016 4.67 3.85 <0.001 6 28 −24 4.65 3.84 <0.001 6 8 −12 4.54 3.77 <0.001 6 40 −24 Region Side Cluster size Cluster p-value T-value Z-value equivalent Voxel p-value MNI coordinates (x, y, z) GM: Good Sleepers > Idiopathic Hypersomnia  Inferior frontal gyrus Left 233 0.044 4.48 3.71 <0.001 −57 21 12 4.40 3.66 <0.001 −48 22 10 GM: Idiopathic Hypersomnia > Good Sleepers  Middle occipital gyrus Left 500 <0.001 6.74 4.86 <0.001 −6 −69 14 4.19 3.53 <0.001 −12 −80 0  Precuneus Right 264 0.025 5.09 4.06 <0.001 2 −54 52 GM: Regression with subjective daytime sleepiness score (positive)  Middle occipital gyrus Left 343 0.006 6.37 4.70 0.000 −9 −68 14 CT: Idiopathic Hypersomnia > Good Sleepers  Precuneus Left 2403 0.046 3.75 <0.001 −14 −64 33 SC from MPFC: Interaction  Precuneus Left 7145 0.005 3.75 <0.001 −14 −64 33 FC: Good Sleepers > Idiopathic Hypersomnia  Orbitofrontal cortex Right 62 0.079 6.41 4.77 <0.001 6 36 −28 FC: Regression with subjective daytime sleepiness score (negative)  Fusiform gyrus Left 73 0.045 5.21 4.16 <0.001 −50 −24 −24  Medial prefrontal cortex Left 74 0.043 4.86 3.97 <0.001 −10 52 4 4.41 3.69 <0.001 6 56 4  Orbitofrontal cortex Right 95 0.016 4.67 3.85 <0.001 6 28 −24 4.65 3.84 <0.001 6 8 −12 4.54 3.77 <0.001 6 40 −24 Statistics table of the group differences and regressions with self-reported daytime sleepiness measured with the ESS at the whole group level for gray matter volume (GM), cortical thickness (CT), structural covariance (SC) and functional connectivity (FC) controlling for age, sex and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Open in new tab Table 2. Statistics table of whole-brain analyses Region Side Cluster size Cluster p-value T-value Z-value equivalent Voxel p-value MNI coordinates (x, y, z) GM: Good Sleepers > Idiopathic Hypersomnia  Inferior frontal gyrus Left 233 0.044 4.48 3.71 <0.001 −57 21 12 4.40 3.66 <0.001 −48 22 10 GM: Idiopathic Hypersomnia > Good Sleepers  Middle occipital gyrus Left 500 <0.001 6.74 4.86 <0.001 −6 −69 14 4.19 3.53 <0.001 −12 −80 0  Precuneus Right 264 0.025 5.09 4.06 <0.001 2 −54 52 GM: Regression with subjective daytime sleepiness score (positive)  Middle occipital gyrus Left 343 0.006 6.37 4.70 0.000 −9 −68 14 CT: Idiopathic Hypersomnia > Good Sleepers  Precuneus Left 2403 0.046 3.75 <0.001 −14 −64 33 SC from MPFC: Interaction  Precuneus Left 7145 0.005 3.75 <0.001 −14 −64 33 FC: Good Sleepers > Idiopathic Hypersomnia  Orbitofrontal cortex Right 62 0.079 6.41 4.77 <0.001 6 36 −28 FC: Regression with subjective daytime sleepiness score (negative)  Fusiform gyrus Left 73 0.045 5.21 4.16 <0.001 −50 −24 −24  Medial prefrontal cortex Left 74 0.043 4.86 3.97 <0.001 −10 52 4 4.41 3.69 <0.001 6 56 4  Orbitofrontal cortex Right 95 0.016 4.67 3.85 <0.001 6 28 −24 4.65 3.84 <0.001 6 8 −12 4.54 3.77 <0.001 6 40 −24 Region Side Cluster size Cluster p-value T-value Z-value equivalent Voxel p-value MNI coordinates (x, y, z) GM: Good Sleepers > Idiopathic Hypersomnia  Inferior frontal gyrus Left 233 0.044 4.48 3.71 <0.001 −57 21 12 4.40 3.66 <0.001 −48 22 10 GM: Idiopathic Hypersomnia > Good Sleepers  Middle occipital gyrus Left 500 <0.001 6.74 4.86 <0.001 −6 −69 14 4.19 3.53 <0.001 −12 −80 0  Precuneus Right 264 0.025 5.09 4.06 <0.001 2 −54 52 GM: Regression with subjective daytime sleepiness score (positive)  Middle occipital gyrus Left 343 0.006 6.37 4.70 0.000 −9 −68 14 CT: Idiopathic Hypersomnia > Good Sleepers  Precuneus Left 2403 0.046 3.75 <0.001 −14 −64 33 SC from MPFC: Interaction  Precuneus Left 7145 0.005 3.75 <0.001 −14 −64 33 FC: Good Sleepers > Idiopathic Hypersomnia  Orbitofrontal cortex Right 62 0.079 6.41 4.77 <0.001 6 36 −28 FC: Regression with subjective daytime sleepiness score (negative)  Fusiform gyrus Left 73 0.045 5.21 4.16 <0.001 −50 −24 −24  Medial prefrontal cortex Left 74 0.043 4.86 3.97 <0.001 −10 52 4 4.41 3.69 <0.001 6 56 4  Orbitofrontal cortex Right 95 0.016 4.67 3.85 <0.001 6 28 −24 4.65 3.84 <0.001 6 8 −12 4.54 3.77 <0.001 6 40 −24 Statistics table of the group differences and regressions with self-reported daytime sleepiness measured with the ESS at the whole group level for gray matter volume (GM), cortical thickness (CT), structural covariance (SC) and functional connectivity (FC) controlling for age, sex and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Open in new tab Figure 1. Open in new tabDownload slide Group differences in gray matter volume. The left panel depicts regions with lower volume in idiopathic hypersomnia compared to good sleepers. The right panel depicts regions with greater volume in hypersomnia compared to good sleepers. The figures represent T-maps. T-map of group difference in gray matter volume controlling for age, sex and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, R: right, A: anterior, P: posterior, TIV: total intracranial volume, X,Y,Z represent coordinates in MNI standard space. Figure 1. Open in new tabDownload slide Group differences in gray matter volume. The left panel depicts regions with lower volume in idiopathic hypersomnia compared to good sleepers. The right panel depicts regions with greater volume in hypersomnia compared to good sleepers. The figures represent T-maps. T-map of group difference in gray matter volume controlling for age, sex and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, R: right, A: anterior, P: posterior, TIV: total intracranial volume, X,Y,Z represent coordinates in MNI standard space. The voxel-wise regression showed no significant correlation with the DMN, but there was a significant positive correlation between gray matter volume in the middle occipital gyrus and self-reported daytime sleepiness (Table 2). No significant voxel-wise association was found between gray matter volume and objective daytime sleepiness from the MSLT in IH. When no correction for depression score was applied, no region showed smaller gray matter volume in IH compared to good sleepers. Participants with IH had larger gray matter volume in the precuneus compared to good sleepers, which corresponds to the posterior part of the DMN (Supplementary Figure 2, Supplementary Table 1). The voxel-wise regression showed a significant positive correlation between gray matter volume in the precuneus and subjective daytime sleepiness (Supplementary Figure 3, Supplementary Table 1). Cortical thickness Participants with IH had thicker precuneus, which corresponds to the posterior DMN; no regions showing lower cortical thickness were observed (Figure 2, Table 2). Figure 2. Open in new tabDownload slide Cortical thickness maps. The left panel depicts the T-map of cortical thickness differences between idiopathic hypersomnia and good sleepers. Cold colors represent thinner cortex and warm colors represent thicker cortex in idiopathic hypersomnia compared to good sleepers. The right panel depicts the p-map of regions with significantly thicker cortex in hypersomnia compared to good sleepers, controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied at the vertex level, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, A: anterior, P: posterior. Figure 2. Open in new tabDownload slide Cortical thickness maps. The left panel depicts the T-map of cortical thickness differences between idiopathic hypersomnia and good sleepers. Cold colors represent thinner cortex and warm colors represent thicker cortex in idiopathic hypersomnia compared to good sleepers. The right panel depicts the p-map of regions with significantly thicker cortex in hypersomnia compared to good sleepers, controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied at the vertex level, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, A: anterior, P: posterior. The vertex-wise regression showed no significant association between cortical thickness and subjective or objective daytime sleepiness. No significant cortical thickness difference between groups was observed without controlling for depression scores. Structural covariance Structural covariance results showed that, in both groups, thickness of the precuneus was correlated with thickness of the rest of the DMN, i.e. in the inferior parietal lobe, medial prefrontal cortex, inferior frontal gyrus, and superior temporal gyrus; thickness of the medial prefrontal cortex was correlated with thickness of the rest of the DMN, i.e. in the medial prefrontal cortex, postcentral gyrus, inferior frontal gyrus and inferior temporal gyrus (Figure 3). Figure 3. Open in new tabDownload slide Structural covariance maps. The top left panel depicts the R-map of structural covariance from the precuneus in idiopathic hypersomnia. The top right panel depicts the R-map of structural covariance from the precuneus in good sleepers. The bottom left panel depicts the R-map of structural covariance from the medial prefrontal cortex (MPFC) in idiopathic hypersomnia. The bottom right panel depicts the R-map of structural covariance from the MPFC in good sleepers. Abbreviations: L: left, A: anterior, P: posterior. Figure 3. Open in new tabDownload slide Structural covariance maps. The top left panel depicts the R-map of structural covariance from the precuneus in idiopathic hypersomnia. The top right panel depicts the R-map of structural covariance from the precuneus in good sleepers. The bottom left panel depicts the R-map of structural covariance from the medial prefrontal cortex (MPFC) in idiopathic hypersomnia. The bottom right panel depicts the R-map of structural covariance from the MPFC in good sleepers. Abbreviations: L: left, A: anterior, P: posterior. In addition, there was a structural covariance by group interaction: structural covariance between the left medial prefrontal cortex (MNI coordinates x = −2, y = 44, z = 16) and left precuneus was greater in IH participants compared to good sleepers (Figure 4). There was no significant group difference in structural covariance from the precuneus, only structural covariance within the precuneus showed a subthreshold difference between groups (T > 2 uncorrected). No regions showed higher structural covariance in good sleepers compared to IH, nor any vertex-wise regression with subjective or objective daytime sleepiness. Figure 4. Open in new tabDownload slide Group difference in structural covariance. Structural covariance interaction T-map from the medial prefrontal cortex (right panel). A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Figure 4. Open in new tabDownload slide Group difference in structural covariance. Structural covariance interaction T-map from the medial prefrontal cortex (right panel). A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. No significant structural covariance difference between groups was observed without controlling for depression scores. Functional connectivity at rest During the resting-state fMRI acquisition, none of the participants fell asleep. There was no functional connectivity difference from the “posterior DMN” between groups. The “anterior DMN” was functionally connected with itself and with the rest of the DMN in both groups (Figure 5). Functional connectivity at rest was lower between the anterior DMN and the orbitofrontal cortex network in IH participants compared to good sleepers (Figure 6, Table 2). The network has been visually identified as the orbitofrontal network from the Cambridge parcellation [48] and includes the orbitofrontal cortex bilaterally and the fusiform gyrus. No regions showed stronger functional connectivity with the DMN at rest in IH compared to good sleepers. Figure 5. Open in new tabDownload slide Functional connectivity maps. The left panel depicts the R-map of functional connectivity from the anterior default-mode network (anterior DMN, i.e. medial prefrontal cortex) in idiopathic hypersomnia and in good sleepers. The right panel depicts the T-map of regions showing lower functional connectivity from the anterior DMN in idiopathic hypersomnia compared to good sleepers controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, R: right, A: anterior, P: posterior, X,Y,Z represent coordinates in MNI standard space. Figure 5. Open in new tabDownload slide Functional connectivity maps. The left panel depicts the R-map of functional connectivity from the anterior default-mode network (anterior DMN, i.e. medial prefrontal cortex) in idiopathic hypersomnia and in good sleepers. The right panel depicts the T-map of regions showing lower functional connectivity from the anterior DMN in idiopathic hypersomnia compared to good sleepers controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. Abbreviations: L: left, R: right, A: anterior, P: posterior, X,Y,Z represent coordinates in MNI standard space. Figure 6. Open in new tabDownload slide Voxel-wise regression between functional connectivity and self-reported daytime sleepiness. The left panel represents the T-map of the whole brain regression between functional connectivity and self-reported daytime sleepiness from the Epworth Sleepiness Scale (ESS), controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. The right panel shows the functional connectivity strength between the default-mode network (DMN) and the orbitofrontal network (ORB)—which was visually identified as the orbitofrontal network from the Cambridge parcellation—plotted against Epworth sleepiness score from the voxel with the maximum T-value. Abbreviations: L: left, R: right, A: anterior, P: posterior, X,Y,Z represent coordinates in MNI standard space. Figure 6. Open in new tabDownload slide Voxel-wise regression between functional connectivity and self-reported daytime sleepiness. The left panel represents the T-map of the whole brain regression between functional connectivity and self-reported daytime sleepiness from the Epworth Sleepiness Scale (ESS), controlling for age, sex, and depression score. A statistical threshold of p < 0.001 corresponding to a T > 3.5 was first applied, followed by an extent threshold of p < 0.05 corrected for multiple comparisons at the cluster level. The right panel shows the functional connectivity strength between the default-mode network (DMN) and the orbitofrontal network (ORB)—which was visually identified as the orbitofrontal network from the Cambridge parcellation—plotted against Epworth sleepiness score from the voxel with the maximum T-value. Abbreviations: L: left, R: right, A: anterior, P: posterior, X,Y,Z represent coordinates in MNI standard space. The voxel-wise regression showed no significant positive correlation between functional connectivity at rest from the anterior DMN and subjective daytime sleepiness in both groups. There was a significant negative correlation between functional connectivity at rest within the anterior DMN and self-reported daytime sleepiness in both groups (i.e. inferior temporal gyrus, medial prefrontal cortex, orbitofrontal cortex, Figure 6, Table 2). No significant regression was found between the functional connectivity at rest from the anterior DMN and objective daytime sleepiness in IH. Without controlling for depression score, functional connectivity at rest was lower between the anterior DMN and the fronto-parietal network in IH participants compared to good sleepers (Supplementary Figure 4, left panel; Supplementary Table 1). Functional connectivity was lower between the anterior DMN and the rest of the DMN (namely the medial prefrontal cortex, the inferior temporal gyrus, and postcentral gyrus) in IH participants compared to good sleepers (Supplementary Figure 4, right panel; Supplementary Table 1). The voxel-wise regression showed no significant positive correlation between functional connectivity at rest from the anterior DMN and self-reported daytime sleepiness in both groups. There was a significant negative correlation between the anterior DMN-fusiform functional connectivity and self-reported daytime sleepiness in both groups (Supplementary Figure 5, Supplementary Table 1). Discussion Given the importance of the DMN in alertness, sleep physiology, and based on previous findings in IH, we investigated the structural and functional modifications in the DMN with multimodal MRI to better characterize the pathophysiology of IH. We showed that functional connectivity at rest within the anterior (but not the posterior) DMN, i.e. the medial prefrontal cortex was lower in participants with IH. Functional connectivity was negatively correlated with subjective daytime sleepiness. In contrast, structural analyses showed that participants with IH had larger posterior DMN structure compared to good sleepers, namely increased gray matter volume and thicker precuneus. In addition, cortical thickness of the medial prefrontal cortex was more strongly correlated with cortical thickness of the posterior DMN (namely the precuneus) in participants with IH compared to good sleepers. Overall, the DMN hubs—the precuneus and medial prefrontal cortex—demonstrated significant changes in IH, and functional connectivity in the DMN correlates with self-reported but not objective sleepiness severity (as measured by the MSLT). Functional connectivity can be seen as a surrogate measure of regional cerebral blood flow for short-range connections [50, 51]. Therefore, the present findings of altered anterior DMN functional connectivity in IH are consistent with previous SPECT findings from our group that showed lower regional cerebral blood flow in the DMN (i.e. in the medial prefrontal cortex, posterior cingulate cortex, and putamen [5]. In narcolepsy with cataplexy, lower regional cerebral blood flow and metabolism in the medial prefrontal cortex has been described specifically within the inferior parietal lobe and anterior cingulate cortex [52–57], along with regions outside the DMN [52, 58–61], see [62] for a review. However, only one study investigated functional connectivity in those patients and showed lower functional connectivity outside the DMN, within the executive network (left medial frontal gyrus) and the salience network (right caudate) [63]. On the other hand, structural differences in the DMN have also been found in narcolepsy with cataplexy where gray matter volume in the medial prefrontal cortex, cingulate cortex, orbitofrontal cortex, and inferior parietal lobule was lower [64–69]. In addition, several regions outside the DMN also showed a smaller gray matter volume, particularly the hypothalamus, in line with a loss of hypocretinergic neurons [52, 69–74], see ref. [62] for a review. Our findings show a partial overlap between regions altered in IH and in narcolepsy with cataplexy but with opposite volume differences that could be explained by the absence of a loss of hypocretinergic neurons in IH and a structural compensation of the functional changes in other brain regions. In primary insomnia, reduced metabolism, measured with FDG-PET, was found in the medial prefrontal cortex at wake [75]. However, fMRI evidence demonstrated that this reduced activity was not solely restricted to the anterior DMN, but rather extended to the posterior DMN at wake [76]. Structural differences in the DMN have also been found, corresponding to lower gray matter volume in the orbitofrontal cortex and precuneus [41, 77, 78], although other studies have also found increased gray matter volume in the rostral anterior cingulate cortex [79]. Additionally, lower structural covariance between the frontal and parietal regions of the DMN has been found in primary insomnia, potentially reflecting a lower propensity to sleep [41]. The structural changes in primary insomnia contrast with our present findings in IH, in which we found greater structural covariance within medial regions (i.e. the medial prefrontal cortex and the precuneus), which might reflect a greater propensity to sleep. Therefore, the greater cortical thickness and structural covariance are a feature of IH that makes it distinct from other sleep disorders such as primary insomnia or narcolepsy with cataplexy. Functional connectivity changes have also been studied in the context of sleepiness associated with sleep deprivation. Sleep deprivation has been shown to reduce functional connectivity within the DMN (between the anterior and posterior parts) at rest [80]. These connectivity changes are dependent on the level of sleep pressure, i.e. functional connectivity decreases as a function of the duration prior wakefulness [80–82]. Lower anterior to posterior DMN connectivity has been shown related to increased daytime sleepiness in young healthy adults [83] and to worse sleep quality in adolescents [84]. In our current study, we observed a more focused difference in functional connectivity, which was restricted to the anterior part of the DMN in association with greater daytime sleepiness in IH, in contrast to the anterior to posterior connectivity change found after sleep deprivation. There are some limitations to this study. First, there was a group difference in BDI score, with IH showing higher depression score than controls. That is why we controlled for BDI score in all our analyses. For completeness, results without accounting for BDI are presented in the supplementary data and show gray matter volume and functional connectivity differences that were similar to the corrected results. However, there was no difference in cortical thickness and structural covariance between groups when BDI was not accounted for. Second, the present study has a limited sample size, which warrants replication in a larger sample. Given the limited number of participants with IH (N = 12), our study might have been underpowered to evidence correlations between DMN connectivity and objective daytime sleepiness. Finally, the lack of association between MSLT and connectivity could be explained by the fact that the MSLT was conducted as part of the clinical diagnostic procedure of participants and was not repeated immediately prior to the imaging session. In conclusion, the DMN—a brain network key to alertness and sleep—is affected in IH. Differences in the DMN in IH compared to controls are strikingly distinct from those reported in other sleep disorders such as narcolepsy with cataplexy and chronic insomnia. The greater gray matter volume and cortical thickness in the posterior DMN in IH may be a compensatory mechanism to the lower regional cerebral blood flow and functional connectivity in the anterior DMN. These opposite changes between functional and structural modalities have also been reported in conditions such as in multiple sclerosis, where structural white matter alterations were accompanied by higher functional connectivity and less cognitive efficiency [85]. Future studies are needed to further characterize the structural and functional changes that are specific to IH. This could be achieved by comparing a larger sample of individuals with IH to samples of narcolepsy participants (with and without cataplexy), as well as participants with sleepiness due to sleep deprivation. Ultimately these studies will be important for identification of neural biomarkers that are specific to each central disorder of hypersomnolence. Funding This work was supported by the Sleep Research Society Foundation. T.T.D-V. was also supported by the Canadian Institutes of Health Research (MOP 142191, PJT 153115 and PJT 156125), the Natural Sciences and Engineering Research Council of Canada (RGPIN 436006-2013), the Fonds de Recherche du Québec – Santé, the Canada Foundation for Innovation and Concordia University. H.K. was supported by the National Institutes of Health (P41EB015922) and BrightFocus Foundation (A2019052S). Conflict of interest statement. J.M. has received grants or support from Merck and GSK, was on the advisory board of Jazz Pharmaceuticals, Valeant Pharmaceuticals, and UCB Canada, and was a consultant for Valeant Pharmaceuticals. The other authors have no conflicts of interest to disclose. Acknowledgments We thank Dr Nathan Cross for critically reviewing the manuscript. References 1. Ohayon MM . From wakefulness to excessive sleepiness: what we know and still need to know . Sleep Med Rev. 2008 ; 12 ( 2 ): 129 – 141 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Dauvilliers Y , et al. Psychological health in central hypersomnias: the French Harmony study . J Neurol Neurosurg Psychiatry. 2009 ; 80 ( 6 ): 636 – 641 . Google Scholar Crossref Search ADS PubMed WorldCat 3. Trotti LM , et al. Brain MRI findings in patients with idiopathic hypersomnia . Clin Neurol Neurosurg. 2017 ; 157 : 19 – 21 . Google Scholar Crossref Search ADS PubMed WorldCat 4. Dauvilliers Y , et al. [18F]Fludeoxyglucose-positron emission tomography evidence for cerebral hypermetabolism in the awake state in narcolepsy and idiopathic hypersomnia . Front Neurol. 2017 ; 8 : 350 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Boucetta S , et al. Altered regional cerebral blood flow in idiopathic hypersomnia . Sleep 2017 ; 40 ( 10 ). doi: https://doi.org/10.1093/sleep/zsx140 . WorldCat 6. Raichle ME , et al. A default mode of brain function . Proc Natl Acad Sci USA . 2001 ; 98 : 676 – 682 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Laird AR , et al. Investigating the functional heterogeneity of the default mode network using coordinate-based meta-analytic modeling . J Neurosci. 2009 ; 29 ( 46 ): 14496 – 14505 . Google Scholar Crossref Search ADS PubMed WorldCat 8. Mohan A , et al. The significance of the Default Mode Network (DMN) in neurological and neuropsychiatric disorders: a review . Yale J Biol Med. 2016 ; 89 ( 1 ): 49 – 57 . Google Scholar PubMed WorldCat 9. Sämann PG , et al. Development of the brain’s default mode network from wakefulness to slow wave sleep . Cereb Cortex. 2011 ; 21 ( 9 ): 2082 – 2093 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Horovitz SG , et al. Low frequency BOLD fluctuations during resting wakefulness and light sleep: a simultaneous EEG-fMRI study . Hum Brain Mapp. 2008 ; 29 ( 6 ): 671 – 682 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Horovitz SG , et al. Decoupling of the brain’s default mode network during deep sleep . Proc Natl Acad Sci U S A. 2009 ; 106 ( 27 ): 11376 – 11381 . Google Scholar Crossref Search ADS PubMed WorldCat 12. American Academy of Sleep Medicine . International Classification of Sleep Disorders revised: Diagnostic and coding manual . American Academy of Sleep Medicine ; 2012 . WorldCat COPAC 13. American Psychiatric Association . Diagnostic and Statistical Manual of Mental Disorders (DSM-5®) . Washington, DC : American Psychiatric Pub ; 2013 . WorldCat COPAC 14. Johns MW . A new method for measuring daytime sleepiness: the Epworth sleepiness scale . Sleep. 1991 ; 14 ( 6 ): 540 – 545 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Buysse DJ , et al. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research . Psychiatry Res. 1989 ; 28 ( 2 ): 193 – 213 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Horne JA , et al. A self-assessment questionnaire to determine morningness-eveningness in human circadian rhythms . Int J Chronobiol. 1976 ; 4 ( 2 ): 97 – 110 . Google Scholar PubMed WorldCat 17. Beck AT , et al. An inventory for measuring clinical anxiety: psychometric properties . J Consult Clin Psychol. 1988 ; 56 ( 6 ): 893 – 897 . Google Scholar Crossref Search ADS PubMed WorldCat 18. BECK AT , et al. An inventory for measuring depression . Arch Gen Psychiatry. 1961 ; 4 : 561 – 571 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Baldwin DS , et al. Symptoms of fatigue and sleepiness in major depressive disorder . J Clin Psychiatry. 2006 ; 67 ( Suppl 6 ): 9 – 15 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Vandeputte M , et al. Sleep disorders and depressive feelings: a global survey with the Beck depression scale . Sleep Med. 2003 ; 4 ( 4 ): 343 – 345 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Greicius MD , et al. Resting-state functional connectivity in major depression: abnormally increased contributions from subgenual cingulate cortex and thalamus . Biol Psychiatry. 2007 ; 62 ( 5 ): 429 – 437 . Google Scholar Crossref Search ADS PubMed WorldCat 22. Whitfield-Gabrieli S , et al. Default mode network activity and connectivity in psychopathology . Annu Rev Clin Psychol. 2012 ; 8 : 49 – 76 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Ashburner J , et al. Voxel-based morphometry–the methods . Neuroimage. 2000 ; 11 ( 6 Pt 1 ): 805 – 821 . Google Scholar Crossref Search ADS PubMed WorldCat 24. CAT - A computational anatomy toolbox for the analysis of structural MRI Data. In: Human Brain Mapping Conference 2016 : 1 – 1 . http://www.neuro.uni-jena.de/hbm2016/GaserHBM2016.pdf WorldCat COPAC 25. Collins DL , et al. Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space . J Comput Assist Tomogr. 1994 ; 18 ( 2 ): 192 – 205 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Sled JG , et al. A nonparametric method for automatic correction of intensity nonuniformity in MRI data . IEEE Trans Med Imaging. 1998 ; 17 ( 1 ): 87 – 97 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Zijdenbos A , Forghani R , Evans A . Automatic quantification of MS lesions in 3D MRI brain data sets: Validation of INSECT. In: Wells WM , Colchester A , Delp S , eds. Berlin, Heidelberg : Springer Berlin Heidelberg ; 1998 : 439 – 448 . COPAC 28. Collins DL , et al. ANIMAL+INSECT: Improved Cortical Structure Segmentation. In: Kuba A , Šáamal M , Todd-Pokropek A , eds. Information Processing in Medical Imaging . IPMI 1999. Lecture Notes in Computer Science, vol. 1613. Berlin, Heidelberg : Springer ; 1999 . Google Preview WorldCat COPAC 29. Smith SM . Fast robust automated brain extraction . Hum Brain Mapp. 2002 ; 17 ( 3 ): 143 – 155 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Tohka J , et al. Fast and robust parameter estimation for statistical partial volume models in brain MRI . Neuroimage. 2004 ; 23 ( 1 ): 84 – 97 . Google Scholar Crossref Search ADS PubMed WorldCat 31. Ad-Dab’bagh Y , et al. The CIVET Image-Processing Environment: A Fully Automated Comprehensive Pipeline for Anatomical Neuroimaging Research . In 2006 . http://www.bic.mni.mcgill.ca/users/yaddab/Yasser-HBM2006-Poster.pdf Google Preview WorldCat COPAC 32. Sherif T , et al. CBRAIN: a web-based, distributed computing platform for collaborative neuroimaging research . Front Neuroinform 2014 ; 8 : 54 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Tristan G , et al. Interoperability between the CBRAIN and VIP web platforms for neuroimage analysis . Front Neuroinform 2014 ; 8 . WorldCat 34. Grabner G , et al. Symmetric atlasing and model-based segmentation: an application to the hippocampus in older adults . Med Image Comput Comput Assist Interv. 2006 ; 9 ( Pt 2 ): 58 – 66 . Google Scholar PubMed WorldCat 35. Boucher M , et al. Depth potential function for folding pattern representation, registration and analysis . Med Image Anal. 2009 ; 13 ( 2 ): 203 – 214 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Robbins S , et al. Tuning and comparing spatial normalization methods . Med Image Anal. 2004 ; 8 ( 3 ): 311 – 323 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Lyttelton O , et al. An unbiased iterative group registration template for cortical surface analysis . Neuroimage. 2007 ; 34 ( 4 ): 1535 – 1544 . Google Scholar Crossref Search ADS PubMed WorldCat 38. Gong G , et al. Convergence and divergence of thickness correlations with diffusion connections across the human cerebral cortex . Neuroimage. 2012 ; 59 ( 2 ): 1239 – 1248 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Lerch JP , et al. Mapping anatomical correlations across cerebral cortex (MACACC) using cortical thickness from MRI . Neuroimage. 2006 ; 31 ( 3 ): 993 – 1003 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Suh S , et al. Cortical thinning and altered cortico-cortical structural covariance of the default mode network in patients with Persistent Insomnia Symptoms . Sleep. 2016 ; 39 ( 1 ): 161 – 171 . Google Scholar Crossref Search ADS PubMed WorldCat 41. Bellec P , et al. A neuroimaging analysis kit for Matlab and Octave. In Proceedings of the 17th International Conference on Functional Mapping of the Human Brain ; 2011 . Google Preview WorldCat COPAC 42. Collins DL , et al. Animal: validation and applications of nonlinear registration-based segmentation . Int J Pattern Recognit Artif Intell . 1997 ; 11 : 1271 – 1294 . Google Scholar Crossref Search ADS WorldCat 43. Fonov V , et al. ; Brain Development Cooperative Group . Unbiased average age-appropriate atlases for pediatric studies . Neuroimage. 2011 ; 54 ( 1 ): 313 – 327 . Google Scholar Crossref Search ADS PubMed WorldCat 44. Power JD , et al. Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion . Neuroimage. 2012 ; 59 ( 3 ): 2142 – 2154 . Google Scholar Crossref Search ADS PubMed WorldCat 45. Lund TE , et al. Non-white noise in fMRI: does modelling have an impact? Neuroimage. 2006 ; 29 ( 1 ): 54 – 66 . Google Scholar Crossref Search ADS PubMed WorldCat 46. Giove F , et al. Images-based suppression of unwanted global signals in resting-state functional connectivity studies . Magn Reson Imaging. 2009 ; 27 ( 8 ): 1058 – 1064 . Google Scholar Crossref Search ADS PubMed WorldCat 47. Worsley KJ , et al. Detecting changes in nonisotropic images . Hum Brain Mapp. 1999 ; 8 ( 2-3 ): 98 – 101 . Google Scholar Crossref Search ADS PubMed WorldCat 48. Liu H , et al. Evidence from intrinsic activity that asymmetry of the human brain is controlled by multiple factors . Proc Natl Acad Sci USA ., 2009 ; 106 ( 48 ): 20499 – 20503 . doi: https://doi.org/10.1073/pnas.0908073106 Google Scholar Crossref Search ADS PubMed WorldCat 49. Flandin G , et al. Analysis of family-wise error rates in statistical parametric mapping using random field theory . Hum Brain Mapp. 2019 ; 40 ( 7 ): 2052 – 2054 . Google Scholar Crossref Search ADS PubMed WorldCat 50. Liang X , et al. Coupling of functional connectivity and regional cerebral blood flow reveals a physiological basis for network hubs of the human brain . Proc Natl Acad Sci USA . 2013 ; 110 : 1929 – 1934 . Google Scholar Crossref Search ADS PubMed WorldCat 51. Qiu M , et al. Multi-modal analysis of functional connectivity and cerebral blood flow reveals shared and unique effects of propofol in large-scale brain networks . Neuroimage. 2017 ; 148 : 130 – 140 . Google Scholar Crossref Search ADS PubMed WorldCat 52. Joo EY , et al. Cerebral perfusion abnormality in narcolepsy with cataplexy . Neuroimage. 2005 ; 28 ( 2 ): 410 – 416 . Google Scholar Crossref Search ADS PubMed WorldCat 53. Hong SB , et al. Cerebral perfusion changes during cataplexy in narcolepsy patients . Neurology. 2006 ; 66 ( 11 ): 1747 – 1749 . Google Scholar Crossref Search ADS PubMed WorldCat 54. Chabas D , et al. Functional imaging of cataplexy during status cataplecticus . Sleep. 2007 ; 30 ( 2 ): 153 – 156 . Google Scholar Crossref Search ADS PubMed WorldCat 55. Joo EY , et al. Glucose hypometabolism of hypothalamus and thalamus in narcolepsy . Ann Neurol. 2004 ; 56 ( 3 ): 437 – 440 . Google Scholar Crossref Search ADS PubMed WorldCat 56. Ponz A , et al. Abnormal activity in reward brain circuits in human narcolepsy with cataplexy . Ann Neurol. 2010 ; 67 ( 2 ): 190 – 200 . Google Scholar Crossref Search ADS PubMed WorldCat 57. Ponz A , et al. Reduced amygdala activity during aversive conditioning in human narcolepsy . Ann Neurol. 2010 ; 67 ( 3 ): 394 – 398 . Google Scholar PubMed WorldCat 58. Asenbaum S , et al. Technetium-99m-HMPAO SPECT imaging of cerebral blood flow during REM sleep in narcoleptics . J Nucl Med. 1995 ; 36 ( 7 ): 1150 – 1155 . Google Scholar PubMed WorldCat 59. Dauvilliers Y , et al. A brain PET study in patients with narcolepsy-cataplexy . J Neurol Neurosurg Psychiatry. 2010 ; 81 ( 3 ): 344 – 348 . Google Scholar Crossref Search ADS PubMed WorldCat 60. Schwartz S , et al. Abnormal activity in hypothalamus and amygdala during humour processing in human narcolepsy with cataplexy . Brain. 2008 ; 131 ( Pt 2 ): 514 – 522 . Google Scholar Crossref Search ADS PubMed WorldCat 61. Reiss AL , et al. Anomalous hypothalamic responses to humor in cataplexy . PLoS One. 2008 ; 3 ( 5 ): e2225 . Google Scholar Crossref Search ADS PubMed WorldCat 62. O’Byrne JN , Salimi A , Dang-Vu TT . Neuroimaging of Narcolepsy . In: Goswami M , Thorpy M , Pandi-Perumal S , eds. Narcolepsy. Switzerland : Springer , Cham; 2016 . doi: https://doi.org/10.1007/978-3-319-23739-8_13 COPAC 63. Xiao F , et al. Independent component analysis and graph theoretical analysis in patients with narcolepsy . Neurosci Bull . 2018 ; 70 : 335 – 13 . WorldCat 64. Kaufmann C , et al. Reduced cortical gray matter in narcolepsy: preliminary findings with voxel-based morphometry . Neurology. 2002 ; 58 ( 12 ): 1852 – 1855 . Google Scholar Crossref Search ADS PubMed WorldCat 65. Brenneis C , et al. Voxel-based morphometry in narcolepsy . Sleep Med. 2005 ; 6 ( 6 ): 531 – 536 . Google Scholar Crossref Search ADS PubMed WorldCat 66. Joo EY , et al. Analysis of cortical thickness in narcolepsy patients with cataplexy . Sleep. 2011 ; 34 ( 10 ): 1357 – 1364 . Google Scholar Crossref Search ADS PubMed WorldCat 67. Joo EY , et al. Gray matter concentration abnormality in brains of narcolepsy patients . Korean J Radiol. 2009 ; 10 ( 6 ): 552 – 558 . Google Scholar Crossref Search ADS PubMed WorldCat 68. Schaer M , et al. Cortical morphometry in narcolepsy with cataplexy . J Sleep Res. 2012 ; 21 ( 5 ): 487 – 494 . Google Scholar Crossref Search ADS PubMed WorldCat 69. Scherfler C , et al. White and gray matter abnormalities in narcolepsy with cataplexy . Sleep. 2012 ; 35 ( 3 ): 345 – 351 . Google Scholar Crossref Search ADS PubMed WorldCat 70. Draganski B , et al. Hypothalamic gray matter changes in narcoleptic patients . Nat Med. 2002 ; 8 ( 11 ): 1186 – 1188 . Google Scholar Crossref Search ADS PubMed WorldCat 71. Buskova J , et al. Reduced hypothalamic gray matter in narcolepsy with cataplexy . Neuro Endocrinol Lett. 2006 ; 27 ( 6 ): 769 – 772 . Google Scholar PubMed WorldCat 72. Kim SJ , et al. Gray matter deficits in young adults with narcolepsy . Acta Neurol Scand. 2009 ; 119 ( 1 ): 61 – 67 . Google Scholar Crossref Search ADS PubMed WorldCat 73. Brabec J , et al. Volume of the amygdala is reduced in patients with narcolepsy - a structural MRI study . Neuro Endocrinol Lett. 2011 ; 32 ( 5 ): 652 – 656 . Google Scholar PubMed WorldCat 74. Joo EY , et al. Hippocampal volume and memory in narcoleptics with cataplexy . Sleep Med. 2012 ; 13 ( 4 ): 396 – 401 . Google Scholar Crossref Search ADS PubMed WorldCat 75. Nofzinger EA , et al. Functional neuroimaging evidence for hyperarousal in insomnia . Am J Psychiatry. 2004 ; 161 ( 11 ): 2126 – 2128 . Google Scholar Crossref Search ADS PubMed WorldCat 76. Tagliazucchi E , et al. The large-scale functional connectivity correlates of consciousness and arousal during the healthy and pathological human sleep cycle . Neuroimage. 2017 ; 160 : 55 – 72 . Google Scholar Crossref Search ADS PubMed WorldCat 77. Altena E , et al. Reduced orbitofrontal and parietal gray matter in chronic insomnia: a voxel-based morphometric study . Biol Psychiatry. 2010 ; 67 ( 2 ): 182 – 185 . Google Scholar Crossref Search ADS PubMed WorldCat 78. Joo EY , et al. Brain gray matter deficits in patients with chronic primary insomnia . Sleep. 2013 ; 36 ( 7 ): 999 – 1007 . Google Scholar Crossref Search ADS PubMed WorldCat 79. Winkelman JW , et al. Increased rostral anterior cingulate cortex volume in chronic primary insomnia . Sleep. 2013 ; 36 ( 7 ): 991 – 998 . Google Scholar Crossref Search ADS PubMed WorldCat 80. De Havas JA , et al. Sleep deprivation reduces default mode network connectivity and anti-correlation during rest and task performance . Neuroimage. 2012 ; 59 ( 2 ): 1745 – 1751 . Google Scholar Crossref Search ADS PubMed WorldCat 81. Sämann PG , et al. Increased sleep pressure reduces resting state functional connectivity . MAGMA. 2010 ; 23 ( 5-6 ): 375 – 389 . Google Scholar Crossref Search ADS PubMed WorldCat 82. Yeo BT , et al. Functional connectivity during rested wakefulness predicts vulnerability to sleep deprivation . Neuroimage. 2015 ; 111 : 147 – 158 . Google Scholar Crossref Search ADS PubMed WorldCat 83. Ward AM , et al. Daytime sleepiness is associated with decreased default mode network connectivity in both young and cognitively intact elderly subjects . Sleep. 2013 ; 36 ( 11 ): 1609 – 1615 . Google Scholar Crossref Search ADS PubMed WorldCat 84. Tashjian SM , et al. Sleep quality and adolescent default mode network connectivity . Soc Cogn Affect Neurosci. 2018 ; 13 ( 3 ): 290 – 299 . Google Scholar Crossref Search ADS PubMed WorldCat 85. Hawellek DJ , et al. Increased functional connectivity indicates the severity of cognitive impairment in multiple sclerosis . Proc Natl Acad Sci U S A. 2011 ; 108 ( 47 ): 19066 – 19071 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Garbazza,, Corrado;Hackethal,, Sandra

doi: 10.1093/sleep/zsz155pmid: 31318026

Our attention as readers of Sleep has been caught by a recently published article entitled “Neurocognitive and behavioral significance of periodic lim movements during sleep in adolescents with attention-deficit/hyperactivity disorder” [1] (ADHD), which examins a possible association between periodic leg movements during sleep (PLMS) and adverse neurobehavioral outcomes in adolescents with ADHD. Study participants were recruited among the Penn State Child Cohort, a random general population sample. Four hundred twenty-one adolescents (17.0 ± 2.3 years, 53.9% male) underwent night sleep assessment by full polysomnography, as well as a physical examination and neurocognitive evaluation. PLMS were defined as abnormally high based on a PLM index (PLMI) of ≥5 events per hour of sleep, whereas the diagnosis of ADHD was determined through parent- or self-report. Past or current pharmacologic treatment (i.e. stimulant, psychoactive, sleep or asthma/allergy medication) was also assessed, but not further specified. Among adolescents with ADHD, n = 71 reported current and n = 27 past treatment for the disorder. The authors then subdivided the cohort into four groups based on the presence of ADHD (+/−) and PLMI ≥ 5 (+/−): ADHD+/PLMI+ (n = 35), ADHD+/PLMI− (n = 63), ADHD−/PLMI+ (n = 68), and ADHD−/PLMI− (controls, n = 255). Statistical analysis was performed by using the chi-squared test and analysis of variance for categorical and continuous variables, respectively. To test whether the presence of PLMI ≥ 5 had an influence on the neurobehavioral phenotype of ADHD, full-factorial multivariable-adjusted general linear models were applied after controlling for potential confounders “(i.e. sex, race, age, body mass index (BMI) percentile, morningness–eveningness questionnaire (MEQ), insomnia, excessive daytime sleepiness (EDS), sleep onset latency, number of awakenings, and Apnea–Hypopnea Index (AHI))”(p. 3). No correction for multiple testing was performed. The authors report adolescents with ADHD (n = 98, age 16.8 ± 2.1 years) to have a significantly higher PLMI (5.4 ± 7.2) and percentage of PLMI ≥ 5 (35.7%) compared to controls (3.4 ± 5.6, p = 0.006 and 21.1%, p = 0.003). ADHD was also significantly associated with a worse performance in all measures of neurocognitive and behavioral functioning. Patients with both ADHD and PLMI ≥ 5 (n = 35) showed a significant interaction between the two conditions, with more pronounced deficits in control interference (Stroop test) and elevated internalizing behavior (C/ABCL). Furthermore, statistical analysis revealed a positive linear relationship between ADHD severity and externalizing behavior in both the ADHD alone and ADHD+/PLMI+ group. For other neurocognitive functions the presence of PLMI ≥ 5 in ADHD did not modify the outcomes. On the basis of the collected data, the authors hypothesize adolescents with ADHD and PLMI ≥ 5/h to represent a distinct phenotype with “increased anxiety, physiological difficulty falling asleep, and control interference deficits.” (p. 7). They also interpret the sleep pattern observed in this group, (with significant sleep fragmentation, high number of awakenings in stage 1 sleep, increased sleep onset latency and slightly decreased stage 2 sleep) as a defining feature of the proposed phenotype, and even suggest different pathophysiological mechanisms of sleep disturbances in ADHD with and without PLMS. PLM are motor phenomena arising during rest wakefulness (PLMW) and sleep (PLMS), characterized by periodic, rather stereotyped movements of the upper and mainly of the lower limbs. They may be associated with cortical arousals and autonomic activations, and/or with an awakening, thus resulting in sleep disruption [2]. PLMS frequently occur in restless leg syndrome (RLS), representing a supportive diagnostic criterion, but also in other sleep disorders, such as narcolepsy, rapid eye movement (REM) sleep behavior disorder, obstructive sleep apnea syndrome, or periodic limb movement disorder (PLMD) [3–9], as well as in the general healthy population, with a prevalence of 25.3%–36.4%, increasing with age, according to the most recent large-scale studies [10–12]. Although the complex clinical significance and implications of PLMS have not been completely understood, in the last decade several studies on the time structure of PLMS have been published, as well as on their pathophysiological correlates, including the association with sympathetic activation [13–15] and cardiovascular risk [13–16]. However, the results of all research investigations conducted on PLMS and their clinical impact strictly depend on the criteria used to define PLMS, which still represent an ongoing and dynamic process. Therefore, considerable progress has been made in the development of consensus rules and advanced analysis methods for the measurement of PLMS, in order to provide researchers with adequate instruments to correctly evaluate the intrinsic features of this phenomenon. In this Journal Club, by critically reviewing the work by Frye et al. [1], we intend to highlight some methodological issues that should be accurately taken into account when measuring PLMS, because they may significantly influence the results of the PLMS analysis and its interpretation. First, when deciding to detect PLMS, it is necessary to choose one of the two currently existing sets of scoring criteria. Frye et al. [1] adopted the American Academy of Sleep Medicine (AASM) rules, published in 2007 and regularly updated [17, 18], that partially resume the first international criteria established in 1993 [19], which, in turn, were a revision of the first scoring criteria proposed by Coleman in 1982 [20]. However, another set of rules has been introduced by a task force of the International Restless Legs Syndrome Study Group (IRLSSG) and endorsed by the World Association of Sleep Medicine (WASM) in 2006 [21]. The IRLSSG/WASM criteria have been then thoroughly revised in 2016 [22], in the light of the considerable new findings arising from research in recent years. These rules provide a complete and detailed set of necessary features to unambiguously score PLMS. As an example, by following the AASM rules [17, 18] or the 2006 IRLSSG/WASM criteria [21], it is possible to classify as PLMS also nonperiodic activity, which is part of the whole motor phenomenon, but neurophysiologically and neurobiologically different from the genuine PLMS [23, 24]. Conversely, the 2016 IRLSSG/WASM criteria [22] for scoring leg motor activity during sleep proved to be particularly powerful in separating and identifying the different types of movements [25]. A second aspect to consider is related to the cutoff value of the PLMS index, i.e. the number of PLMS per hour of sleep. Although there is no clear threshold beyond which this index should be considered pathological [26], because the clinical context and the patient’s sleep-related complaints are probably more important than an absolute cutoff value [3], the evidence that the PLMS index is generally increasing with age in normal controls has induced most authors to consider normative values of greater than 5/h in children and greater than 15/h in adults as abnormal and likely to be associated with symptoms of sleep disturbances [3]. In the work by Frye et al. [1], the pediatric cutoff value of PLMI ≥ 5/h was applied to a population of adolescents aged 12–23 years old (mean age = 17.0 ± 2.3). Although, according to mean age, even the adoption of a PLMS index greater than 15/h would have been conceivable, this choice may be justified by the fact that, as shown by previous studies, leg movements (LMs) and PLMS indices decrease with age up to 18 years [26, 27]. Moreover, the authors found an identical distribution of PLMS in the 12–17 and 18–23 years age groups (PLMS index of 4.9/h for the 75% percentile), providing a rationale for using the pediatric cutoff value in both groups [1]. Third, since the debate about the cutoff value for an abnormal PLMS index remains unresolved and an increased index can be found in a variety of clinical conditions besides RLS, it should be considered that the PLMS index alone may not be the most reliable and specific tool to analyze LM activity during sleep. Therefore, it may be useful to estimate also other advanced variables, in addition to the PLMS index. In recent years, new data-driven parameters were established to better describe the intrinsic features of LMs and characterize the genuine, periodic component of the LM phenomenon, rather than other motor activities included in the “PLMS” analysis [25,28,29]: the distribution of intermovement intervals (IMI), which in RLS subjects show a typical bimodal pattern, with a first, smaller peak at with one peak at 2–4 seconds and second, main peak at around 20–22 seconds, mostly representing the occurrence of PLMS; the periodicity index (PI), calculated as the ratio between the number of intermovement intervals contained in regular uninterrupted sequences of at least 4 LMs, separated by 10- to 90-second intervals and the total number of intermovement intervals recorded. The PI is independent from both the absolute number of LMs recorded and the PLMS index, thus providing a reliable information about the time structure of LM activity [25]; the stage segregation of PLMS and their distribution across the night, which in adult RLS subjects follows a circadian trend, with a progressive reduction across the night reaching a minimum around the nadir in body temperature and occurring more in NREM vs. REM sleep [28]. The evaluation of these three types of measurements in different conditions [29–33] and in response to treatment [34–40] showed that the pattern of LMs during sleep is very significantly influenced by age [41], with periodicity being a less common phenomenon in pediatric individuals [26, 27, 42]. Interestingly, Bruni et al. [43] and Ferri et al. [44], demonstrated that PLMS in children with ADHD, differently from those observed in RLS patients, have a low PI, a scarce circadian decrement across the night, and are not responsive to L-DOPA treatment, leading the authors to hypothesize different generating mechanisms of LMs in ADHD and RLS [44]. The hypothesis that an elevated number of PLMS during the night in ADHD may mirror the typical daytime hyperactivity of these patients and explain their overall reduced sleep quality, led researchers to focus on the pattern of nocturnal motor activity in ADHD [44, 45]. However, considering the PLMS index alone, several studies did not reach consistent conclusions, with some of them suggesting that sleep in ADHD is characterized by a significantly higher number of PLMS compared to healthy subjects [46–50], whereas others did not confirm these findings [51–53]. Recently, in a large study analyzing the frequency of PLMS in a young adult population of ADHD patients with and without RLS compared to controls, Lopez et al.[54] found no statistically significant higher prevalence of PLMS index ≥5 between groups, but a higher periodicity (PI) of LMs and lower ferritin levels in the subgroup of ADHS patients with RLS. Frye et al. [1] report adolescents with ADHD to have a significantly higher PLMS index (5.4 ± 7.3) compared to controls (3.4 ± 5.6, p = 0.006). However, for the abovementioned reasons, it should here be noted that the clinical value of this difference between groups is uncertain and likely to be influenced by some confounding factors: assessment of disease: ADHD assessment was performed only by asking the participants “Have you ever been treated for a psychiatric/behavioral disorder?” or their parents “Has your child ever been treated for a psychiatric/behavioral disorder?,” followed, in case of positive answer, by the option to specify whether that disorder was ADHD and whether it was being currently treated or it had been treated in the past. Not applying formal Diagnostic and Statistical Manual of Mental Disorders (Fifth Edition) diagnostic criteria for ADHD [55], might bear the risk to overestimate the prevalence of disease, as stated by the authors, and therefore include other clinical conditions mimicking ADHD symptoms, such as RLS; concomitant medications: according to Table 1, when separating the subjects based on the presence of PLMS (independent of ADHD status), the PLMS positive group was also taking significantly more stimulant and psychoactive medications. However, when evaluating the influence of PLMS in ADHD on neurocognitive outcomes, the authors reportedly controlled for a whole range of variables but not for medication, although a supplementary sensitivity analysis suggested similar results in the unmedicated group; comorbid conditions: given the frequent, age-independent association between ADHD and sleep-related movement disorders, such as RLS and PLMD [56–59], which may even be based on a possible common central dopaminergic dysfunction [60, 61] or brain iron deficiency [62–64], a careful assessment of RLS and iron status is mandatory when studying PLMS in ADHD. This was not the case of the present study, where a diagnosis of RLS based on the five IRLSSG diagnostic criteria [3] was not performed, but parents were instead asked “does your child describe restlessness of the legs when in bed?,” “does your child have brief kicks of one leg or both legs during sleep?,” and “does your child have repeated kicks or jerks of the legs at regular intervals?.” As a result, in only 26%–27% of patients with ADHD and PLMS, the parents endorsed observing LMs during their child’s sleep or symptoms of RLS. Moreover, despite the evidence for a strong association between PLM and iron metabolism alterations in children, iron status was not assessed in the study population. Table 1. PLMS analysis checklist Chekclist for the analysis of PLMS • Choose the set of scoring rules The AASM and the IRLSSG/WASM are the internationally recognized criteria. Specify which version of these scoring rules has been used and if the PLMS scoring was performed automatically (followed/not followed by human supervision) or manually. • Consider the cutoff value of the PLMS index according to age PLMS describing parameters, including the PLMS index, need correction for age, especially in young children, adolescents and elderly people. This might be reflected in age-sensitive cutoff values. • Measure other parameters than the PLMS index The PLMS Index alone has a low specificity, e.g. for the diagnosis of RLS, and a high night-to-night variability. The informative value of the PLMS analysis can be increased by computing additional parameters, such as the periodicity index (PI), intermovement intervals (IMI) distribution, number and length of PLMS sequences, and time distribution of PLMS across the night. • Control for medications Several drugs, including most antidepressants, some neuroleptics, lithium and anti-dopaminergic medications in general, may induce or exacerbate PLMS. • Control for comorbid conditions Several sleep disorders, such as restless legs syndrome (RLS), REM sleep behaviour disorder (RBD), narcolepsy, and obstructive sleep apnea syndrome (OSAS), as well as neurodegenerative diseases, e.g. Parkinson’s disease (PD) and multiple system atrophy (MSA), are associated with PLMS. • Assess iron status Low ferritin levels may be associated with an increased number of PLMS. Therefore, iron status should be systematically assessed and iron supplementation may be considered, especially in RLS patients. Chekclist for the analysis of PLMS • Choose the set of scoring rules The AASM and the IRLSSG/WASM are the internationally recognized criteria. Specify which version of these scoring rules has been used and if the PLMS scoring was performed automatically (followed/not followed by human supervision) or manually. • Consider the cutoff value of the PLMS index according to age PLMS describing parameters, including the PLMS index, need correction for age, especially in young children, adolescents and elderly people. This might be reflected in age-sensitive cutoff values. • Measure other parameters than the PLMS index The PLMS Index alone has a low specificity, e.g. for the diagnosis of RLS, and a high night-to-night variability. The informative value of the PLMS analysis can be increased by computing additional parameters, such as the periodicity index (PI), intermovement intervals (IMI) distribution, number and length of PLMS sequences, and time distribution of PLMS across the night. • Control for medications Several drugs, including most antidepressants, some neuroleptics, lithium and anti-dopaminergic medications in general, may induce or exacerbate PLMS. • Control for comorbid conditions Several sleep disorders, such as restless legs syndrome (RLS), REM sleep behaviour disorder (RBD), narcolepsy, and obstructive sleep apnea syndrome (OSAS), as well as neurodegenerative diseases, e.g. Parkinson’s disease (PD) and multiple system atrophy (MSA), are associated with PLMS. • Assess iron status Low ferritin levels may be associated with an increased number of PLMS. Therefore, iron status should be systematically assessed and iron supplementation may be considered, especially in RLS patients. Main variables to consider when performing an accurate analysis of periodic leg movements during sleep (PLMS) in sleep medicine and research. Open in new tab Table 1. PLMS analysis checklist Chekclist for the analysis of PLMS • Choose the set of scoring rules The AASM and the IRLSSG/WASM are the internationally recognized criteria. Specify which version of these scoring rules has been used and if the PLMS scoring was performed automatically (followed/not followed by human supervision) or manually. • Consider the cutoff value of the PLMS index according to age PLMS describing parameters, including the PLMS index, need correction for age, especially in young children, adolescents and elderly people. This might be reflected in age-sensitive cutoff values. • Measure other parameters than the PLMS index The PLMS Index alone has a low specificity, e.g. for the diagnosis of RLS, and a high night-to-night variability. The informative value of the PLMS analysis can be increased by computing additional parameters, such as the periodicity index (PI), intermovement intervals (IMI) distribution, number and length of PLMS sequences, and time distribution of PLMS across the night. • Control for medications Several drugs, including most antidepressants, some neuroleptics, lithium and anti-dopaminergic medications in general, may induce or exacerbate PLMS. • Control for comorbid conditions Several sleep disorders, such as restless legs syndrome (RLS), REM sleep behaviour disorder (RBD), narcolepsy, and obstructive sleep apnea syndrome (OSAS), as well as neurodegenerative diseases, e.g. Parkinson’s disease (PD) and multiple system atrophy (MSA), are associated with PLMS. • Assess iron status Low ferritin levels may be associated with an increased number of PLMS. Therefore, iron status should be systematically assessed and iron supplementation may be considered, especially in RLS patients. Chekclist for the analysis of PLMS • Choose the set of scoring rules The AASM and the IRLSSG/WASM are the internationally recognized criteria. Specify which version of these scoring rules has been used and if the PLMS scoring was performed automatically (followed/not followed by human supervision) or manually. • Consider the cutoff value of the PLMS index according to age PLMS describing parameters, including the PLMS index, need correction for age, especially in young children, adolescents and elderly people. This might be reflected in age-sensitive cutoff values. • Measure other parameters than the PLMS index The PLMS Index alone has a low specificity, e.g. for the diagnosis of RLS, and a high night-to-night variability. The informative value of the PLMS analysis can be increased by computing additional parameters, such as the periodicity index (PI), intermovement intervals (IMI) distribution, number and length of PLMS sequences, and time distribution of PLMS across the night. • Control for medications Several drugs, including most antidepressants, some neuroleptics, lithium and anti-dopaminergic medications in general, may induce or exacerbate PLMS. • Control for comorbid conditions Several sleep disorders, such as restless legs syndrome (RLS), REM sleep behaviour disorder (RBD), narcolepsy, and obstructive sleep apnea syndrome (OSAS), as well as neurodegenerative diseases, e.g. Parkinson’s disease (PD) and multiple system atrophy (MSA), are associated with PLMS. • Assess iron status Low ferritin levels may be associated with an increased number of PLMS. Therefore, iron status should be systematically assessed and iron supplementation may be considered, especially in RLS patients. Main variables to consider when performing an accurate analysis of periodic leg movements during sleep (PLMS) in sleep medicine and research. Open in new tab In summary, although these study limitations have been correctly stated by the authors in the discussion section of the paper, their impact on the outcomes of the PLMS analysis may have been underestimated. Thus, concluding that the findings “demonstrat(e) the neurobehavioral significance of PLMS in the context of ADHD and suggest potential treatment targets at the sleep and neurobehavioral levels” might represent an arguable interpretation of the results. In general, when analyzing PLMS in research studies, we suggest following a step-by-step approach, as summarized in Table 1, in order to gain more detailed information from the data, that may be critical to understand the complex pathophysiology of sleep-related motor phenomena and their clinical implications. Conflict of interest statement. None declared. References 1. Frye SS , et al. Neurocognitive and behavioral significance of periodic limb movements during sleep in adolescents with attention-deficit/hyperactivity disorder . Sleep . 2018 ; 41 ( 10 ). doi: https://doi.org/10.1093/sleep/zsy129 WorldCat 2. Figorilli M , et al. The clinical importance of periodic leg movements in sleep . Curr Treat Options Neurol. 2017 ; 19 ( 3 ): 10 . Google Scholar Crossref Search ADS PubMed WorldCat 3. ICSD-3 Online Version—American Academy of Sleep Medicine (AASM) [Internet]. AASM . 2015 . http://www.aasmnet.org/store/product.aspx?pid=849. Accessed December 17, 2018. COPAC 4. Allen RP , et al. ; International Restless Legs Syndrome Study Group . Restless legs syndrome/Willis-Ekbom disease diagnostic criteria: updated International Restless Legs Syndrome Study Group (IRLSSG) consensus criteria–history, rationale, description, and significance . Sleep Med. 2014 ; 15 ( 8 ): 860 – 873 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Dauvilliers Y , et al. Periodic leg movements during sleep and wakefulness in narcolepsy . J Sleep Res. 2007 ; 16 ( 3 ): 333 – 339 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Fantini ML , et al. Periodic leg movements in REM sleep behavior disorder and related autonomic and EEG activation . Neurology . 2002 ; 59 ( 12 ): 1889 – 1894 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Baran AS , et al. Change in periodic limb movement index during treatment of obstructive sleep apnea with continuous positive airway pressure . Sleep . 2003 ; 26 ( 6 ): 717 – 720 . Google Scholar Crossref Search ADS PubMed WorldCat 8. Chervin RD . Periodic leg movements and sleepiness in patients evaluated for sleep-disordered breathing . Am J Respir Crit Care Med. 2001 ; 164 ( 8 Pt 1 ): 1454 – 1458 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Al-Alawi A , et al. Prevalence, risk factors and impact on daytime sleepiness and hypertension of periodic leg movements with arousals in patients with obstructive sleep apnea . J Clin Sleep Med. 2006 ; 2 ( 3 ): 281 – 287 . Google Scholar PubMed WorldCat 10. Leary EB , et al. Periodic limb movements in sleep: prevalence and associated sleepiness in the Wisconsin Sleep Cohort . Clin Neurophysiol. 2018 ; 129 ( 11 ): 2306 – 2314 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Szentkirályi A , et al. Prevalence and associated risk factors of periodic limb movement in sleep in two German population-based studies . Sleep . 2018 Dec 14. doi: https://doi.org/10.1093/sleep/zsy237 WorldCat 12. Haba-Rubio J , et al. Prevalence and determinants of periodic limb movements in the general population . Ann Neurol. 2016 ; 79 ( 3 ): 464 – 474 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Pennestri MH , et al. Nocturnal blood pressure changes in patients with restless legs syndrome . Neurology. 2007 ; 68 ( 15 ): 1213 – 1218 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Siddiqui F , et al. Rise of blood pressure with periodic limb movements in sleep and wakefulness . Clin Neurophysiol. 2007 ; 118 ( 9 ): 1923 – 1930 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Ferrillo F , et al. Changes in cerebral and autonomic activity heralding periodic limb movements in sleep . Sleep Med. 2004 ; 5 ( 4 ): 407 – 412 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Koo BB , et al. ; Osteoporotic Fractures in Men (MrOS) Study Group . Association of incident cardiovascular disease with periodic limb movements during sleep in older men: outcomes of sleep disorders in older men (MrOS) study . Circulation. 2011 ; 124 ( 11 ): 1223 – 1231 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Iber C , et al. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology, and Technical Specifications .1st ed. Westchester, IL : American Academy of Sleep Medicine ; 2007 . Google Preview WorldCat COPAC 18. Berry RB , et al. The AASM Manual for the Scoring of Sleep and Associated Events: Rules, Terminology and Technical Specifications, Version 2.2 . Darien, IL : American Academy of Sleep Medicine ; 2015 . Google Preview WorldCat COPAC 19. Recording and scoring leg movements . The atlas task force . Sleep . 1993 ; 16 ( 8 ): 748 – 759 . Crossref Search ADS PubMed WorldCat 20. Coleman RM . Periodic movements in sleep (nocturnal myoclonus) and restless legs syndrome. In: Guilleminault C , ed. Sleeping and Waking Disorders: Indications and Techniques . Menlo Park, CA : Addison-Wesley ; 1982 : 265 – 295 . Google Preview WorldCat COPAC 21. Zucconi M , et al. ; International Restless Legs Syndrome Study Group (IRLSSG) . The official World Association of Sleep Medicine (WASM) standards for recording and scoring periodic leg movements in sleep (PLMS) and wakefulness (PLMW) developed in collaboration with a task force from the International Restless Legs Syndrome Study Group (IRLSSG) . Sleep Med. 2006 ; 7 ( 2 ): 175 – 183 . Google Scholar Crossref Search ADS PubMed WorldCat 22. Ferri R , et al. ; International and European Restless Legs Syndrome Study Groups (IRLSSG and EURLSSG) . World Association of Sleep Medicine (WASM) 2016 standards for recording and scoring leg movements in polysomnograms developed by a joint task force from the International and the European Restless Legs Syndrome Study Groups (IRLSSG and EURLSSG) . Sleep Med. 2016 ; 26 : 86 – 95 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Ferri R , et al. Quantifying leg movement activity during sleep . Sleep Med Clin. 2016 ; 11 ( 4 ): 413 – 420 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Ferri R . The time structure of leg movement activity during sleep: The theory behind the practice . Sleep Med . 2012 ; 13 ( 4 ): 433 – 441 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Ferri R , et al. Putting the periodicity back into the periodic leg movement index: an alternative data-driven algorithm for the computation of this index during sleep and wakefulness . Sleep Med. 2015 ; 16 ( 10 ): 1229 – 1235 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Scholle S , et al. Leg movements and periodic leg movements during sleep in the development across childhood and adolescence from 1 to 18 years . Sleep Med. 2014 ; 15 ( 9 ): 1068 – 1074 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Ferri R , et al. Leg movement activity during sleep in school-age children and adolescents: a detailed study in normal controls and participants with restless legs syndrome and narcolepsy type 1 . Sleep . 2018 ; 41 ( 4 ). doi: https://doi.org/10.1093/sleep/zsy010 WorldCat 28. Ferri R , et al. Diagnostic accuracy of the standard and alternative periodic leg movement during sleep indices for restless legs syndrome . Sleep Med. 2016 ; 22 : 97 – 99 . Google Scholar Crossref Search ADS PubMed WorldCat 29. Ferri R , et al. New approaches to the study of periodic leg movements during sleep in restless legs syndrome . Sleep . 2006 ; 29 ( 6 ): 759 – 769 . Google Scholar PubMed WorldCat 30. Ferri R , et al. Different periodicity and time structure of leg movements during sleep in narcolepsy/cataplexy and restless legs syndrome . Sleep . 2006 ; 29 ( 12 ): 1587 – 1594 . Google Scholar Crossref Search ADS PubMed WorldCat 31. Manconi M , et al. Time structure analysis of leg movements during sleep in REM sleep behavior disorder . Sleep . 2007 ; 30 ( 12 ): 1779 – 1785 . Google Scholar Crossref Search ADS PubMed WorldCat 32. Ferri R , et al. Periodic leg movements during sleep and periodic limb movement disorder in patients presenting with unexplained insomnia . Clin Neurophysiol. 2009 ; 120 ( 2 ): 257 – 263 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Ferri R , et al. Sleep polygraphic study of children and adolescents with narcolepsy/cataplexy . Dev Neuropsychol. 2009 ; 34 ( 5 ): 523 – 538 . Google Scholar Crossref Search ADS PubMed WorldCat 34. Manconi M , et al. Periodic leg movements in Cheyne-Stokes respiration . Eur Respir J. 2008 ; 32 ( 6 ): 1656 – 1662 . Google Scholar Crossref Search ADS PubMed WorldCat 35. Manconi M , et al. First night efficacy of pramipexole in restless legs syndrome and periodic leg movements . Sleep Med. 2007 ; 8 ( 5 ): 491 – 497 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Manconi M , et al. . Defining the boundaries of the response of sleep leg movements to a single dose of dopamine agonist . Sleep . 2008 ; 31 ( 9 ): 1229 – 1237 . Google Scholar PubMed WorldCat 37. Ferri R , et al. Acute dopamine-agonist treatment in restless legs syndrome: effects on sleep architecture and NREM sleep instability . Sleep . 2010 ; 33 ( 6 ): 793 – 800 . Google Scholar Crossref Search ADS PubMed WorldCat 38. Manconi M , et al. Effects of acute dopamine-agonist treatment in restless legs syndrome on heart rate variability during sleep . Sleep Med. 2011 ; 12 ( 1 ): 47 – 55 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Manconi M , et al. Pramipexole versus ropinirole: polysomnographic acute effects in restless legs syndrome . Mov Disord. 2011 ; 26 ( 5 ): 892 – 895 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Manconi M , et al. Preferential D2 or preferential D3 dopamine agonists in restless legs syndrome . Neurology. 2011 ; 77 ( 2 ): 110 – 117 . Google Scholar Crossref Search ADS PubMed WorldCat 41. Ferri R , et al. Age-related changes in periodic leg movements during sleep in patients with restless legs syndrome . Sleep Med. 2008 ; 9 ( 7 ): 790 – 798 . Google Scholar Crossref Search ADS PubMed WorldCat 42. Miano S , et al. Sleep breathing and periodic leg movement pattern in Angelman Syndrome: a polysomnographic study . Clin Neurophysiol. 2005 ; 116 ( 11 ): 2685 – 2692 . Google Scholar PubMed WorldCat 43. Bruni O , et al. New approaches to the study of leg movements during sleep in ADHD children, In: Proceedings of the 20th Meeting of the Associated Sleep Societies . Salt Lake City ; 2006 . 44. Ferri R , et al. Time structure of leg movement activity during sleep in attention-deficit/hyperactivity disorder and effects of levodopa . Sleep Med. 2013 ; 14 ( 4 ): 359 – 366 . Google Scholar Crossref Search ADS PubMed WorldCat 45. Garbazza C , et al. Leg Movement Activity During Sleep in Adults With Attention-Deficit/Hyperactivity Disorder . Front Psychiatry. 2018 ; 9 : 179 . Google Scholar Crossref Search ADS PubMed WorldCat 46. Huang YS , et al. Sleep disorders in Taiwanese children with attention deficit/hyperactivity disorder . J Sleep Res. 2004 ; 13 ( 3 ): 269 – 277 . Google Scholar Crossref Search ADS PubMed WorldCat 47. Sadeh A , et al. Sleep in children with attention-deficit hyperactivity disorder: a meta-analysis of polysomnographic studies . Sleep Med Rev. 2006 ; 10 ( 6 ): 381 – 398 . Google Scholar Crossref Search ADS PubMed WorldCat 48. Picchietti DL , et al. Further studies on periodic limb movement disorder and restless legs syndrome in children with attention-deficit hyperactivity disorder . Mov Disord . 1999 ; 14 ( 6 ): 1000 – 1007 . Google Scholar Crossref Search ADS PubMed WorldCat 49. Martinez S , et al. Periodic leg movements in prepubertal children with sleep disturbance . Dev Med Child Neurol . 2004 ; 46 ( 11 ): 765 – 770 . Google Scholar Crossref Search ADS PubMed WorldCat 50. Golan N , et al. Sleep disorders and daytime sleepiness in children with attention-deficit/hyperactive disorder . Sleep . 2004 ; 27 ( 2 ): 261 – 266 . Google Scholar Crossref Search ADS PubMed WorldCat 51. Gruber R , et al. Sleep disturbances in prepubertal children with attention deficit hyperactivity disorder: a home polysomnography study . Sleep . 2009 ; 32 ( 3 ): 343 – 350 . Google Scholar Crossref Search ADS PubMed WorldCat 52. Prihodova I , et al. Sleep disorders and daytime sleepiness in children with attention-deficit/hyperactivity disorder: a two-night polysomnographic study with a multiple sleep latency test . Sleep Med. 2010 ; 11 ( 9 ): 922 – 928 . Google Scholar Crossref Search ADS PubMed WorldCat 53. Cooper J , et al. No evidence of sleep apnea in children with attention deficit hyperactivity disorder . Clin Pediatr (Phila). 2004 ; 43 ( 7 ): 609 – 614 . Google Scholar Crossref Search ADS PubMed WorldCat 54. Lopez R , et al. Restless legs syndrome and iron deficiency in adults with attention deficit hyperactivity disorder . Sleep . 2019 . doi: https://doi.org/10.1093/sleep/zsz027 WorldCat 55. American Psychiatric Association . Diagnostic and Statistical Manual of Mental Disorders. 5th ed. Arlington, VA : American Psychiatric Association ; 2013 . WorldCat COPAC 56. Picchietti DL , et al. Periodic limb movement disorder and restless legs syndrome in children with attention-deficit hyperactivity disorder . J Child Neurol. 1998 ; 13 ( 12 ): 588 – 594 . Google Scholar Crossref Search ADS PubMed WorldCat 57. Chervin RD , et al. Associations between symptoms of inattention, hyperactivity, restless legs, and periodic leg movements . Sleep . 2002 ; 25 ( 2 ): 213 – 218 . Google Scholar PubMed WorldCat 58. Gaultney JF , et al. Parent-reported periodic limb movement, sleep disordered breathing, bedtime resistance behaviors, and ADHD . Behav Sleep Med. 2005 ; 3 ( 1 ): 32 – 43 . Google Scholar Crossref Search ADS PubMed WorldCat 59. Wagner ML , et al. Symptoms of attention-deficit/hyperactivity disorder in adults with restless legs syndrome . Sleep . 2004 ; 27 ( 8 ): 1499 – 1504 . Google Scholar Crossref Search ADS PubMed WorldCat 60. Cortese S , et al. Restless legs syndrome and attention-deficit/hyperactivity disorder: a review of the literature . Sleep . 2005 ; 28 ( 8 ): 1007 – 1013 . Google Scholar Crossref Search ADS PubMed WorldCat 61. Allen R . Dopamine and iron in the pathophysiology of restless legs syndrome (RLS) . Sleep Med. 2004 ; 5 ( 4 ): 385 – 391 . Google Scholar Crossref Search ADS PubMed WorldCat 62. Konofal E , et al. Effects of iron supplementation on attention deficit hyperactivity disorder in children . Pediatr Neurol. 2008 ; 38 ( 1 ): 20 – 26 . Google Scholar Crossref Search ADS PubMed WorldCat 63. Konofal E , et al. Impact of restless legs syndrome and iron deficiency on attention-deficit/hyperactivity disorder in children . Sleep Med. 2007 ; 8 ( 7-8 ): 711 – 715 . Google Scholar Crossref Search ADS PubMed WorldCat 64. Earley CJ , et al. Insight into the pathophysiology of restless legs syndrome . J Neurosci Res. 2000 ; 62 ( 5 ): 623 – 628 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Dauvilliers, Yves; Arnulf, Isabelle; Szakacs, Zoltan; Leu-Semenescu, Smaranda; Lecomte, Isabelle; Scart-Gres, Catherine; Lecomte, Jeanne-Marie; Schwartz, Jean-Charles; HARMONY III study group ; Arnulf, Isabelle; Bastuji, Hélène; Dauvilliers, Yves; Vieccherini, Marie Françoise; Pepin, Jean Louis; Quera Salva, Maria Antonia; Stoll, Anne Thibault; Szakacs, Zoltan

May, Anna M; May, Ryan D; Bena, James; Wang, Lu; Monahan, Ken; Stone, Katie L; Barrett-Connor, Elizabeth; Koo, Brian B; Winkelman, John W; Redline, Susan; Mittleman, Murray A; Mehra, Reena; for the Osteoporotic Fractures in Men (MrOS) Study Group ; Stone, K L; Bauer, D C; Cummings, S R; Goldschlager, N; Varosy, P; Yaffe, K;

Kessler,, Riley;Knutson, Kristen, L;Mokhlesi,, Babak;Anderson, Samantha, L;Shah,, Monica;Meltzer, David, O;Arora, Vineet, M

doi: 10.1093/sleep/zsz153pmid: 31310317

Abstract Study Objectives Although sleep disturbance is common in acutely ill patients during and after a hospitalization, how hospitalization affects sleep in general medicine patients has not been well characterized. We describe how sleep and activity patterns vary during and after hospitalization in a small population of older, predominately African American general medicine patients. Methods Patients wore a wrist accelerometer during hospitalization and post-discharge to provide objective measurements of sleep duration, efficiency, and physical activity. Random effects linear regression models clustered by subject were used to test associations between sleep and activity parameters across study days from hospitalization through post-discharge. Results We recorded 404 nights and 384 days from 54 patients. Neither nighttime sleep duration nor sleep efficiency increased from hospitalization through post-discharge (320.2 vs. 320.2 min, p = 0.99; 74.0% vs. 71.7%, p = 0.24). Daytime sleep duration also showed no significant change (26.3 vs. 25.8 min/day, p = 0.5). Daytime physical activity was significantly less in-hospital compared to post-discharge (128.6 vs. 173.2 counts/min, p < 0.01) and increased 23.3 counts/min (95% CI = 16.5 to 30.6, p < 0.01) per hospital day. A study day and post-discharge period interaction was observed demonstrating slowed recovery of activity post-discharge (β 3 = −20.8, 95% CI = −28.8 to −12.8, p < 0.01). Conclusions Nighttime sleep duration and efficiency and daytime sleep duration were similar in-hospital and post-discharge. Daytime physical activity, however, was greater post-discharge and increased more rapidly during hospitalization than post-discharge. Interventions, both in hospital and at home, to restore patient sleep and sustain activity improvements may improve patient recovery from illness. sleep, activity, hospital, home, older adults Statement of Significance Disrupted sleep during hospitalization can hinder recovery from illness and continued disruption after patients return home may leave them vulnerable to readmission. No study to date has documented sleep and activity patterns both during hospitalization on a general medicine floor and immediately after discharge. We show that our patients’ sleep disturbance persisted post-discharge with no significant change in sleep duration, sleep efficiency, or daytime sleep. There was, however, a trajectory of increasing physical activity across study days. It is important to investigate whether patient sleep and activity patterns are eventually restored after discharge or if hospitalization is associated with chronic sleep disorders. Future research on how sleep and activity trajectories could predict patient outcomes is warranted. Introduction Patients are in a particularly vulnerable state during the time just after discharge from the hospital. Not only are they recovering from their illnesses, but they also continue to be affected by stressors associated with hospitalization. Altered sleep patterns, reduced activity, poor nutrition, pain and discomfort, new medications, and mental stress can all affect recovery [1, 2]. The cumulative effect may result in posthospital syndrome, a high-risk period after discharge when patients may acquire adverse medical conditions leading to re-hospitalization [1, 2]. More than one in six Medicare patients discharged from the hospital are readmitted within 30 days and many are readmitted for a different illness than their index admission, indicating a period of increased vulnerability [1–3]. Thus it has become important to consider the factors associated with a hospital admission that may impair recovery post-discharge and predispose patients to readmission. Both sleep and physical activity patterns are disrupted during hospitalization and may affect patient recovery [4–8]. Sleep deprivation is harmful to immune [9, 10], metabolic [11, 12], and cardiovascular health [13, 14], allostasis [2], and cognition and memory [15, 16]. These effects may be more significant in older adults who may have preexisting vulnerabilities. Disruptions to the circadian sleep–wake cycle can further exacerbate these problems by imposing a jetlag-like syndrome leading to excessive fatigue and daytime sleepiness that can limit active participation in one’s medical care and rehabilitation [17]. Furthermore, lack of physical activity can leave the patient deconditioned, limiting self-care and increasing the chance of falls. Thus acute disturbance of sleep and physical activity patterns associated with hospitalization may hinder recovery and continued disruption post-discharge may increase risk for readmission. Previous studies have demonstrated sleep disturbance weeks to months after critical illness requiring an intensive care unit (ICU) stay and after cardiac surgery [18–22]. However, no study has used actigraphy to characterize sleep and activity patterns in patients both during their inpatient stays on a general medicine service and immediately post-discharge. Characterizing these patterns is an important step in understanding how they may influence a patient’s recovery and risk for 30-day readmission and is a critical step in targeting interventions to address posthospital syndrome. This study aims to provide insight into immediate post-discharge sleep and activity patterns in a cohort of predominately African American patients cared for on an inpatient general medicine service. Methods Subjects Subjects were a subset of a larger sleep study of hospitalized patients conducted at the University of Chicago Medical Center [23]. Patients eligible for actigraphy monitoring were ambulatory adults age 50 years or older hospitalized on the general medicine or hematology/oncology services and living in the community before admission. Patients were considered ambulatory if they reported being able to walk across the room independently or with a walker. As our sample was drawn from a general medicine service, no neurology patients were included in the study (Table 1). Exclusion criteria included being transferred from the ICU or another hospital, being in droplet or airborne isolation, cognitive impairment as measured by the Mini-Mental State Examination or Short Portable Mental Status Questionnaire [24, 25], having a documented sleep disorder (including chronic insomnia, obstructive sleep apnea, sleep disordered breathing, and restless legs syndrome), being on bed rest, and having been hospitalized for greater than 72 hours prior to enrollment. Exclusion criteria were designed to ensure participants’ actigraphy results reflected their sleep and activity as any of the aforementioned conditions may affect patients’ mobility or sleep/wake patterns. Patients were eligible if they wore the actigraphy monitor for at least one night in the hospital and at least one night post-discharge. Patients received a total of $50; they received the first $25 as they were discharged from the hospital and the second $25 after completing a 2-week follow-up assessment by phone. The University of Chicago institutional review board approved this study and all patients provided written consent. Table 1. Sample characteristics (N = 54) Characteristic Value Age (years) 61 ± 9.4 African American, n (%) 43 (79.6%) Female, n (%) 32 (59.3%) Major diagnostic category for primary diagnoses  Diseases and disorders of the digestive system, hepatobiliary system and pancreas 16  Other* 11  Diseases and disorders of the respiratory system 10  Diseases and disorders of the circulatory system, blood, blood forming organs, immunologic disorders 7  Endocrine, nutritional, and metabolic diseases and disorders 5  Diseases and disorders of the kidney and urinary tract 5 Mean global PSQI score (mean ± SD)† 9.5 ± 5.3 Number of patients with global PSQI >5‡ 36 (75%) Baseline self-reported sleep duration (mean ± SD)§  Weekday (minutes) 348 ± 114  Weekend (minutes) 362 ± 120 Mean ESS score (mean ± SD) 8.2 ± 5.5 Number of patients with ESS >10|| 19 (35.2%) Total number of nights recorded 404 Total number of days recorded 384 Mean length of stay in the hospital (days, mean ± SD) 3.6 ± 1.6 Mean follow-up time post-discharge (nights, mean ± SD) 6.4 ± 1.6 Number of patients readmitted within 30-days, n (%) 3 (5.6%) Mean AHRQ Elixhauser Comorbidity Index (mean ± SD, min, max)  30-day readmission score 15.3 ± 12.1, −3, 47  In-hospital mortality score 3.7 ± 8.8, −11, 28 Characteristic Value Age (years) 61 ± 9.4 African American, n (%) 43 (79.6%) Female, n (%) 32 (59.3%) Major diagnostic category for primary diagnoses  Diseases and disorders of the digestive system, hepatobiliary system and pancreas 16  Other* 11  Diseases and disorders of the respiratory system 10  Diseases and disorders of the circulatory system, blood, blood forming organs, immunologic disorders 7  Endocrine, nutritional, and metabolic diseases and disorders 5  Diseases and disorders of the kidney and urinary tract 5 Mean global PSQI score (mean ± SD)† 9.5 ± 5.3 Number of patients with global PSQI >5‡ 36 (75%) Baseline self-reported sleep duration (mean ± SD)§  Weekday (minutes) 348 ± 114  Weekend (minutes) 362 ± 120 Mean ESS score (mean ± SD) 8.2 ± 5.5 Number of patients with ESS >10|| 19 (35.2%) Total number of nights recorded 404 Total number of days recorded 384 Mean length of stay in the hospital (days, mean ± SD) 3.6 ± 1.6 Mean follow-up time post-discharge (nights, mean ± SD) 6.4 ± 1.6 Number of patients readmitted within 30-days, n (%) 3 (5.6%) Mean AHRQ Elixhauser Comorbidity Index (mean ± SD, min, max)  30-day readmission score 15.3 ± 12.1, −3, 47  In-hospital mortality score 3.7 ± 8.8, −11, 28 *Diseases and disorders of the ear, nose, mouth, and throat; diseases and disorders of the musculoskeletal system and connective tissue; diseases and disorders of the skin, subcutaneous tissue, and breast; myeloproliferative diseases and disorders, poorly differentiated neoplasms; infectious and parasitic diseases, systemic or unspecified sites; mental diseases and disorders; alcohol/drug use and alcohol-/drug-induced organic mental disorders; human immunodeficiency virus infections. †N = 48. ‡PSQI > 5 represents poor sleep quality. §Based on self-reported sleep from month before admission on the PSQI. ||ESS > 10 represents excessive daytime sleepiness. Open in new tab Table 1. Sample characteristics (N = 54) Characteristic Value Age (years) 61 ± 9.4 African American, n (%) 43 (79.6%) Female, n (%) 32 (59.3%) Major diagnostic category for primary diagnoses  Diseases and disorders of the digestive system, hepatobiliary system and pancreas 16  Other* 11  Diseases and disorders of the respiratory system 10  Diseases and disorders of the circulatory system, blood, blood forming organs, immunologic disorders 7  Endocrine, nutritional, and metabolic diseases and disorders 5  Diseases and disorders of the kidney and urinary tract 5 Mean global PSQI score (mean ± SD)† 9.5 ± 5.3 Number of patients with global PSQI >5‡ 36 (75%) Baseline self-reported sleep duration (mean ± SD)§  Weekday (minutes) 348 ± 114  Weekend (minutes) 362 ± 120 Mean ESS score (mean ± SD) 8.2 ± 5.5 Number of patients with ESS >10|| 19 (35.2%) Total number of nights recorded 404 Total number of days recorded 384 Mean length of stay in the hospital (days, mean ± SD) 3.6 ± 1.6 Mean follow-up time post-discharge (nights, mean ± SD) 6.4 ± 1.6 Number of patients readmitted within 30-days, n (%) 3 (5.6%) Mean AHRQ Elixhauser Comorbidity Index (mean ± SD, min, max)  30-day readmission score 15.3 ± 12.1, −3, 47  In-hospital mortality score 3.7 ± 8.8, −11, 28 Characteristic Value Age (years) 61 ± 9.4 African American, n (%) 43 (79.6%) Female, n (%) 32 (59.3%) Major diagnostic category for primary diagnoses  Diseases and disorders of the digestive system, hepatobiliary system and pancreas 16  Other* 11  Diseases and disorders of the respiratory system 10  Diseases and disorders of the circulatory system, blood, blood forming organs, immunologic disorders 7  Endocrine, nutritional, and metabolic diseases and disorders 5  Diseases and disorders of the kidney and urinary tract 5 Mean global PSQI score (mean ± SD)† 9.5 ± 5.3 Number of patients with global PSQI >5‡ 36 (75%) Baseline self-reported sleep duration (mean ± SD)§  Weekday (minutes) 348 ± 114  Weekend (minutes) 362 ± 120 Mean ESS score (mean ± SD) 8.2 ± 5.5 Number of patients with ESS >10|| 19 (35.2%) Total number of nights recorded 404 Total number of days recorded 384 Mean length of stay in the hospital (days, mean ± SD) 3.6 ± 1.6 Mean follow-up time post-discharge (nights, mean ± SD) 6.4 ± 1.6 Number of patients readmitted within 30-days, n (%) 3 (5.6%) Mean AHRQ Elixhauser Comorbidity Index (mean ± SD, min, max)  30-day readmission score 15.3 ± 12.1, −3, 47  In-hospital mortality score 3.7 ± 8.8, −11, 28 *Diseases and disorders of the ear, nose, mouth, and throat; diseases and disorders of the musculoskeletal system and connective tissue; diseases and disorders of the skin, subcutaneous tissue, and breast; myeloproliferative diseases and disorders, poorly differentiated neoplasms; infectious and parasitic diseases, systemic or unspecified sites; mental diseases and disorders; alcohol/drug use and alcohol-/drug-induced organic mental disorders; human immunodeficiency virus infections. †N = 48. ‡PSQI > 5 represents poor sleep quality. §Based on self-reported sleep from month before admission on the PSQI. ||ESS > 10 represents excessive daytime sleepiness. Open in new tab Data collection Objective measurements of sleep and activity were made using actigraphy following a protocol previously described by our group [26]. Patients wore Acitwatch Spectrum Pro (Phillips Respironics, Bend, OR) wrist actigraphy monitors to collect data on sleep duration and quality and activity. The monitor collects acceleration data with a 32 Hertz sampling frequency and returns estimates of sleep timing and quality and has been validated against polysomnography in both healthy individuals and insomniacs [27, 28]. The intensity of physical activity is expressed in activity counts using a proprietary metric developed by Phillips Respironics and derived within the Actiware software from collected accelerometry data. Actiware 5 software was used to calculate sleep duration, sleep efficiency, and activity levels [29]. To estimate nighttime sleep duration and efficiency, the assumed sleep period was determined by the Karolinska Sleep Diary while patients were in the hospital and a sleep log while they were at home [30–32]. The Actiware software was used to derive objective estimates of sleep duration and efficiency based on periods with low activity over all 15-second epochs during the assumed sleep period and time in bed, respectively, based on self reported measures from the Karolinska Sleep Diary and home sleep logs. Sleep duration is the total time spent asleep and sleep efficiency is the ratio of time asleep to time in bed. Similar to previous reports, daytime sleep duration was calculated from Actiware defined minor rest intervals that occurred during the patient reported wake period or between 7 am and 11 pm if patient reported wake time was not available [33]. A value of zero was entered into the database for days when no daytime sleep occurred. Activity was measured as average activity counts per minute over the assumed wake period. Because patients were able to remove the Actiwatch, the number of days and nights recorded per participant represents the total number of days and nights over the study period with adequate sleep and activity data rather than the number of consecutive days and nights they were in the study. To maximize our study population, we included patients who had data from at least one in-hospital and one post-discharge night allowing for intervening non-recorded nights. Our term “study day” does reflect consecutive days in the study. Baseline sleep characteristics Baseline sleep characteristics were assessed upon enrollment using the Epworth Sleepiness Scale (ESS) to measure excessive daytime sleepiness in common situations and the Pittsburgh Sleep Quality Index (PSQI) to measure baseline sleep quality and hygiene over the previous month [34–36]. Baseline patient characteristics Demographic information (age, race, ethnicity, sex) and length of stay were collected from patient charts through an ongoing study of patients admitted to the University of Chicago inpatient general medicine service [23]. To account for severity of disease, we calculated Agency for Healthcare Research and Quality (AHRQ) Elixhauser Comorbidity Index scores for 30-day readmission and in-hospital mortality using 29 binary Elixhauser comorbidity variables according to the algorithm developed by the Healthcare Cost and Utilization Project [37]. Elixhauser comorbidity measures are a group of clinical conditions that can be used in administrative data sets to assign weights to preadmission diseases to derive a single comorbidity score that can be used as an adjustment factor to control for disease severity and can help indentify risk for mortality and 30-day readmission [37]. Data analysis All data were collected and entered into REDCap [38], a secure web application used to create a secure database. Descriptive statistics were used to summarize daytime and nighttime sleep duration, sleep efficiency, and physical activity in-hospital and post-discharge. The primary outcome used for physical activity was average activity counts per minute, which is a frequently reported activity measure from actigraphy [27, 39]. To test for differences between in-hospital and post-discharge nighttime sleep duration, sleep efficiency, daytime sleep duration, and daytime physical activity, the average value for each variable was calculated per subject both in-hospital and post-discharge. When the data were normally distributed (nighttime sleep duration), a paired t-test was used to compare the means. When the data were not normally distributed (nighttime sleep efficiency, daytime sleep duration, and daytime physical activity), a Wilcoxon matched-pairs signed rank test was used to test for a difference between in-hospital and post-discharge variables. To characterize the association between sleep duration, sleep efficiency, daytime sleep duration, and daytime physical activity across each day of the study from the inpatient stay through post-discharge, random effects linear regression models clustered by subject were used to account for multiple observations per participant. We controlled for the following covariates in our final regression model: age (<65 or ≥65 years), sex, ethnicity, and 30-day readmission score for AHRQ Elixhauser Comorbidity Index. When the covariates were significant, we added an interaction term using cross product terms to specifically test the differential effect of study days in the post-discharge period compared to study days while in the hospital. Statistical significance was set at p < 0.05. All data were analyzed in STATA 14.0 (Stata Corp., College Station, TX). Results From October 2012 to November 2017, 404 nights (27% in-hospital, 73% post-discharge) and 384 days (28% in-hospital, 72% post-discharge) were recorded from 54 patients (Table 1). Average number of in-hospital nights and post-discharge nights per participant were 2.1 ± 1.3 and 5.4 ± 2.1 nights, respectively. Average daytime physical activity periods per participant were 2.0 ± 1.2 and 5.1 ± 1.9 days, respectively. Daytime sleep data were available from 45 patients over 333 total days with an average of 2.1 ± 1.3 days in-hospital and 5.3 ± 1.6 days post-discharge per participant. The study population had an average age of 61 ± 9.4 years and was predominately female (59.3%) and African American (79.6%) (Table 1). Average global PSQI at baseline was 9.5 ± 5.3 of 21 with 75% of patients having a score higher than five, representing poor baseline sleep quality. Average baseline ESS score was 8.2 ± 5.5 of 24, with 35.2% of patients recording scores greater than 10, representing excessive daytime sleepiness. Average AHRQ Elixhauser Cormorbidity Index score for 30-day readmission was 15.3 ± 12.1 and for in-hospital mortality was 3.7 ± 8.8. Nighttime sleep duration and efficiency Actigraphy data demonstrated no significant difference in average sleep duration in-hospital compared to post-discharge (320.2 ± 128.5 vs. 320.2 ± 103.0 min; p = 0.99) (Figure 1). Similar to sleep duration, there was no difference in sleep efficiency between settings (median in-hospital = 74.0% vs. median post-discharge = 71.7%, z = −1.2, p = 0.24) (Figure 2). In random effects linear regression models clustered by subject, neither sleep duration nor sleep efficiency changed significantly across study days from hospitalization through post-discharge (1.2 min, 95% CI = −2.5 to 5.0, p = 0.5; −.5%, 95% CI = −1.1 to .1, p = 0.1). None of the covariates were significant in either model. Figure 1. Open in new tabDownload slide Nighttime sleep duration. Left: There is no significant difference between mean in-hospital sleep duration and mean post-discharge sleep duration (320.2 ± 128.53 vs. 320.2 ± 103.03 min, p = 0.99). Right: There is no significant change in sleep duration by study day, where day 0 is the day of discharge (p = 0.5). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Figure 1. Open in new tabDownload slide Nighttime sleep duration. Left: There is no significant difference between mean in-hospital sleep duration and mean post-discharge sleep duration (320.2 ± 128.53 vs. 320.2 ± 103.03 min, p = 0.99). Right: There is no significant change in sleep duration by study day, where day 0 is the day of discharge (p = 0.5). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Figure 2. Open in new tabDownload slide Nighttime sleep efficiency. Left: There is no significant difference between median in-hospital sleep efficiency and median post-discharge sleep efficiency (74.0% vs. 71.7% z = −1.2, p = 0.24). Right: There is no significant change in sleep efficiency by study day, where day 0 is the day of discharge (p = 0.1). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Figure 2. Open in new tabDownload slide Nighttime sleep efficiency. Left: There is no significant difference between median in-hospital sleep efficiency and median post-discharge sleep efficiency (74.0% vs. 71.7% z = −1.2, p = 0.24). Right: There is no significant change in sleep efficiency by study day, where day 0 is the day of discharge (p = 0.1). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Daytime sleep duration and physical activity Like nighttime sleep parameters, daytime sleep duration was similar across settings. Median daytime sleep duration calculated based on 45 patients over a total of 333 days, was not different between in-hospital (26.3 min/day) and post-discharge (25.8 min/day) settings (z = 0.6, p = 0.5) (Figure 3). Random effects linear regression clustered by subject demonstrated no significant change across days of the study (−1.2 min, 95% CI = −3.2 to 0.7, p = 0.2). Daytime physical activity, however, did show significant change. Calculated from all 54 patients over a total of 384 days, daytime physical activity was lower in-hospital (median = 128.6 counts/min) than post-discharge (median = 173.2 counts/min) (z = −4.2, p < 0.01) (Figure 4). Random effects linear regression clustered by subject showed an increase of 23.5 counts/min (95% CI = 16.5 to 30.6, p < 0.01) per hospital day. A significant interaction between study day and post-discharge period was observed (β 3 = −20.8, 95% CI = −28.8 to −12.8, p < 0.01) demonstrating a slower rate of increase after discharge (Table 2). Age was the only significant covariate in the regression model. Table 2. Random effects linear regression models for daytime physical activity β Coefficient P-value 95% Confidence interval Model 1 Study day 9.6 <0.01 6.9 to 12.3 Constant 164.3 <0.01 141.9 to 186.6 Model 2 Study day 23.5 <0.01 16.5 to 30.6 Post-discharge period 2.8 0.77 −16.5 to 22.2 Study day × Post-discharge period −20.8 <0.01 −28.8 to −12.8 Age* −56.4 0.03 −106.3 to −6.5 Sex 19.9 0.37 −23.3 to 63.1 Ethnicity −4.7 0.86 −58.0 to 48.5 AHRQ Elixhauser Comorbidity Index: 30-day readmission score 1.2 0.17 −0.5 to 3.0 Constant 165.5 <0.01 101.5 to 229.5 β Coefficient P-value 95% Confidence interval Model 1 Study day 9.6 <0.01 6.9 to 12.3 Constant 164.3 <0.01 141.9 to 186.6 Model 2 Study day 23.5 <0.01 16.5 to 30.6 Post-discharge period 2.8 0.77 −16.5 to 22.2 Study day × Post-discharge period −20.8 <0.01 −28.8 to −12.8 Age* −56.4 0.03 −106.3 to −6.5 Sex 19.9 0.37 −23.3 to 63.1 Ethnicity −4.7 0.86 −58.0 to 48.5 AHRQ Elixhauser Comorbidity Index: 30-day readmission score 1.2 0.17 −0.5 to 3.0 Constant 165.5 <0.01 101.5 to 229.5 N = 54. *Age ≥ 65 years. Open in new tab Table 2. Random effects linear regression models for daytime physical activity β Coefficient P-value 95% Confidence interval Model 1 Study day 9.6 <0.01 6.9 to 12.3 Constant 164.3 <0.01 141.9 to 186.6 Model 2 Study day 23.5 <0.01 16.5 to 30.6 Post-discharge period 2.8 0.77 −16.5 to 22.2 Study day × Post-discharge period −20.8 <0.01 −28.8 to −12.8 Age* −56.4 0.03 −106.3 to −6.5 Sex 19.9 0.37 −23.3 to 63.1 Ethnicity −4.7 0.86 −58.0 to 48.5 AHRQ Elixhauser Comorbidity Index: 30-day readmission score 1.2 0.17 −0.5 to 3.0 Constant 165.5 <0.01 101.5 to 229.5 β Coefficient P-value 95% Confidence interval Model 1 Study day 9.6 <0.01 6.9 to 12.3 Constant 164.3 <0.01 141.9 to 186.6 Model 2 Study day 23.5 <0.01 16.5 to 30.6 Post-discharge period 2.8 0.77 −16.5 to 22.2 Study day × Post-discharge period −20.8 <0.01 −28.8 to −12.8 Age* −56.4 0.03 −106.3 to −6.5 Sex 19.9 0.37 −23.3 to 63.1 Ethnicity −4.7 0.86 −58.0 to 48.5 AHRQ Elixhauser Comorbidity Index: 30-day readmission score 1.2 0.17 −0.5 to 3.0 Constant 165.5 <0.01 101.5 to 229.5 N = 54. *Age ≥ 65 years. Open in new tab Figure 3. Open in new tabDownload slide Daytime sleep duration. Left: There is no significant difference between median in-hospital daytime sleep duration and median post-discharge daytime sleep duration (26.3 vs. 25.8 min/day, z = 0.6, p = 0.5). Right: There is no significant change in daytime sleep duration by study day, where day 0 is the day of discharge (p = 0.2). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Figure 3. Open in new tabDownload slide Daytime sleep duration. Left: There is no significant difference between median in-hospital daytime sleep duration and median post-discharge daytime sleep duration (26.3 vs. 25.8 min/day, z = 0.6, p = 0.5). Right: There is no significant change in daytime sleep duration by study day, where day 0 is the day of discharge (p = 0.2). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission. Figure 4. Open in new tabDownload slide Daytime physical activity. Left: Median post-discharge daytime physical activity is greater than median in-hospital daytime physical activity (173.2 vs. 128.6 counts/min, z = −4.2, p < 0.01). Right: Daytime physical activity increased 23.5 counts/min (95% CI = 16.5 to 30.6, p < 0.01) per hospital day. A significant interaction between study day and post-discharge period was observed (β3 = −20.8, 95% CI = −28.8 to −12.8, p < 0.01). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission and an interaction term for Study day × Post-discharge. Figure 4. Open in new tabDownload slide Daytime physical activity. Left: Median post-discharge daytime physical activity is greater than median in-hospital daytime physical activity (173.2 vs. 128.6 counts/min, z = −4.2, p < 0.01). Right: Daytime physical activity increased 23.5 counts/min (95% CI = 16.5 to 30.6, p < 0.01) per hospital day. A significant interaction between study day and post-discharge period was observed (β3 = −20.8, 95% CI = −28.8 to −12.8, p < 0.01). Best fit lines are derived from a linear regression model adjusted for age, sex, ethnicity, and AHRQ Elixhauser Comorbidity Index score for 30-day readmission and an interaction term for Study day × Post-discharge. Discussion This is the first study to the authors’ knowledge to characterize sleep and activity patterns in older community-dwelling adults during an inpatient general medicine stay and immediately following discharge to home. Neither sleep duration nor sleep efficiency was significantly improved post-discharge relative to during hospitalization and neither showed significant change across days of the study. Similar to previous findings in critically ill patients, our results suggest that sleep disturbance experienced by general medicine patients while in the hospital may persist once patients are discharged. Although the average post-discharge follow-up time for our study was 6.4 ± 1.6 nights, further research will be necessary to determine how long this effect lasts. Importantly, the nighttime sleep durations we observed both in the hospital and at home are shorter than the recommended 7–9 hours of sleep per night for adults [40]. For patients recovering from illness, this sleep deprivation may be particularly detrimental. The baseline short sleep durations reported on the PSQI by our patient population, suggest they are chronically sleep deprived, as are 65% of Americans who get less than 7 hours of sleep per weeknight [40]. Given the harms of sleep deprivation, it is concerning that we did not observe improved sleep post-discharge when patients should be continuing to recover and are no longer directly subject to the sleep disruptions of the hospital. Even when daytime sleep is factored in, total sleep time remains less than the recommended 7–9 hours per day. Furthermore, the presence of daytime sleep may be a signal that nighttime sleep is not sufficient and/or daytime sleep could delay nighttime sleep and reduce homeostatic sleep drive contributing to short nighttime sleep durations and altered circadian rhythms. Daytime physical activity increased significantly across days of the study and may be a marker of recovery. Average activity in the hospital (132 counts/min) fell between the thresholds for sitting and watching television (67 counts/min) and sitting and eating (177 counts/min) whereas average post-discharge activity (193 counts/min) fell just above the threshold for sitting and being active with one’s hands (190 counts/min) [41]. Age was a significant covariate of daytime physical activity, which likely represents that patients older than 65 years have lower physical activity levels in general. The weaker effect on daytime physical activity post-discharge may reflect an unexpected attenuated trajectory of recovery once patients leave the hospital. Therefore, it may be important to consider more intensive rehabilitation to bridge hospital to home care to maintain the positive trajectory. This may also help patients achieve higher levels of activity intensity at home, beyond that observed in our study. Several limitations to this study must be acknowledged. This is a single center study conducted within a small group of older adult, predominantly African American patients, which limits generalizability. Being a prospective cohort study, it can be used to define associations but not causality. Without baseline pre-hospitalization actigraphy data, it is difficult to know how the observed sleep and activity patterns differ from the patients’ habitual patterns. Although our patients are medical rather than surgical, previous studies have used scheduled operations for pre- and postoperative actigraphy monitoring to show decreased nighttime sleep duration and efficiency and increased daytime sleep during the acute phase of recovery in the hospital compared to baseline [42, 43]. Furthermore, the assumed sleep and wake periods based on self-reported wake and bed times may not be accurate representations of the actual sleep and wake periods. Additionally patients wore actigraphy monitors only on their wrists, rather than on the thigh or waist, limiting the amount of movement data we could collect. Finally, we had to exclude many patients who did not wear the actigraphy monitor for at least one in-hospital and one post-discharge night. It is possible there is a difference in sleep and activity patterns between those who did and did not wear the actigraphy monitor. Because we used only one in-hospital and one post-discharge night, not all days and nights used in the analysis are consecutive, which may have affected our results. There were more nights than days recorded because recordings started on the nights following enrollment and discharge for in-hospital and post-discharge periods, respectively. When patients did remove the actigraphy monitor, it was more likely to be during the day such that we had data from the previous night but not the following day resulting in more nights recorded. Similar to critically ill patients, our results suggest that general medicine patients continue to experience sleep disruption post-discharge. Furthermore, they may even show a slower trajectory to recovery once they are discharged. These findings are important in the context of posthospital syndrome. Continued disturbance to sleep and activity patterns may be an important, and modifiable, contributor to the increased vulnerability after discharge from the hospital. Interventions focused on limiting sleep disturbance in the hospital and emphasizing the importance of sleep after discharge may be able to improve patient recovery and reduce readmissions. Funding We acknowledge funding from the National Heart, Lung, and Blood Institute (NHLBI) (5R25HL116372 and 1K24HL136859), Society for Hospital Medicine Student Hospitalist Scholar Grant, the Pritzker School of Medicine Scholarship and Discovery program, and National Center for Advancing Translational Sciences (NCATS) (UL1 TR000430). Conflict of interest statement. None declared. References 1. Krumholz HM . Post-hospital syndrome—an acquired, transient condition of generalized risk . N Engl J Med. 2013 ; 368 ( 2 ): 100 – 102 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Goldwater DS , et al. Is posthospital syndrome a result of hospitalization-induced allostatic overload? J Hosp Med . 2018 ; 13 ( 5 ): E1 – E9 . doi: https://doi.org/10.12788/jhm.2986 Google Scholar Crossref Search ADS PubMed WorldCat 3. Yale New Haven Health Services Corporation Center for Outcomes Research and Evaluation for Centers for Medicare & Medicaid Services . Medicare Hospital Quality Chartbook 2017: Performance Report on Outcome Measures. https://www.cms.gov/Medicare/Quality-Initiatives-Patient-Assessment-Instruments/HospitalQualityInits/OutcomeMeasures.html. Accessed May 25, 2018. 4. Fisher SR , et al. Ambulatory activity of older adults hospitalized with acute medical illness . J Am Geriatr Soc. 2011 ; 59 ( 1 ): 91 – 95 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Callen BL , et al. Frequency of hallway ambulation by hospitalized older adults on medical units of an academic hospital . Geriatr Nurs. 2004 ; 25 ( 4 ): 212 – 217 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Brown CJ , et al. The underrecognized epidemic of low mobility during hospitalization of older adults . J Am Geriatr Soc. 2009 ; 57 ( 9 ): 1660 – 1665 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Yoder JC , et al. Noise and sleep among adult medical inpatients: far from a quiet night . Arch Intern Med. 2012 ; 172 ( 1 ): 68 – 70 . Google Scholar Crossref Search ADS PubMed WorldCat 8. Missildine K , et al. Sleep in hospitalized elders: a pilot study . Geriatr Nurs. 2010 ; 31 ( 4 ): 263 – 271 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Prather AA , et al. Sleep and antibody response to hepatitis B vaccination . Sleep. 2012 ; 35 ( 8 ): 1063 – 1069 . Google Scholar PubMed WorldCat 10. Irwin MR , et al. Sleep health: reciprocal regulation of sleep and innate immunity . Neuropsychopharmacology. 2017 ; 42 ( 1 ): 129 – 155 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Spiegel K , et al. Impact of sleep debt on metabolic and endocrine function . Lancet. 1999 ; 354 ( 9188 ): 1435 – 1439 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Knutson KL , et al. The metabolic consequences of sleep deprivation . Sleep Med Rev. 2007 ; 11 ( 3 ): 163 – 178 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Arora VM , et al. Objective sleep duration and quality in hospitalized older adults: associations with blood pressure and mood . J Am Geriatr Soc. 2011 ; 59 ( 11 ): 2185 – 2186 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Khan MS , et al. The effects of insomnia and sleep loss on cardiovascular disease . Sleep Med Clin. 2017 ; 12 ( 2 ): 167 – 177 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Chambers AM . The role of sleep in cognitive processing: focusing on memory consolidation . Wiley Interdiscip Rev Cogn Sci . 2017 ; 8 ( 3 ): 1 – 14 . doi: https://doi.org/10.1002/wcs.1433 Google Scholar Crossref Search ADS WorldCat 16. Deak MC , et al. Sleep and cognition . Wiley Interdiscip Rev Cogn Sci. 2010 ; 1 ( 4 ): 491 – 500 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Stewart NH , et al. Sleep in hospitalized older adults . Sleep Med Clin. 2018 ; 13 ( 1 ): 127 – 135 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Altman MT , et al. Sleep disturbance after hospitalization and critical illness: a systematic review . Ann Am Thorac Soc. 2017 ; 14 ( 9 ): 1457 – 1468 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Redeker NS . Symptoms reported by older and middle-aged adults after coronary bypass surgery . Clin Nurs Res. 1993 ; 2 ( 2 ): 148 – 159 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Redeker NS , et al. Sleep patterns in women after coronary artery bypass surgery . Appl Nurs Res. 1996 ; 9 ( 3 ): 115 – 122 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Schaefer KM , et al. Sleep disturbances post coronary artery bypass surgery . Prog Cardiovasc Nurs. 1996 ; 11 ( 1 ): 5 – 14 . Google Scholar PubMed WorldCat 22. Solverson KJ , et al. Assessment of sleep quality post-hospital discharge in survivors of critical illness . Respir Med. 2016 ; 114 : 97 – 102 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Meltzer D , et al. Effects of physician experience on costs and outcomes on an academic general medicine service: results of a trial of hospitalists . Ann Intern Med. 2002 ; 137 ( 11 ): 866 – 874 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Folstein MF , et al. “Mini-mental state”. A practical method for grading the cognitive state of patients for the clinician . J Psychiatr Res. 1975 ; 12 ( 3 ): 189 – 198 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Pfeiffer E . A short portable mental status questionnaire for the assessment of organic brain deficit in elderly patients . J Am Geriatr Soc. 1975 ; 23 ( 10 ): 433 – 441 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Adachi M , et al. Perceived control and sleep in hospitalized older adults: a sound hypothesis? J Hosp Med. 2013 ; 8 ( 4 ): 184 – 190 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Murphy SL . Review of physical activity measurement using accelerometers in older adults: considerations for research design and conduct . Prev Med. 2009 ; 48 ( 2 ): 108 – 114 . Google Scholar Crossref Search ADS PubMed WorldCat 28. Jean-Louis G , et al. The actigraph data analysis software: I. A novel approach to scoring and interpreting sleep-wake activity . Percept Mot Skills. 1997 ; 85 ( 1 ): 207 – 216 . Google Scholar Crossref Search ADS PubMed WorldCat 29. Chae KY , et al. Evaluation of immobility time for sleep latency in actigraphy . Sleep Med. 2009 ; 10 ( 6 ): 621 – 625 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Akerstedt T , et al. Subjective and objective sleepiness in the active individual . Int J Neurosci. 1990 ; 52 ( 1-2 ): 29 – 37 . Google Scholar Crossref Search ADS PubMed WorldCat 31. Alessi CA , et al. More daytime sleeping predicts less functional recovery among older people undergoing inpatient post-acute rehabilitation . Sleep. 2008 ; 31 ( 9 ): 1291 – 1300 . Google Scholar PubMed WorldCat 32. Fung CH , et al. Sleep disturbance among older adults in assisted living facilities . Am J Geriatr Psychiatry. 2012 ; 20 ( 6 ): 485 – 493 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Jakubowski KP , et al. Poor sleep moderates the relationship between daytime napping and inflammation in Black and White men . Sleep Health. 2017 ; 3 ( 5 ): 328 – 335 . Google Scholar Crossref Search ADS PubMed WorldCat 34. Johns MW . A new method for measuring daytime sleepiness: the Epworth Sleepiness Scale . Sleep. 1991 ; 14 ( 6 ): 540 – 545 . Google Scholar Crossref Search ADS PubMed WorldCat 35. Johns MW . Reliability and factor analysis of the Epworth Sleepiness Scale . Sleep. 1992 ; 15 ( 4 ): 376 – 381 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Buysse DJ , et al. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research . Psychiatry Res. 1989 ; 28 ( 2 ): 193 – 213 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Moore BJ , et al. Identifying increased risk of readmission and in-hospital mortality using hospital administrative data: the AHRQ Elixhauser Comorbidity Index . Med Care. 2017 ; 55 ( 7 ): 698 – 705 . Google Scholar Crossref Search ADS PubMed WorldCat 38. Harris PA , et al. Research electronic data capture (REDCap)—a metadata-driven methodology and workflow process for providing translational research informatics support . J Biomed Inform. 2009 ; 42 ( 2 ): 377 – 381 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Troiano RP , et al. Physical activity in the United States measured by accelerometer . Med Sci Sports Exerc. 2008 ; 40 ( 1 ): 181 – 188 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Sleep Health Index: Quarterly Report—Q1 2018 . Alrington, VA: National Sleep Foundation . https://nsf.parc.us.com/client/index.html#/search. Accessed June 6, 2018. WorldCat COPAC 41. Beveridge C , et al. Daytime physical activity and sleep in hospitalized older adults: association with demographic characteristics and disease severity . J Am Geriatr Soc. 2015 ; 63 ( 7 ): 1391 – 1400 . Google Scholar Crossref Search ADS PubMed WorldCat 42. Gögenur I , et al. Disturbances in the circadian pattern of activity and sleep after laparoscopic versus open abdominal surgery . Surg Endosc. 2009 ; 23 ( 5 ): 1026 – 1031 . Google Scholar Crossref Search ADS PubMed WorldCat 43. Redeker NS , et al. Patterns and predictors of sleep pattern disturbance after cardiac surgery . Res Nurs Health. 2004 ; 27 ( 4 ): 217 – 224 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Baker, Fiona, C;Forouzanfar,, Mohamad;Goldstone,, Aimée;Claudatos, Stephanie, A;Javitz,, Harold;Trinder,, John;, De Zambotti, Massimiliano

doi: 10.1093/sleep/zsz175pmid: 31408175

Abstract Hot flashes (HFs) are a hallmark of menopause in midlife women. They are beyond bothersome symptoms, having a profound impact on quality of life and wellbeing, and are a potential marker of cardiovascular (CV) disease risk. Here, we investigated the impact on CV functioning of single nocturnal HFs, considering whether or not they were accompanied by arousals or awakenings. We investigated changes in heart rate (HR, 542 HFs), blood pressure (BP, 261 HFs), and pre-ejection period (PEP, 168 HFs) across individual nocturnal physiological HF events in women in the menopausal transition or post-menopause (age: 50.7 ± 3.6 years) (n = 86 for HR, 45 for BP, 27 for PEP). HFs associated with arousals/awakenings (51.1%), were accompanied by an increase in systolic (SBP; ~6 mmHg) and diastolic (DBP; ~5 mmHg) BP and HR (~20% increase), sustained for several minutes. In contrast, HFs occurring in undisturbed sleep (28.6%) were accompanied by a drop in SBP and a marginal increase in HR, likely components of the heat dissipation response. All HFs were accompanied by decreased PEP, suggesting increased cardiac sympathetic activity, with a prolonged increase for HFs associated with sleep disruption. Older age predicted greater likelihood of HF-related sleep disturbance. HFs were less likely to wake a woman in rapid-eye-movement and slow-wave sleep. Findings show that HFs associated with sleep disruption, which are in the majority and more likely in older women, lead to increases in HR and BP, which could have long-term impact on nocturnal CV restoration in women with multiple HFs. vasomotor symptoms, blood pressure, REM sleep, sympathetic nervous system, menopause Statement of Significance Hot flashes (HFs) are a common symptom that can persist for several years across menopause and are a known marker of cardiovascular (CV) disease risk. They often disrupt sleep, which could impact nocturnal CV recovery. We show that waking up with a HF, beyond the HF itself, is associated with CV activation, including increased blood pressure, heart rate, and cardiac sympathetic activity, which could be a pathway towards increased CV risk in midlife women. Women with multiple nocturnal HFs that wake them up may lose the benefit of the CV “holiday” that accompanies restful sleep. Also, older age predicts HF-associated sleep disruption, suggesting that the CV impact may be more relevant in older women. Introduction Hot flashes (HFs) are a hallmark of the menopausal transition, being reported by up to 80% of women [1–3] and lasting a median duration of 7.4 years [1]. HF frequency varies between women, ranging from hourly to weekly [4], and can occur both day and night (night sweats) [4]. HFs emerge when estradiol levels decline and are a thermoregulatory phenomenon, characterized by a heat dissipation response, including increased cutaneous vasodilation and sweating that last a few minutes [5]. As such, HFs can be identified by a sudden increase in sternal skin conductivity due to increased sweating. Studies in awake women having HFs have shown that vasodilation is mediated by a transient increase in skin sympathetic nerve activity [6] and is accompanied by increased heart rate (HR) and decreased mean arterial pressure [6, 7]. Recent studies have pointed to HFs being beyond “bothersome” symptoms, showing that they are markers of higher risk of subclinical cardiovascular (CV) disease. Specifically, presence and/or severity of HFs has been associated with higher blood pressure (BP), poorer endothelial function and flow-mediated dilation, more aortic calcification, and higher carotid intima-media thickness (reviewed in [8]), with some relationships being stronger in women with an early onset of HFs [9]. Women with HFs are also more likely to have an adverse adipokine profile [10], higher lipid and lipoprotein levels [11], insulin resistance [12], and possibly increased risk for subsequent CV disease events [8]. Most studies have been cross-sectional and relied on self-reported HFs; underlying mechanisms linking HFs with markers of subclinical CV risk are not fully understood, and the HF may itself be just one indicator of underlying risk, reflecting system instability. An unexplored factor that could contribute to increased CV risk in women with multiple HFs is HF-related sleep disturbance. Sleep and the CV system are intimately connected, with sleep providing a “CV holiday” [13], with an overall wake-to-sleep reduction in CV effort (decreases in BP, HR, cardiac output). Arousal from sleep can disrupt this CV holiday [14], which is a pathway to adverse CV health [13]. Higher nocturnal BP is a strong predictor of adverse CV events in hypertensive and general populations [15, 16], and is associated with reduced endothelial function [15]. HFs are associated with poorer self-reported sleep quality and chronic insomnia [17–19], with longitudinal data showing that women reporting moderate-severe HFs are almost three times more likely to report frequent nocturnal awakenings compared to women without HFs [18]. While studies linking self-reported HFs with poor perceived sleep quality are consistent, literature linking objective HFs and objective poor sleep quality are conflicting (see [17] for review), with some studies showing a poorer polysomnographic (PSG)-defined sleep profile [20] and others showing no difference in PSG measures [21] in women with objective HFs compared with asymptomatic women. Analysis of individual HF events and PSG measures in women with HFs are also inconsistent, with some studies reporting that awakenings are more likely to occur before than after an HF [21], and others reporting that HFs occur before an awakening only in the first half of the night [22], or that the majority of HFs coincide with awakenings [23–25], with no differences in the first and second part of the night [23]. In an experimental model of simulated menopause (treatment with a Gonadotropin-releasing hormone [GnRH] agonist), Joffe and colleagues [26] found that the majority (66%) of HFs coincided with wake or N1 sleep. Some of the variability in the findings in the literature might relate to inter- and intra-individual differences in the extent to which HFs disrupt sleep. For example, a more detailed analysis of HF-related sleep disruption revealed that HF-associated wake time was responsible for almost a third of total wakefulness, on average, however there was wide variance between women (some HFs occurred without disturbing sleep, while others were associated with awakenings of different durations, depending on the time taken to fall asleep again) [23]. Awakenings associated with HFs could disrupt CV recovery across the night. However, little work has examined CV changes associated with nocturnal HFs, and none has considered effects of HF-awakenings versus the HF itself. Some studies show that women reporting HFs have higher SBP and DBP [27], with higher DBP being particularly evident at night [28] compared to asymptomatic women. Also, the presence of nocturnal HFs is associated with an increased risk of coronary heart disease [29]. Thurston and colleagues [30] considered HR variability changes around physiological HFs recorded in women across 24 hours, and found a significant reduction in vagal-related activity during an HF, with the reduction being greater for HFs recorded during self-reported sleep periods. We previously reported that HR increased and vagal-related activity decreased in association with HFs in undisturbed sleep, indicating a direct effect of the HF itself [31]. In another study of ambulatory BP in women with severe HFs, half of the HFs self-reported at night were associated with substantially higher SBP and DBP within 15 minutes after the HF relative to average nocturnal values [32]. However, sleep stages and physiological HFs were not measured, therefore, this analysis is restricted to HFs recalled by women. The relationship between HFs, sleep/wake episodes, and CV function, therefore, remains to be determined. Here we aimed to investigate changes in BP and HR recorded continuously across individual HFs, categorized according to whether or not they were associated with simultaneous arousal/awakening (Arousal HF); delayed awakening (Delayed Arousal HF); or without sleep disruption (Sleep HF). In this way, we could differentiate effects on CV functioning of the HF alone from effects of the HF combined with immediate or delayed awakenings. We also assessed pre-ejection period (PEP), derived from impedance cardiography (ICG), and specifically reflective of beta-adrenergic influences upon the heart [33]. We hypothesized that HFs associated with sleep disruption would be associated with a greater CV response compared to HFs in undisturbed sleep. We also investigated if subject characteristics (e.g. menopausal stage, age, body mass index [BMI], depression symptoms), and sleep stage and time of night in which the HF occurred, predicted the likelihood of HF-associated sleep disruption. Methods Sample Eighty-six women who had at least one objectively recorded HF during an overnight laboratory PSG recording were included in this analysis. Sample characteristics are presented in Table 1. Women were participating in a multi-night study, as described elsewhere [34, 35]. Briefly, women aged between 43–60 years old in the early (n = 46, persistent difference of 7 days or more in the length of consecutive cycles) or late (n = 31, increased variability in cycle length and the occurrence of amenorrhea of ≥60 days) menopausal transition or early post-menopause (n = 9, first 1–6 years after final menses), according to Stages of Reproductive Aging Workshop criteria [36], were recruited from the community. Inclusion criteria were: intact uterus and at least one ovary; absence of severe mental or medical conditions requiring medications known to affect sleep and/or the CV system (e.g. antihypertensives, hypnotics, antidepressants), absence of hormone therapy/contraception for the preceding 3 months, absence of sleep-related breathing/leg movement disorders (assessed with clinical PSG). All but three women were nonsmokers. The study was reviewed and approved by the SRI International Institutional Review Board, and all participants provided written, informed consent. Table 1. Characteristics and polysomnographic variables (mean [SD]) for the sample of 86 women included in the analysis Demographics  Sample, No.   •Early menopausal transition 46   •Late menopausal transition 31   •Post-menopause 9 Caucasian, No. 64 Mean (SD) Age, years 50.7 (3.6) Body mass index, kg m−2 24.4 (4.0) Beck depression inventory (BDI-II)a, score 6.6 (5.5) Pittsburgh sleep quality index (PSQI)a, score 7.4 (4.0) Pre-sleep arousal  Presleep arousal scale (PSAS), score   •PSAS-somatic 11.4 (3.1)   •PSAS-cognitive 10.2 (6.0) Polysomnographic variablesb  Time in bed, minutes 439.4 (55.3)  Sleep onset latency, minutes 11.7 (11.1)  Total sleep time (TST), minutes 368.8 (49.2)  Wakefulness after sleep onset, minutes 58.9 (36.0)  Stage N1, %TST 8.8 (4.3)  Stage N2, %TST 53.6 (7.8)  Stage N3, %TST 14.4 (7.9)  Rapid-eye-movement (REM) sleep, %TST 23.1 (4.6)  Arousal Index, No./hour 9.5 (5.3)  Awakening Index, No./hour 3.5 (1.6) Physiological hot flashes  Hot flashes (No. per night) 6.3 (6.1) (median of 5) Demographics  Sample, No.   •Early menopausal transition 46   •Late menopausal transition 31   •Post-menopause 9 Caucasian, No. 64 Mean (SD) Age, years 50.7 (3.6) Body mass index, kg m−2 24.4 (4.0) Beck depression inventory (BDI-II)a, score 6.6 (5.5) Pittsburgh sleep quality index (PSQI)a, score 7.4 (4.0) Pre-sleep arousal  Presleep arousal scale (PSAS), score   •PSAS-somatic 11.4 (3.1)   •PSAS-cognitive 10.2 (6.0) Polysomnographic variablesb  Time in bed, minutes 439.4 (55.3)  Sleep onset latency, minutes 11.7 (11.1)  Total sleep time (TST), minutes 368.8 (49.2)  Wakefulness after sleep onset, minutes 58.9 (36.0)  Stage N1, %TST 8.8 (4.3)  Stage N2, %TST 53.6 (7.8)  Stage N3, %TST 14.4 (7.9)  Rapid-eye-movement (REM) sleep, %TST 23.1 (4.6)  Arousal Index, No./hour 9.5 (5.3)  Awakening Index, No./hour 3.5 (1.6) Physiological hot flashes  Hot flashes (No. per night) 6.3 (6.1) (median of 5) aData unavailable in 1 or 2 women. bPSG measures were averaged across nights if a participant had multiple nights before averaging across the group. Open in new tab Table 1. Characteristics and polysomnographic variables (mean [SD]) for the sample of 86 women included in the analysis Demographics  Sample, No.   •Early menopausal transition 46   •Late menopausal transition 31   •Post-menopause 9 Caucasian, No. 64 Mean (SD) Age, years 50.7 (3.6) Body mass index, kg m−2 24.4 (4.0) Beck depression inventory (BDI-II)a, score 6.6 (5.5) Pittsburgh sleep quality index (PSQI)a, score 7.4 (4.0) Pre-sleep arousal  Presleep arousal scale (PSAS), score   •PSAS-somatic 11.4 (3.1)   •PSAS-cognitive 10.2 (6.0) Polysomnographic variablesb  Time in bed, minutes 439.4 (55.3)  Sleep onset latency, minutes 11.7 (11.1)  Total sleep time (TST), minutes 368.8 (49.2)  Wakefulness after sleep onset, minutes 58.9 (36.0)  Stage N1, %TST 8.8 (4.3)  Stage N2, %TST 53.6 (7.8)  Stage N3, %TST 14.4 (7.9)  Rapid-eye-movement (REM) sleep, %TST 23.1 (4.6)  Arousal Index, No./hour 9.5 (5.3)  Awakening Index, No./hour 3.5 (1.6) Physiological hot flashes  Hot flashes (No. per night) 6.3 (6.1) (median of 5) Demographics  Sample, No.   •Early menopausal transition 46   •Late menopausal transition 31   •Post-menopause 9 Caucasian, No. 64 Mean (SD) Age, years 50.7 (3.6) Body mass index, kg m−2 24.4 (4.0) Beck depression inventory (BDI-II)a, score 6.6 (5.5) Pittsburgh sleep quality index (PSQI)a, score 7.4 (4.0) Pre-sleep arousal  Presleep arousal scale (PSAS), score   •PSAS-somatic 11.4 (3.1)   •PSAS-cognitive 10.2 (6.0) Polysomnographic variablesb  Time in bed, minutes 439.4 (55.3)  Sleep onset latency, minutes 11.7 (11.1)  Total sleep time (TST), minutes 368.8 (49.2)  Wakefulness after sleep onset, minutes 58.9 (36.0)  Stage N1, %TST 8.8 (4.3)  Stage N2, %TST 53.6 (7.8)  Stage N3, %TST 14.4 (7.9)  Rapid-eye-movement (REM) sleep, %TST 23.1 (4.6)  Arousal Index, No./hour 9.5 (5.3)  Awakening Index, No./hour 3.5 (1.6) Physiological hot flashes  Hot flashes (No. per night) 6.3 (6.1) (median of 5) aData unavailable in 1 or 2 women. bPSG measures were averaged across nights if a participant had multiple nights before averaging across the group. Open in new tab Questionnaires Women completed the Beck Depression Inventory (BDI-II) [37] to assess depression symptoms; the Pittsburgh sleep quality index (PSQI) to assess sleep quality [38]; and the presleep arousal scale (PSAS), to assess cognitive and somatic arousal during the falling asleep process [39]. Physiological assessments Recordings were made using dedicated inputs to Compumedics amplifiers (Compumedics, Abbotsford, Victoria, Australia). PSG, electrocardiography (ECG), and skin conductance (SC) were recorded in all women. Beat-to-beat BP and ICG data were only available in a subset of women who completed that part of the protocol. Specifically, beat-to-beat BP data were available in 45 women (21 early menopausal transition, 15 late menopausal transition, 9 early post-menopause) and ICG data were available in 27 women (9 early menopausal transition, 9 late menopausal transition, 9 early post-menopause). There were no differences in demographics or sleep continuity measures between the groups with and without BP or ICG data. Signal processing was performed using customized processing scripts for Matlab (MathWorks, Natick, MA). Sleep. Recordings included electroencephalography (EEG; 256 Hz sampled, 0.3–35 Hz filtered), bilateral electrooculography, submental electromyography, collected according to standard criteria [40] using Compumedics amplifiers and ProFusion software (Compumedics, Abbotsford, Victoria, Australia). Sleep was scored in 30-second epochs (wake, N1, N2, N3, rapid-eye-movement [REM sleep]) by experienced scorers blinded to the presence of any HF events. Brief arousals (≥3 seconds, <15 seconds) were marked according to standard rules (abrupt shift of EEG frequency including alpha, theta and/or frequencies >16 Hz [but not spindles] that lasts ≥ 3 seconds [40]). Heart rate. ECG was recorded via Ag/AgCl Meditrace surface spot electrodes in a modified D2 Einthoven configuration, sampled at 512 Hz. Signals were digitally filtered with a fourth-order Butterworth bandpass filter (upper: 0.5 Hz; lower: 35 Hz), applied in forward and backward directions to avoid phase shifts. Customized algorithms were applied to compute normal-to-normal inter-beat-intervals via automatic detection of R peaks to derive HR. Blood pressure. BP raw waveforms were obtained using Portapres technology (Model-2; TNO TPD Biomedical Instrumentation, Amsterdam, NL), a validated method allowing prolonged noninvasive BP measurements [41]. The BP signal was obtained from the index and middle fingers of the non-dominant hand using inflatable cuffs and photoplethysmography (PPG) sensors to measure the blood volume. The blood volume at zero transmural pressure is estimated via oscillometry, and the cuff pressure is continually varied to maintain this blood volume throughout the cardiac cycle via a fast servo-control system. The applied time-varying cuff pressure is, therefore, equal to arterial BP. The device has been shown to achieve BP errors within Association for the Advancement of Medical Instrumentation limits of 5 mmHg bias and 8 mmHg precision [42]. Despite some limitation in absolute BP estimates, this method allows reliable estimation of BP changes [43], the focus of our investigation. Measurements alternated between fingers every 30 minutes to minimize discomfort. Automatic algorithms were developed to identify and exclude data corresponding to calibration and cuff switching and to detect systolic peaks and diastolic troughs, from which SBP and DBP were derived. Sympathetic activity. ICG dZ/dt (Ω), that is, the rate of change in the impedance waveform on a given beat, was recorded with HIC-4000 Bioelectric ICG (Bio-Impedance Technology, Inc., Chapel Hill, NC), using a dual-spot electrode 4-lead 8-point connection arrangement [44]. A low-intensity (4 mA) alternating current of high frequency (100 kHz) was used for the ICG circuit (sampling rate: 1024 Hz). Signals were digitally filtered with a fourth-order Butterworth bandpass filter (upper: 0.5 Hz; lower: 25 Hz), applied in forward and backward directions. ICG cardiac cycles were identified using the ECG R peaks as reference. Automatic algorithms were developed to detect and remove ICG cycles corrupted by noise and artifacts [45, 46]. B points were then detected on the clean ICG cycles using an automatic B-point detection algorithm [47]. The algorithms used to remove artifact from ICG and ECG signals and to measure PEP have been shown to achieve up to 96% accuracy when compared to expert manual scoring [45–47]. PEP(s) was calculated as the time interval between the Q-point (beginning of the electrical systole; determined by a fixed interval [35 ms] backward from the ECG R-wave peak) on the ECG signal and the B point (opening of the aortic valve) on the ICG dZ/dt signal. PEP is inversely related to beta-adrenergic sympathetic nervous system activity [33, 48]. Post-processing of all CV measures An outlier removal algorithm was developed to remove corrupted cardiac cycles (data points > 10 scaled median absolute deviations from the median). Corrupted data segments <10 seconds were replaced by interpolating the remainder of the data. Uncorrupted CV parameters were then averaged over 30-second bins corresponding to the scored PSG. Thirty-second data segments containing more than 70% corrupted cardiac cycles were identified as corrupted. Corrupted segments of data that were shorter than 300 seconds were replaced by interpolating the remainder of the data; the rest of them were excluded. An expert visually checked the performance of the outlier detection algorithm on approximately 10% of randomly selected data. Hot flashes. SC was sampled at 16 Hz using a BioDerm Skin Conductance Meter (model 2701, UFI) connected to two Ag/AgCl electrodes placed on either side of the sternum, with a 0.5 V constant voltage circuit between them. Physiological HFs were automatically detected based on a ≥1.5 µmho rise in SC within 30 seconds, using a customized algorithm [49] with the threshold set at 1.5 µmho. Algorithms were used to detect HFs and determine the timing of their onset, as well as to reject noise and artifacts by analyzing the SC level and slope [49]. HF candidates accompanied by a very fast change in SC (not physiologically possible) were rejected, and a 10-minute refractory period following an HF was applied during which any HF candidates are removed. To detect HF-onset, the SC signal was filtered, the slope within a 30-second window following the HF was calculated, and the first sample at which the slope reached at least 0.02 μS s−1 was marked as HF-onset [49]. All HFs were then manually checked to examine the shape of each HF: submaximal HFs (lower than the 2 µmho threshold [50]), but with a characteristic HF shape were included, as suggested by others [30, 51, 52]. Indeed, the 2 µmho threshold has poor sensitivity (up to 47% of HFs reported by women did not correspond to a sufficient SC increase) [53], and lowering the threshold increases sensitivity while maintaining specificity [51]. HF categories Given the known increase in skin blood flow even before an increase in sweat rate [4], we marked a 60-second window as “HF-onset,” which encompassed the start of the increase in sweat rate and the 30 seconds immediately prior to that. We then categorized HFs based on the presence of PSG-defined wake time and sleep stage composition and brief arousals during the 4-minute period (Figure 1) encompassing HF-onset, using modified criteria from [31]. The following categories were formed: Figure 1. Open in new tabDownload slide HFs were categorized based on sleep stage composition in 30-second epochs around HF-onset (60 seconds) and across 90-second pre-HF-onset and 90-second post-HF-onset. Bottom panel shows an example of a physiological HF (skin conductance rise of ≥1.5 µmho within 30 seconds) recorded from sternum skin conductance time-aligned with the PSG sleep stages; arrow reflects rise in skin conductance marking HF-onset. The complete analysis window is divided into baseline, HF, early, and late recovery periods. N1, N2, N3, refer to sleep stages; W = wakefulness. Figure 1. Open in new tabDownload slide HFs were categorized based on sleep stage composition in 30-second epochs around HF-onset (60 seconds) and across 90-second pre-HF-onset and 90-second post-HF-onset. Bottom panel shows an example of a physiological HF (skin conductance rise of ≥1.5 µmho within 30 seconds) recorded from sternum skin conductance time-aligned with the PSG sleep stages; arrow reflects rise in skin conductance marking HF-onset. The complete analysis window is divided into baseline, HF, early, and late recovery periods. N1, N2, N3, refer to sleep stages; W = wakefulness. Arousal HF when the 90-second pre-HF-onset period consisted of sleep (N1, N2, N3, or REM sleep) and the HF-onset window contained at least 1 epoch of wake or arousal. The 90-second post-HF-onset period consisted of wake and/or sleep. Delayed Arousal HF when wakefulness did not occur until the post-HF-onset period. Sleep HF when the 90-second pre-HF-onset period, HF-onset, and 90-second post-HF-onset period contained only sleep epochs. There were no brief arousals during HF-onset. Ambiguous HF when the pre-HF-onset period contained all wake or mixed wake/sleep epochs. (Excluded from analysis of CV changes across HF events due to variability in CV measures in the pre-HF-onset period). We used stringent criteria for defining sleep disruption during HF-onset by considering the presence of even brief arousals. Brief arousals were allowed to occur during the rest of the period, regardless of HF category. Data analysis The period of analysis was the 15-minute period around each HF, divided into 30-second epochs (~5 minutes before, and ~ 10 minutes after the HF, Figure 1). Statistical analyses were conducted using Hierarchical linear models (type of repeated-measures mixed model) to account for between-woman and between-night random effects. Repeated-measure models take into account the fact that there are multiple measurements on study participants so that precision is not overestimated (as would occur if individual HFs were treated as independent). Hierarchical linear models using Stata’s mixed commands were used to examine whether values for each CV variable across 30-second epochs of the pre-HF-onset period (90 seconds before HF-onset), HF-onset (60 seconds), and post-HF-onset period (90 seconds after HF-onset) as well as for early (between 90 seconds and 270 seconds after HF-onset) and late (between 300 seconds and 570 seconds after HF-onset) recovery periods differed from baseline (between 90 seconds and 270 seconds before HF-onset) and whether the magnitude of that difference differed between HF types (i.e. Arousal HF, Delayed Arousal HF, Sleep HF). The independent variable was a categorical variable for HF type, where the excluded category (represented by the constant term in the regression equation) was Sleep HF. Statistical significance was calculated using Wald’s test. For tests comparing HF categories, the Wald test compared the expected value for an HF category with that of another category. For analyses examining differences across the HF relative to baseline, the Wald’s test determined whether the difference between baseline and each 30-second epoch was statistically greater than zero for each HF category. HFs with missing CV data were excluded from analysis for that particular model. Repeated-measures mixed ordinal logistic models using Stata’s meologit commands were used for analyses determining whether the proportions of each sleep stage in the pre-HF-onset period differed across HF categories and whether proportions of each HF category differed across hours of the night. Finally, mixed logistic models using Stata’s melogit commands were used to determine if characteristics (menopausal stage; age; BMI; sleep quality from PSQI; depression symptoms from BDI; presleep cognitive and somatic arousal from PSAS) predicted the likelihood of having HF-associated sleep disturbance (combined Arousal and Delayed Arousal HF categories). Results Hot flashes Women contributed 1–13 HFs per night, from 1 to 7 PSG recordings, for a total of 546 physiological HFs detected from 175 nights. A small number of HFs were excluded from analysis for HR (n = 4), BP (n = 10), and ICG (n = 20) due to signal drop-outs (e.g. due to sensor loosening/detachment) around the HF-onset period. Due to differences in the number of women providing data for each signal, and missing data, number of HF events analyzed varied per signal: HR analysis was based on 542 HFs; BP analysis on 261 HFs; and PEP analysis on 168 HFs. PSG variables are presented in Table 1. 51.1% of detected HFs were clearly associated with sleep disruption (27.7% for Arousal HF and 23.4% for Delayed Arousal HF categories). For Arousal HF, the majority (67.6%) were associated with an awakening rather than brief arousal at HF-onset. Of those associated with an arousal, the majority (67.3%) converted to an awakening in the following 90 seconds (post-HF-onset). 28.6% of HFs occurred in undisturbed sleep (Sleep HF category). The remaining HFs (20.3%) were ambiguous (the pre-HF period contained wake or mixed wake-sleep stages such that CV measures were variable before HF-onset) and were excluded from analysis. Figure 2 shows sleep stage composition across the window of analysis, showing the distinction between HF categories. Figure 2. Open in new tabDownload slide Stacked area graphs illustrating sleep stage composition in each 30-second bin (percentage of total) as a function of time across HF-onset for HF categories (Sleep HF, Arousal HF, Delayed Arousal HF). Outer white dotted lines highlight HF characterization windows of pre-HF-onset, HF-onset, and post-HF-onset). Figure 2. Open in new tabDownload slide Stacked area graphs illustrating sleep stage composition in each 30-second bin (percentage of total) as a function of time across HF-onset for HF categories (Sleep HF, Arousal HF, Delayed Arousal HF). Outer white dotted lines highlight HF characterization windows of pre-HF-onset, HF-onset, and post-HF-onset). HR across HFs Figure 3 shows HR plotted across each HF category. While HR appears to be higher during baseline for Sleep HF relative to other categories, there was no significant difference in baseline HR, after accounting for between-women and night-to-night variability. Comparisons of HR in pre-HF-onset, HF-onset, and post-HF-onset periods relative to baseline for each HF category are shown in Table 2. For Arousal HF, HR began to increase relative to baseline before HF-onset (p < 0.001), and peaked during HF-onset and 30 seconds after (+12.0 bpm, on average). For Delayed Arousal HF, HR was significantly higher during HF-onset relative to baseline (p < 0.001), peaking 60 seconds post-HF-onset (+11.9 bpm, on average). For Sleep HF, HR began to increase 30 seconds before HF-onset relative to baseline (p = 0.01), peaking between 60–90 seconds post-HF-onset (+2.9 bpm, on average). Figure 3. Open in new tabDownload slide Heart rate (unadjusted mean ± standard error [SEM]) across HFs according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Analysis adjusted for between-women and between-night variability showed no significant differences between categories in baseline HR. Figure 3. Open in new tabDownload slide Heart rate (unadjusted mean ± standard error [SEM]) across HFs according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Analysis adjusted for between-women and between-night variability showed no significant differences between categories in baseline HR. The magnitude of the HR increase during HF-onset was greater for Arousal HF than for Sleep HF (χ 2 = 158.7, p < 0.001) and Delayed Arousal HF (χ 2 = 130.0, p < 0.001). The magnitude of the HR increase after HF-onset was greater for Delayed Arousal HF compared to Sleep HF (χ 2 = 99.7, p < 0.001; Figure 3). Finally, HR remained significantly higher in the early recovery period compared to baseline (p < 0.001) for all HF types (Arousal HF χ 2 = 42.7; Delayed Arousal HF χ 2 = 49.1; Sleep HF χ 2 = 41.8). HR continued to be higher in the late recovery period relative to baseline for Sleep HF (χ 2 = 11.4, p < 0.01) and Arousal HF (χ 2 = 9.6, p < 0.01) but not for Delayed Arousal HF. Table 2. Results of statistical comparisons of changes in heart rate, systolic and diastolic blood pressure, and pre-ejection period across pre-HF-onset, HF-onset, and post-HF-onset, relative to baseline for three different categories of HFs recorded during overnight sleep (Arousal HF, Delayed Arousal HF, Sleep HF) HF category Heart rate Systolic blood pressure Diastolic blood pressure Pre-ejection perioda Relative to baseline Pre-HF- onset HF-onset Post-HF- onset Pre-HF-onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post-HF-onset Arousal HF ↑ (χ 2 = 14.1, p < 0.001) ↑ (χ 2 = 120, p < 0.001) ↑ (χ 2 = 128, p < 0.001) ns ns ↑ (χ 2 = 4.5, p = 0.03) ns ↑ (χ 2 = 4.5, p = 0.03) ↑ (χ 2 = 19.6, p < 0.001) ns ↓ (χ 2 = 16.4, p < 0.001) ↓ (χ 2 = 18.7 p < 0.001) Delayed Arousal HF ns ↑ (χ 2 = 26.3 p < 0.001) ↑ (χ 2 = 251, p < 0.001) ns ↓ (χ 2 = 5.5, p = 0.02) ↑ (χ 2 = 3.9, p = 0.047) ns ↓ (χ 2 = 5.6, p = 0.02) ↑ (χ 2 = 7.9, p < 0.01) ns ↓ (trend) p = 0.07 ↓ (χ 2 = 11.7, p < 0.001) Sleep HF ↑ (χ 2 = 6.6, p = 0.01) ↑ (χ 2 = 8.9, p < 0.01) ↑ (χ 2 = 24, p < 0.001) ns ↓ (χ 2 = 7.6, p < 0.01 ↓ (χ 2 = 4.8, p = 0.03) ns ns ns ns ↓ (χ 2 = 5.7, p = 0.02) ↓ (χ 2 = 4.9, p = 0.03) HF category Heart rate Systolic blood pressure Diastolic blood pressure Pre-ejection perioda Relative to baseline Pre-HF- onset HF-onset Post-HF- onset Pre-HF-onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post-HF-onset Arousal HF ↑ (χ 2 = 14.1, p < 0.001) ↑ (χ 2 = 120, p < 0.001) ↑ (χ 2 = 128, p < 0.001) ns ns ↑ (χ 2 = 4.5, p = 0.03) ns ↑ (χ 2 = 4.5, p = 0.03) ↑ (χ 2 = 19.6, p < 0.001) ns ↓ (χ 2 = 16.4, p < 0.001) ↓ (χ 2 = 18.7 p < 0.001) Delayed Arousal HF ns ↑ (χ 2 = 26.3 p < 0.001) ↑ (χ 2 = 251, p < 0.001) ns ↓ (χ 2 = 5.5, p = 0.02) ↑ (χ 2 = 3.9, p = 0.047) ns ↓ (χ 2 = 5.6, p = 0.02) ↑ (χ 2 = 7.9, p < 0.01) ns ↓ (trend) p = 0.07 ↓ (χ 2 = 11.7, p < 0.001) Sleep HF ↑ (χ 2 = 6.6, p = 0.01) ↑ (χ 2 = 8.9, p < 0.01) ↑ (χ 2 = 24, p < 0.001) ns ↓ (χ 2 = 7.6, p < 0.01 ↓ (χ 2 = 4.8, p = 0.03) ns ns ns ns ↓ (χ 2 = 5.7, p = 0.02) ↓ (χ 2 = 4.9, p = 0.03) ns, not significant. aDecreased PEP reflects increased sympathetic activity. Open in new tab Table 2. Results of statistical comparisons of changes in heart rate, systolic and diastolic blood pressure, and pre-ejection period across pre-HF-onset, HF-onset, and post-HF-onset, relative to baseline for three different categories of HFs recorded during overnight sleep (Arousal HF, Delayed Arousal HF, Sleep HF) HF category Heart rate Systolic blood pressure Diastolic blood pressure Pre-ejection perioda Relative to baseline Pre-HF- onset HF-onset Post-HF- onset Pre-HF-onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post-HF-onset Arousal HF ↑ (χ 2 = 14.1, p < 0.001) ↑ (χ 2 = 120, p < 0.001) ↑ (χ 2 = 128, p < 0.001) ns ns ↑ (χ 2 = 4.5, p = 0.03) ns ↑ (χ 2 = 4.5, p = 0.03) ↑ (χ 2 = 19.6, p < 0.001) ns ↓ (χ 2 = 16.4, p < 0.001) ↓ (χ 2 = 18.7 p < 0.001) Delayed Arousal HF ns ↑ (χ 2 = 26.3 p < 0.001) ↑ (χ 2 = 251, p < 0.001) ns ↓ (χ 2 = 5.5, p = 0.02) ↑ (χ 2 = 3.9, p = 0.047) ns ↓ (χ 2 = 5.6, p = 0.02) ↑ (χ 2 = 7.9, p < 0.01) ns ↓ (trend) p = 0.07 ↓ (χ 2 = 11.7, p < 0.001) Sleep HF ↑ (χ 2 = 6.6, p = 0.01) ↑ (χ 2 = 8.9, p < 0.01) ↑ (χ 2 = 24, p < 0.001) ns ↓ (χ 2 = 7.6, p < 0.01 ↓ (χ 2 = 4.8, p = 0.03) ns ns ns ns ↓ (χ 2 = 5.7, p = 0.02) ↓ (χ 2 = 4.9, p = 0.03) HF category Heart rate Systolic blood pressure Diastolic blood pressure Pre-ejection perioda Relative to baseline Pre-HF- onset HF-onset Post-HF- onset Pre-HF-onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post- HF- onset Pre-HF- onset HF-onset Post-HF-onset Arousal HF ↑ (χ 2 = 14.1, p < 0.001) ↑ (χ 2 = 120, p < 0.001) ↑ (χ 2 = 128, p < 0.001) ns ns ↑ (χ 2 = 4.5, p = 0.03) ns ↑ (χ 2 = 4.5, p = 0.03) ↑ (χ 2 = 19.6, p < 0.001) ns ↓ (χ 2 = 16.4, p < 0.001) ↓ (χ 2 = 18.7 p < 0.001) Delayed Arousal HF ns ↑ (χ 2 = 26.3 p < 0.001) ↑ (χ 2 = 251, p < 0.001) ns ↓ (χ 2 = 5.5, p = 0.02) ↑ (χ 2 = 3.9, p = 0.047) ns ↓ (χ 2 = 5.6, p = 0.02) ↑ (χ 2 = 7.9, p < 0.01) ns ↓ (trend) p = 0.07 ↓ (χ 2 = 11.7, p < 0.001) Sleep HF ↑ (χ 2 = 6.6, p = 0.01) ↑ (χ 2 = 8.9, p < 0.01) ↑ (χ 2 = 24, p < 0.001) ns ↓ (χ 2 = 7.6, p < 0.01 ↓ (χ 2 = 4.8, p = 0.03) ns ns ns ns ↓ (χ 2 = 5.7, p = 0.02) ↓ (χ 2 = 4.9, p = 0.03) ns, not significant. aDecreased PEP reflects increased sympathetic activity. Open in new tab SBP and DBP Across HFs. There was no significant difference in baseline levels of SBP or DBP between HF categories. As shown in Figure 4 and Table 2, for Arousal HF, SBP was higher relative to baseline post-HF-onset (p = 0.03), with the maximum increase occurring 90 seconds after HF-onset (+6.7 mmHg, on average). As shown in Table 2, for Delayed Arousal HF, SBP was significantly lower during HF-onset relative to baseline (−8.3 mmHg, on average) before it began to rise again, being significantly higher than baseline 90 seconds post-HF-onset (p = 0.047, +3.9 mmHg, on average). For Sleep HF, SBP was lower during HF-onset (p < 0.01, −5.4 mmHg, on average) relative to baseline. Figure 4. Open in new tabDownload slide Systolic blood pressure (unadjusted mean ± standard error [SEM]) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Figure 4. Open in new tabDownload slide Systolic blood pressure (unadjusted mean ± standard error [SEM]) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Changes in DBP were similar to those in SBP across HFs (Figure 5 and Table 2). For Arousal HF, DBP was higher relative to baseline during HF-onset (p = 0.03), with the maximum increase 90 seconds after HF-onset (+5.4 mmHg, on average). For Delayed Arousal HF, DBP was lower during HF-onset relative to baseline (p = 0.02) and then began to rise after HF-onset, being higher than baseline 60 seconds after HF-onset (p < 0.01). DBP did not change across the HF relative to baseline for Sleep HF. Figure 5. Open in new tabDownload slide Diastolic blood pressure (unadjusted mean ± standard error [SEM]) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Figure 5. Open in new tabDownload slide Diastolic blood pressure (unadjusted mean ± standard error [SEM]) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Between HF analyses showed that the magnitude of the increase in SBP after HF-onset was greater for Arousal HF (χ 2 = 11.3, p < 0.001) and Delayed Arousal HF (χ 2 = 6.1, p = 0.01) across the post-HF-onset period, relative to Sleep HF. Similarly, the increase in DBP after HF-onset was greater for Arousal HF (χ 2 = 7.6, p < 0.01) and tended to be greater for Delayed Arousal HF (χ 2 = 3.1, p = 0.08), relative to Sleep HF. There were no significant differences between baseline and recovery periods for SBP for any of the HF categories. For DBP, in early recovery, levels remained higher than baseline for Arousal HF (χ 2 = 3.9, p = 0.047) but returned to baseline levels in late recovery. For Delayed Arousal HF and Sleep HF, DBP no longer differed from baseline during recovery periods. PEP across HFs. As shown in Figure 6 and Table 2, for Arousal HF, PEP significantly shortened (increase in sympathetic activity) during HF-onset relative to baseline (p < 0.001), and continued to be shorter 30–90 seconds post-HF-onset (p < 0.001). For Delayed Arousal HF, PEP was significantly shorter 30–90 seconds post-HF-onset relative to baseline (p < 0.001). Finally, for Sleep HF, PEP was significantly shorter during HF-onset (p = 0.02) and 30 seconds post-HF-onset (p = 0.03) relative to baseline. There was no difference in the magnitude of PEP shortening (increase in sympathetic activity) across HF categories. Finally, PEP levels in early recovery were still lower than baseline for both Arousal HF (χ 2 = 6.9, p < 0.01) and Delayed Arousal HF (χ 2 = 7.1, p < 0.01) but returned to baseline levels during the late recovery period. For Sleep HF, PEP during recovery had returned to baseline levels. Figure 6. Open in new tabDownload slide Pre-ejection period (unadjusted mean ± standard error [SEM], an index inversely related to cardiac sympathetic nervous system activity; lower values = greater sympathetic activity) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Figure 6. Open in new tabDownload slide Pre-ejection period (unadjusted mean ± standard error [SEM], an index inversely related to cardiac sympathetic nervous system activity; lower values = greater sympathetic activity) across hot flashes according to category (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash). Factors predicting HF-associated arousal Sleep stage composition in the pre-HF period differed between HF categories: The pre-HF period for Sleep HFs was less likely to contain N2 sleep (48% vs 70%) and more likely to contain REM sleep (24% vs 6%) than Arousal and Delayed Arousal HFs (p < 0.01, Figure 2). The pre-HF-onset period for Sleep HF was also less likely than Arousal HF to contain N1 sleep (3% vs 7%, p = 0.036) and more likely to contain N3 sleep (26% vs 15%, p = 0.018). Sleep stage composition in the pre-HF-onset period did not differ between Arousal and Delayed Arousal HFs. The distribution of HF categories across hours of the night is shown in Supplementary Figure 1. Of the factors, we considered as potential predictors of HF-associated sleep disturbance (Arousal and Delayed Arousal HF events, combined) age (χ 2 = 3.9, p = 0.047) and BMI (χ 2 = 4.53, p = 0.03) were significant: older age and higher BMI were associated with a greater likelihood of having HF-associated sleep disturbance. Follow-up analysis with age and BMI in the same model showed that each had an independent effect and that there was no age × BMI interaction effect for predicting HF-associated sleep disturbance. Discussion We show here that HFs accompanied by arousal from sleep are associated with tachycardia and an increase in BP that persists for several minutes. This HF phenotype is the most disruptive and predominant type of HF occurring at night. In contrast, when HFs occur in undisturbed sleep, HR increases to a lesser extent, and BP drops at HF-onset, likely reflecting decreased total peripheral resistance, a component of the heat dissipation response. Together, these findings show that HF-associated sleep disturbance is linked with substantial CV activation over and above the effect of the HF itself. Our finding of a pronounced increase in HR and BP in association with HF-associated arousal from sleep compliments the large body of literature showing CV activation at spontaneous arousals as well as auditory- and respiratory-evoked arousals [13]. For example, Trinder and colleagues [54] report an increase in HR averaged over 20 beats of 4.1 beats per minute, accompanied by an increase in SBP averaged over 20 beats of 6.4 mmHg following an arousal to a nonthreatening auditory stimulus in young participants. An arousal from sleep evokes an increase in peripheral and cardiac sympathetic activity, which increases vasoconstriction and HR, and leads to a surge in BP [55–57]. Similarly, we found a combination of increased BP, likely reflecting peripheral vasoconstriction, and reduced PEP, despite increased BP, suggesting both peripheral and central sympathetic activation across HF-arousal events. Arousals from sleep are a normal component of the night, but if frequent, they can dampen the reduction in CV activity that typically occurs during sleep [13]. Carrington and Trinder found that experimental arousals triggered by auditory tones (~every 1–2 minutes) across the first 90 minutes blunted the nocturnal BP dip [14]. Further, work in patients with obstructive sleep apnea has shown that a higher arousal index, beyond respiratory-related disturbances, is associated with higher diurnal and 24-hour SBP [58] and higher peripheral sympathetic activity during wakefulness in addition to increased sympathetic activity during sleep [57, 59]. HFs, and thus HF-associated awakenings, occur at a much lower frequency than do spontaneous brief arousals in healthy individuals; women had between 1–13 HFs per night in our study. However, arousal duration is an important factor to consider over and above frequency. The CV activation response at an arousal appears to have two components, the first comprising a transient “reflex” surge in BP and HR, peaking within 3–6 seconds of the stimulus, and with values returning to baseline within 10 seconds [13]. The second component depends on arousal duration; CV measures remain elevated as wakefulness persists, only returning to baseline when sleep returns [13, 54]. We found that most HF-arousal events were associated with awakenings (>15 seconds duration) rather than brief arousals and even when HF-onset was accompanied by a brief arousal, it was usually followed by a full awakening. Awakening with a HF can last from several minutes to more than an hour post HF, with HFs being responsible for, on average, almost a third of total wake time (ranging from 0% to more than 70% of total wakefulness) [23]. As such, a wake-associated increase in HR and BP following a HF could be sustained for several minutes to more than an hour post-HF. The return to sleep could depend on several factors such as discomfort due to sweating, cognitive arousal, and amount of homeostatic sleep pressure, with potentially varying effects between women on nocturnal CV recovery. Similar to our findings, others [32] reported a marked rise in BP within 15 minutes of severe nocturnal self-reported HFs, which our current data suggest were likely related to HF-associated awakenings. Here, we focused on CV changes around individual nocturnal HFs, and it remains an open question whether having multiple HF-wake events across the night over several years impacts the CV system. We previously found that the nocturnal BP profile was altered in women in the menopausal transition with insomnia disorder, who were more likely to have physiological nocturnal HFs than controls [60]. However, having HFs (yes/no) was not a significant factor in any models, although sample size was small and it is becoming increasingly apparent that HF characteristics beyond frequency are important, including amount of HF-associated wakefulness, age when HFs emerge, and HF trajectory patterns across the menopausal transition [9, 17, 23]. As we showed previously [23], a minority of HFs occurred without any awakenings. These HFs allowed us to examine CV changes associated with HFs independent of superimposed arousal effects. HR was higher 30 seconds before the rise in SC and remained higher than baseline for the duration of the HF, which supports our previous findings of higher HR during nocturnal HFs [31, 49], and that of others for HFs recorded when women are awake [4, 7, 30, 61]. Increased HR during HF-onset was accompanied by a shorter PEP, which returned to baseline levels soon after HF-onset. We showed previously that there is also a decline in cardiac vagal activity during Sleep HFs [31]. Taken together, these findings suggest that increased HR during a HF may be mediated by a combination of cardiac vagal withdrawal and sympathetic activation. We also found a decline in SBP and DBP across HF-onset, which likely reflects a decrease in total peripheral resistance due to the substantial increase in blood flow, particularly to the cutaneous vessels to dissipate heat, and which is apparent even in advance of a rise in SC [4, 6]. This increase in skin blood flow is neurally mediated via sympathetic cholinergic nerves that are responsible for active vasodilation [6]. Similar to our findings of a drop in SBP and DBP during HFs in asleep women, Low and colleagues [6, 7] found a decline in MAP, of about 9 mmHg during HFs in awake women, although not all women showed this response. As speculated by those authors [7], the drop in BP probably triggers a baroreflex response to increase HR (beyond the initial HR increase when an HF is triggered) such that HR continues to be higher during recovery periods, but BP returns to baseline. The CV response to a HF in undisturbed sleep, therefore, represents a normal thermoregulatory response and appears to be similar to responses to HFs while awake. In our study, we distinguished a sub-category of HF-disturbed sleep events: HFs in which waking lagged HF-onset. Examination of CV measures around these HFs shows components of the other two HF categories: an initial change associated with HF-onset (increase in HR and drop in BP, reflecting a heat dissipation response), and the subsequent changes associated with an awakening (substantial increase in HR and BP). It is unknown why an arousal/awakening coincides with HF-onset (defined as a sudden rise in SC) for some HFs, but is delayed or even absent in others. Possibly, the graded waking response across HFs reflects differences in intensity of the HF trigger, with Arousal HFs having the most intense trigger. Arousal threshold to various sensory stimuli is influenced by the intensity of sensory stimulation, with the percentage of experimentally induced arousals increasing with stimulus intensity [62]. Further work is needed to determine what combination of physiological measures could best be used to measure intensity/severity of an HF. While the mechanism of an HF it not completely understood, evidence from a series of studies by Freedman and colleagues strongly supports involvement of the central sympathetic nervous system in the initiation of HFs [5]. Clonidine, an α 2-adrenergic agonist, reduces central noradrenergic activation and increases the sweating threshold, thus reducing HFs [63]. On the other hand, Yohimbine treatment, which increases brain norepinephrine, increases the number of HFs [63]. With sufficient increase in central sympathetic activation, an arousal could ensue coincident with HF-onset. For HFs with a delayed awakening, the initial stimulus may not be great enough to initiate an arousal. However, the cascade of changes that characterize a HF (sweating, change in skin blood flow and temperature, drop in BP) may trigger an awakening. Other factors that determine likelihood of an arousal from sleep include sleep stage and sleep depth, stimulus modality, and individual characteristics (like age) [62, 64]. Sleep HFs were more likely to arise in N3 sleep. N3 (slow-wave sleep) is associated with a higher arousal threshold, at least to auditory stimuli [62, 65]. Possibly, HF-related arousal is also less likely in N3 sleep. Sleep HFs were also more likely to arise in REM sleep, which could be due to lower awakening responsiveness in REM sleep [62, 66] and/or the unique thermoregulatory nature of the HF stimulus: HFs are less likely in REM than NREM sleep [22, 23], possibly due to the lower sensitivity of the thermoregulatory system, and associated decrease in sweating responses, during REM sleep [67]. Any HFs that do occur in REM sleep, therefore, may be less intense (i.e. less likely to be associated with EEG arousal) than those in other sleep stages. We found that older age was associated with greater likelihood of having HF-associated sleep disturbance. Sleep is more vulnerable to disruption in older adults, with auditory awakening thresholds declining across adulthood [64]. Similarly, sleep might be more vulnerable to HF-related arousal in older women. Alternatively, greater exposure to HFs over time as a woman moves across the menopausal transition may lead to greater sensitization to HF-associated arousal; or HFs may be more severe in older women. We cannot differentiate these factors in our dataset. Menopausal status did not predict having HF-wake events; however, our sample mainly consisted of women in the menopausal transition. A higher BMI was also associated with a greater likelihood of having HF-associated sleep disturbance, however, since we had a cutoff for BMI of 32 Kg m−2, further research is needed to determine the nature of this relationship, including whether body weight or specific aspects of body composition (e.g. percentage body fat) may account for these findings. There is a growing body of research investigating whether HFs are associated with CV disease risk, with most studies relying on self-reports of HF frequency, severity, and bother, and some distinguishing relationships between diurnal HFs versus night sweats (reviewed in [8, 68, 69]). Findings for HF-BP relationships are mixed, with some showing relationships between HFs and higher BP [70, 71] and others showing no relationship between HFs and BP [32, 72]. Recent work by Thurston and colleagues has investigated the association between self-reported and/or physiological HFs (day and night) and preclinical CV risk markers [8]. Higher HF frequency, particularly during the day, was associated with greater carotid intima-media thickness and plaque [9]. Further work is still needed to determine whether HFs themselves provoke unfavorable changes in the vasculature, either directly or indirectly, or whether an underlying vulnerability, such as in estrogen-sympathetic nervous system dynamics, underlies both HFs and an unfavorable CV status. In this context, our data show that nocturnal HFs, whether or not linked with an arousal, are associated with increased cardiac sympathetic activation, although activation appears to be more sustained when an awakening coincides with an HF. When associated with an arousal/awakening, there is also an increase in BP, which could have an additional detrimental impact in women with frequent HFs persisting over several years. HF-associated awake time is a critical contributor to total wakefulness across the night [23] such that just a few HF events could have a sustained disruptive effect on sleep, and consequently CV activity. Our results should be considered in context of the study limitations. We used a convenience sample of women recruited from the community, mostly in the menopausal transition, who had at least one physiological HF during the nocturnal recording. We did not include any women seeking treatment for their symptoms. Another limitation is that we used an indirect (although validated [33, 48]) measure of cardiac sympathetic nervous system activity, derived from ICG. While reliability of absolute measures of PEP with ICG, and BP with Portapres have been questioned, they are valid at detecting changes across an event [55], such as HFs, and have the advantage of not disturbing sleep. Also, PEP and BP measures are sensitive to motion artifact; we, therefore, applied algorithms to identify and remove sections of compromised data and also manually checked the data to ensure the signals were reliable. We also used a validated algorithm to identify the time of the opening of the aortic valve [47]. We carefully categorized HFs to explore CV physiology around HF-onset depending on their association with arousals/awakenings. However, this approach meant we excluded ~20% of HFs that were ambiguous due to wake or mixed wake/sleep composition in the approach to the HF, and consequently with unstable baseline CV measures. Our conclusions, therefore, may not extend to all nocturnal HFs. Further work may examine HFs more continuously or categorize HFs in different ways, and investigate other individual or group differences within HF categories. Future analyses also could investigate further the CV responses to HFs, considering aspects of the SC signal not included here, such as HF end-points (to enable determination of HF duration), peak amplitude, and rate of change in SC. These measures could indicate physiological severity of the HF, although magnitude of the rise in SC does not necessarily correspond to subjective HF intensity or distress [53], and it is possible that a combination of physiological measures may better reflect HF severity. Future work is also needed to investigate the additive effect of multiple HFs across the night, taking into account their association with wakefulness. Ultimately, interventions that manipulate occurrence of HFs may unveil the impact of HFs on nocturnal CV restoration, extending our results here focused on transient CV activation in association with single HF events. Also, beat-to-beat analysis of HR and BP in association with more refined EEG and SC signal analyses is necessary to determine the temporal dynamics and interactions between physiological signals across a HF. A better characterization of HFs may further advance understanding of their physiological mechanisms, severity, and impact and may ultimately lead to the determination of distinct CV risk profiles in women with HFs. In conclusion, we have shown the different patterns of CV changes across nocturnal HFs, depending on their association with or without arousal from sleep. When HFs are associated with sleep disturbance, BP shows a sustained increase, which could potentially dampen nocturnal CV recovery in women with multiple HF-wake events, ultimately increasing risk for CV disease. Supplementary material Supplementary material is available at SLEEP online. Supplementary Figure 1. Stacked area graph representing the proportions of hot flash categories (Arousal Hot Flash, Delayed Arousal Hot Flash, Sleep Hot Flash, Ambiguous Hot Flash) within each hour of the night. The proportion of Sleep HFs was higher in hour-2 compared to hour-7 (χ 2= 4.5, p=0.03). The proportion of Arousal HFs was lower in hour-2 than hour-5 (χ 2 = 4.5, p=0.03). Funding The study is supported by the National Institutes of Health (HL103688, F.C.B.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Acknowledgments We thank the women who participated in this study. Conflict of interest statement. Authors declare no conflict of interest. M.d.Z. and F.C.B. have received research funding unrelated to this work from Ebb Therapeutics Inc., Fitbit Inc., and International Flavors & Fragrances Inc. References 1. Avis NE , et al. ; Study of Women’s Health Across the Nation . Duration of menopausal vasomotor symptoms over the menopause transition . JAMA Intern Med. 2015 ; 175 ( 4 ): 531 – 539 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Gold EB , et al. Longitudinal analysis of the association between vasomotor symptoms and race/ethnicity across the menopausal transition: study of women’s health across the nation . Am J Public Health. 2006 ; 96 ( 7 ): 1226 – 1235 . Google Scholar Crossref Search ADS PubMed WorldCat 3. Woods NF , et al. Anxiety, hormonal changes, and vasomotor symptoms during the menopause transition . Menopause. 2005 ; 12 ( 3 ): 242 – 245 . Google Scholar Crossref Search ADS PubMed WorldCat 4. Kronenberg F . Hot flashes: epidemiology and physiology . Ann N Y Acad Sci. 1990 ; 592 : 52 – 86 ; discussion 123 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Freedman RR . Menopausal hot flashes: mechanisms, endocrinology, treatment . J Steroid Biochem Mol Biol. 2014 ; 142 : 115 – 120 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Low DA , et al. Mechanisms of cutaneous vasodilation during the postmenopausal hot flash . Menopause . 2011 ; 18 ( 4 ): 359 – 365 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Low DA , et al. Cutaneous and hemodynamic responses during hot flashes in symptomatic postmenopausal women . Menopause. 2008 ; 15 ( 2 ): 290 – 295 . Google Scholar Crossref Search ADS PubMed WorldCat 8. El Khoudary SR , et al. Cardiovascular implications of the menopause transition: endogenous sex hormones and vasomotor symptoms . Obstet Gynecol Clin North Am. 2018 ; 45 ( 4 ): 641 – 661 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Thurston RC , et al. ; Appendix . Trajectories of vasomotor symptoms and carotid intima media thickness in the study of women’s health across the nation . Stroke. 2016 ; 47 ( 1 ): 12 – 17 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Thurston RC , et al. Adipokines, adiposity, and vasomotor symptoms during the menopause transition: findings from the Study of Women’s Health Across the Nation . Fertil Steril. 2013 ; 100 ( 3 ): 793 – 800 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Thurston RC , et al. Vasomotor symptoms and lipid profiles in women transitioning through menopause . Obstet Gynecol. 2012 ; 119 ( 4 ): 753 – 761 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Thurston RC , et al. Vasomotor symptoms and insulin resistance in the study of women’s health across the nation . J Clin Endocrinol Metab. 2012 ; 97 ( 10 ): 3487 – 3494 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Trinder J , et al. Sleep and cardiovascular regulation . Pflugers Arch. 2012 ; 463 ( 1 ): 161 – 168 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Carrington MJ , et al. Blood pressure and heart rate during continuous experimental sleep fragmentation in healthy adults . Sleep . 2008 ; 31 ( 12 ): 1701 – 1712 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Routledge FS , et al. Insomnia symptoms are associated with abnormal endothelial function . J Cardiovasc Nurs. 2017 ; 32 ( 1 ): 78 – 85 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Salles GF , et al. ; ABC-H Investigators . Prognostic effect of the nocturnal blood pressure fall in hypertensive patients: The Ambulatory Blood Pressure Collaboration in Patients With Hypertension (ABC-H) meta-analysis . Hypertension. 2016 ; 67 ( 4 ): 693 – 700 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Baker FC , et al. . Sleep problems during the menopausal transition: prevalence, impact, and management challenges . Nat Sci Sleep . 2018 ; 10 : 73 – 95 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Kravitz HM , et al. Sleep during the perimenopause: a SWAN story . Obstet Gynecol Clin North Am. 2011 ; 38 ( 3 ): 567 – 586 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Ohayon MM . Severe hot flashes are associated with chronic insomnia . Arch Intern Med. 2006 ; 166 ( 12 ): 1262 – 1268 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Woodward S , et al. The thermoregulatory effects of menopausal hot flashes on sleep . Sleep . 1994 ; 17 ( 6 ): 497 – 501 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Freedman RR , et al. Lack of sleep disturbance from menopausal hot flashes . Fertil Steril . 2004 ; 82 ( 1 ): 138 – 144 . Google Scholar Crossref Search ADS PubMed WorldCat 22. Freedman RR , et al. Effects of REM sleep and ambient temperature on hot flash-induced sleep disturbance . Menopause . 2006 ; 13 ( 4 ): 576 – 583 . Google Scholar Crossref Search ADS PubMed WorldCat 23. de Zambotti M , et al. Magnitude of the impact of hot flashes on sleep in perimenopausal women . Fertil Steril. 2014 ; 102 ( 6 ): 1708 – 1715.e1 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Erlik Y , et al. Association of waking episodes with menopausal hot flushes . JAMA . 1981 ; 245 ( 17 ): 1741 – 1744 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Gonen R , et al. The association between mid-sleep waking episodes and hot flushes in post-menopausal women . J Psychosom Obst Gynecol . 1986 ; 5 : 113 – 117 . Google Scholar Crossref Search ADS WorldCat 26. Joffe H , et al. . A gonadotropin-releasing hormone agonist model demonstrates that nocturnal hot flashes interrupt objective sleep . Sleep . 2013 ; 36 ( 12 ): 1977 – 1985 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Gast GC , et al. Menopausal complaints are associated with cardiovascular risk factors . Hypertension . 2008 ; 51 ( 6 ): 1492 – 1498 . Google Scholar Crossref Search ADS PubMed WorldCat 28. James GD , et al. Ambulatory blood pressure and heart rate in relation to hot flash experience among women of menopausal age . Ann Hum Biol . 2004 ; 31 ( 1 ): 49 – 58 . Google Scholar Crossref Search ADS PubMed WorldCat 29. Gast GC , et al. Vasomotor menopausal symptoms are associated with increased risk of coronary heart disease . Menopause . 2011 ; 18 ( 2 ): 146 – 151 . Google Scholar PubMed WorldCat 30. Thurston RC , et al. Changes in heart rate variability during vasomotor symptoms among midlife women . Menopause . 2016 ; 23 ( 5 ): 499 – 505 . Google Scholar Crossref Search ADS PubMed WorldCat 31. de Zambotti M , et al. Vagal withdrawal during hot flashes occurring in undisturbed sleep . Menopause 2013 ; 20 (11) : 1147 - 1153 . Google Scholar Crossref Search ADS PubMed WorldCat 32. Tuomikoski P , et al. Vasomotor hot flushes and 24-hour ambulatory blood pressure in normotensive women: a placebo-controlled trial on post-menopausal hormone therapy . Ann Med . 2010 ; 42 ( 5 ): 334 – 343 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Sherwood A , et al. Methodological guidelines for impedance cardiography . Psychophysiology . 1990 ; 27 ( 1 ): 1 – 23 . Google Scholar Crossref Search ADS PubMed WorldCat 34. Baker FC , et al. Insomnia in women approaching menopause: beyond perception . Psychoneuroendocrinology . 2015 ; 60 : 96 – 104 . Google Scholar Crossref Search ADS PubMed WorldCat 35. Sassoon SA , et al. Association between personality traits and DSM-IV diagnosis of insomnia in peri- and postmenopausal women . Menopause . 2014 ; 21 ( 6 ): 602 – 611 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Soules MR , et al. Executive summary: Stages of Reproductive Aging Workshop (STRAW) . Fertil Steril . 2001 ; 76 ( 5 ): 874 – 878 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Beck AT , et al. Manual for the Beck Depression Inventory . 2nd ed. San Antonio, TX : The Psychological Corporation ; 1996 . Google Preview WorldCat COPAC 38. Buysse DJ , et al. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research . Psychiatry Res . 1989 ; 28 ( 2 ): 193 – 213 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Nicassio PM , et al. The phenomenology of the pre-sleep state: the development of the pre-sleep arousal scale . Behav Res Ther . 1985 ; 23 ( 3 ): 263 – 271 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Iber C . The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specifications . Am Acad Sleep Med . 2007 . WorldCat 41. Castiglioni P , et al. Broad-band spectral analysis of 24 h continuous finger blood pressure: comparison with intra-arterial recordings . Clin Sci (Lond) . 1999 ; 97 ( 2 ): 129 – 139 . Google Scholar PubMed WorldCat 42. Silke B , et al. Accuracy and precision of blood pressure determination with the Finapres: an overview using re-sampling statistics . J Hum Hypertens. 1998 ; 12 ( 6 ): 403 – 409 . Google Scholar Crossref Search ADS PubMed WorldCat 43. James GD , et al. Measuring arterial blood pressure in humans: auscultatory and automatic measurement techniques for human biological field studies . Am J Hum Biol. 2018 ; 30 (1) : 1 - 16 . WorldCat 44. Sherwood A , et al. Comparison of impedance cardiographic measurements using band and spot electrodes . Psychophysiology . 1992 ; 29 ( 6 ): 734 – 741 . Google Scholar Crossref Search ADS PubMed WorldCat 45. Forouzanfar M , et al. Automatic artifact detection in impedance cardiogram using pulse similarity index . Conf Proc IEEE Eng Med Biol Soc . 2019 . In press. WorldCat 46. Forouzanfar M , et al. Automatic analysis of pre-ejection period during sleep using impedance cardiogram . Psychophysiology . 2019 ; 56 ( 7 ): e13355 . Google Scholar Crossref Search ADS PubMed WorldCat 47. Forouzanfar M , et al. Toward a better noninvasive assessment of preejection period: a novel automatic algorithm for B-point detection and correction on thoracic impedance cardiogram . Psychophysiology . 2018 ; 55 ( 8 ): e13072 . Google Scholar Crossref Search ADS PubMed WorldCat 48. Cacioppo JT , et al. Autonomic cardiac control. II. noninvasive indices and basal response as revealed by autonomic blockades . Psychophysiology . 1994 ; 31 ( 6 ): 586 – 598 . Google Scholar Crossref Search ADS PubMed WorldCat 49. Forouzanfar M , et al. Automatic detection of hot flash occurrence and timing from skin conductance activity . Conf Proc IEEE Eng Med Biol Soc . 2018 ; 2018 : 1090 – 1093 . Google Scholar PubMed WorldCat 50. Freedman RR . Laboratory and ambulatory monitoring of menopausal hot flashes . Psychophysiology . 1989 ; 26 ( 5 ): 573 – 579 . Google Scholar Crossref Search ADS PubMed WorldCat 51. Savard MH , et al. Relationship between objectively recorded hot flashes and sleep disturbances among breast cancer patients: investigating hot flash characteristics other than frequency . Menopause . 2013 ; 20 ( 10 ): 997 – 1005 . Google Scholar Crossref Search ADS PubMed WorldCat 52. Thurston RC , et al. Support Vector Machines to improve physiologic hot flash measures: application to the ambulatory setting . Psychophysiology . 2011 ; 48 ( 7 ): 1015 – 1021 . Google Scholar Crossref Search ADS PubMed WorldCat 53. Carpenter JS , et al. Is sternal skin conductance monitoring a valid measure of hot flash intensity or distress? Menopause . 2005 ; 12 ( 5 ): 512 – 519 . Google Scholar Crossref Search ADS PubMed WorldCat 54. Trinder J , et al. On the nature of cardiovascular activation at an arousal from sleep . Sleep . 2003 ; 26 ( 5 ): 543 – 551 . Google Scholar PubMed WorldCat 55. Davies RJ , et al. Arterial blood pressure responses to graded transient arousal from sleep in normal humans . J Appl Physiol (1985) . 1993 ; 74 ( 3 ): 1123 – 1130 . Google Scholar Crossref Search ADS PubMed WorldCat 56. Morgan BJ , et al. Neurocirculatory consequences of abrupt change in sleep state in humans . J Appl Physiol (1985) 1996 ; 80 : 1627 – 36 . Google Scholar Crossref Search ADS PubMed WorldCat 57. Tamisier R , et al. Sleep biology updates: hemodynamic and autonomic control in sleep disorders . Metabolism. 2018 ; 84 : 3 – 10 . Google Scholar Crossref Search ADS PubMed WorldCat 58. Chouchou F , et al. ; PROOF Study Group . Sympathetic overactivity due to sleep fragmentation is associated with elevated diurnal systolic blood pressure in healthy elderly subjects: the PROOF-SYNAPSE study . Eur Heart J. 2013 ; 34 ( 28 ): 2122 – 2131, 2131a . Google Scholar Crossref Search ADS PubMed WorldCat 59. Taylor KS , et al. Arousal from sleep and sympathetic excitation during wakefulness . Hypertension . 2016 ; 68 ( 6 ): 1467 – 1474 . Google Scholar Crossref Search ADS PubMed WorldCat 60. de Zambotti M , et al. Altered nocturnal blood pressure profiles in women with insomnia disorder in the menopausal transition . Menopause . 2017 ; 24 ( 3 ): 278 – 287 . Google Scholar Crossref Search ADS PubMed WorldCat 61. Sturdee DW , et al. Physiological aspects of menopausal hot flush . Br Med J . 1978 ; 2 ( 6130 ): 79 – 80 . Google Scholar Crossref Search ADS PubMed WorldCat 62. Kato T , et al. Experimentally induced arousals during sleep: a cross-modality matching paradigm . J Sleep Res. 2004 ; 13 ( 3 ): 229 – 238 . Google Scholar Crossref Search ADS PubMed WorldCat 63. Freedman RR , et al. Alpha 2-adrenergic mechanism in menopausal hot flushes . Obstet Gynecol. 1990 ; 76 ( 4 ): 573 – 578 . Google Scholar PubMed WorldCat 64. Zepelin H , et al. Effects of age on auditory awakening thresholds . J Gerontol . 1984 ; 39 ( 3 ): 294 – 300 . Google Scholar Crossref Search ADS PubMed WorldCat 65. Halász P , et al. The nature of arousal in sleep . J Sleep Res. 2004 ; 13 ( 1 ): 1 – 23 . Google Scholar Crossref Search ADS PubMed WorldCat 66. Rechtschaffen A , et al. Auditory awakening thresholds in REM and NREM sleep stages . Percept Mot Skills. 1966 ; 22 ( 3 ): 927 – 942 . Google Scholar Crossref Search ADS PubMed WorldCat 67. Sagot JC , et al. Sweating responses and body temperatures during nocturnal sleep in humans . Am J Physiol. 1987 ; 252 ( 3 Pt 2 ): R462 – R470 . Google Scholar PubMed WorldCat 68. Franco OH , et al. Vasomotor symptoms in women and cardiovascular risk markers: systematic review and meta-analysis . Maturitas . 2015 ; 81 ( 3 ): 353 – 361 . Google Scholar Crossref Search ADS PubMed WorldCat 69. Sassarini J , et al. Vascular function and cardiovascular risk factors in women with severe flushing . Maturitas . 2015 ; 80 ( 4 ): 379 – 383 . Google Scholar Crossref Search ADS PubMed WorldCat 70. Gerber LM , et al. Hot flashes are associated with increased ambulatory systolic blood pressure . Menopause . 2007 ; 14 ( 2 ): 308 – 315 . Google Scholar Crossref Search ADS PubMed WorldCat 71. Jackson EA , et al. Hot flash frequency and blood pressure: data from the Study of Women’s Health Across the Nation . J Womens Health (Larchmt) . 2016 ; 25 ( 12 ): 1204 – 1209 . Google Scholar Crossref Search ADS PubMed WorldCat 72. Brown DE , et al. Relationship between hot flashes and ambulatory blood pressure: the Hilo Women’s Health Study . Psychosom Med. 2011 ; 73 ( 2 ): 166 – 172 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Congiu,, Patrizia;Mariani,, Sara;Milioli,, Giulia;Parrino,, Liborio;Tamburrino,, Ludovica;Borghero,, Giuseppe;Defazio,, Giovanni;Pereira,, Bruno;Fantini, Maria, L;Puligheddu,, Monica

doi: 10.1093/sleep/zsz164pmid: 31312838

Geng,, Fulei;Liang,, Yingxin;Li,, Yuanyuan;Fang,, Yi;Pham, Tien, Sy;Liu,, Xianchen;Fan,, Fang

doi: 10.1093/sleep/zsz162pmid: 31328781

Abstract Study Objectives To assess insomnia symptoms in adolescents with probable posttraumatic stress disorder (PTSD) and to determine whether there are longitudinal and reciprocal associations between insomnia, PTSD, and depressive symptoms. Methods Participants were 1,492 adolescent survivors who had been exposed to the 2008 Wenchuan earthquake in China. Insomnia, PTSD, and depressive symptoms were measured at 12 months (T1, n = 1407), 18 months (T2, n = 1335), and 24 months (T3, n = 1361) postearthquake by self-report questionnaires. Generalized estimating equation (GEE) models were used to examine the cross-sectional and longitudinal associations of insomnia with PTSD and depression. Results Insomnia, PTSD, and depressive symptoms were common among adolescent survivors. Among participants with probable PTSD, approximately 47% (48.5%, T1; 48.1%, T2; and 44.2%, T3) reported difficulty falling asleep or difficulty maintaining sleep. Cross-sectional analyses showed that insomnia co-occurred with PTSD (odds ratio [OR] = 2.04) and depressive symptoms (OR = 2.10). Longitudinal analyses revealed that probable PTSD (OR = 1.50) and depression (OR = 1.42) predicted the incidence of insomnia; in turn, insomnia predicted the incidence of depression (OR = 1.65) over time. Furthermore, PTSD predicted (OR = 3.11) and was predicted (OR = 3.25) by depressive symptoms. Conclusions There is a bidirectional relationship between insomnia, PTSD, and depressive symptoms. This suggests that insomnia, PTSD, and depression are intertwined over time. insomnia, PTSD, depression, adolescents, longitudinal study Statement of Significance This is the first study to examine cross-sectional and longitudinal relationships between insomnia and posttraumatic stress disorder (PTSD)/depression in a large sample of adolescents with three waves of data collection. The findings provide new evidence of bidirectional relationships among insomnia, PTSD, and depression in adolescents after a traumatic disaster. The reciprocal associations among insomnia, PTSD, and depression provide insights into the improvement of interventions for postdisaster mental health problems, particularly for residual insomnia symptoms in PTSD. Future research needs to validate these findings in other trauma-exposed population as well as to expand upon them to determine possible moderating and mediating variables. Clinical studies should examine whether intervening in merely one condition can relieve other symptoms. Introduction Insomnia symptoms are highly prevalent in posttraumatic stress disorder (PTSD) and are among the most distressing chronic symptoms [1–3]. It is estimated that 40% to 50% patients with PTSD suffer from difficulty initiating or maintaining sleep [4, 5]. PTSD comorbid with insomnia can result in adverse outcomes, including but not limited to suicide, functional impairment, and physical health problems [6–8]. Furthermore, studies have indicated that insomnia, in comparison to other sleep disturbances (e.g. nightmares), is less likely to respond to treatment for PTSD [9, 10]. Indeed, residual insomnia is one of the most intractable problems for current treatment of PTSD [10, 11]. Therefore, understanding the causal relationship between insomnia and PTSD is of great importance for not only the exploration of the mechanisms of their comorbidity but also the improvement of clinical intervention. There have already been a few longitudinal studies examining the associations between PTSD and insomnia [12–16]. Specifically, two studies that similarly took the military as participants both investigated the impact of insomnia before deployment on later development of PTSD after military deployment, but their results came out to be distinct. One found that predeployment insomnia symptoms were significantly associated with higher odds of developing PTSD, depression, and anxiety postdeployment [12], while the other indicated that predeployment nightmares rather than insomnia predicted PTSD symptoms at 6 months postdeployment [13]. Bryant et al. assessed 1,033 traumatically injured patients during hospital admission and at 3 months after injury and found that insomnia immediately prior to trauma predicted subsequent PTSD [14]. In addition, with a sample of injured survivors of motor vehicle accidents, one study reported that insomnia complaints as early as 1 month after the trauma was possible to detect subjects who would later develop chronic PTSD [15]. There was another study reporting that insomnia at 8 weeks postpartum was associated with the maintenance of PTSD symptoms 2 years postpartum [16]. In summary, extant studies have indicated that both pretrauma and peri-trauma insomnia would contribute to the development and maintenance of PTSD. Insomnia and PTSD may persist for a long time [17]. It is unknown whether insomnia could predict PTSD in the long term after trauma. To our best knowledge, there has been only one study examining the bidirectional relationship between insomnia and PTSD, which indicated that insomnia significantly predicted PTSD, whereas PTSD failed to predict insomnia among Iraq combat veterans [18]. PTSD is highly comorbid with depression [19]. Longitudinal studies have shown that PTSD could predict and was also predicted by depression [20, 21]. In general population, the bidirectional relationship between insomnia and depression has already been confirmed [22, 23]. Given that trauma is a common risk factor for both depression and insomnia, a bidirectional relationship between depression and insomnia in trauma-exposed population could be inferred. Thus, the longitudinal interplay of insomnia, PTSD, and depression would be expected. However, no studies to date have examined these associations in traumatized children and adolescents. In our previous studies, we found significant cross-sectional relationships between sleep duration, sleep disturbances, and PTSD symptoms [17]. Furthermore, sleep disturbances predicted the persistence of PTSD and depressive symptoms [24]. Expanding on those findings, the primary aims of the current study were to report the prevalent rates of insomnia in adolescent PTSD and to examine the bidirectional relationships between insomnia, PTSD, and depression in a large cohort of earthquake survivors with three waves of data collection. Specifically, the present study examined (1) the prevalence of insomnia symptoms in PTSD; (2) whether insomnia was associated with PTSD or depressive symptoms cross-sectionally; and (3) whether there were bidirectional relationships between insomnia, PTSD, and depressive symptoms. We hypothesized that insomnia, PTSD, and depressive symptoms were related bidirectionally with each other. Given that PTSD is highly comorbid with depression, PTSD or depressive symptoms would be controlled when examining the associations of insomnia and the other symptoms, which might result in seemingly weakened longitudinal relationships. Methods Participants and procedure Participants were adolescent survivors who participated in the Wenchuan Earthquake Adolescent Health Cohort (WEAHC) study [25]. The purpose of the WEAHC study was to document the long-term mental health outcomes among adolescent survivors, as well as to explore the underlying psychological, neurobiological, and genetic mechanisms. Details of the study aims and design had been described in our previous studies [17, 25–27]. The WEAHC study enrolled 1,573 students from one junior high school and one senior high school in Dujiangyan District, which was directly exposed to the Wenchuan earthquake. Participants were initially assessed at 6 months postearthquake and then were followed for four times at 6 month intervals. Because insomnia, PTSD, and depression symptoms were assessed at 12 (T1), 18 (T2), and 24 (T3) months postearthquake, these three waves of data were used in the current study. There were 1,407, 1,335, and 1,361 students completing the measures of insomnia, PTSD, and depression at T1, T2, and T3, respectively. The response rates were 94.3%, 89.5%, and 91.2% at T1, T2, and T3, respectively. Males were more likely to miss out than females at T1 and T2; participants who reported witnesses of the tragic scenes were more likely to miss out than those who did not at T2. Details of demographics and earthquake exposure are shown in Table 1. Table 1. Demographics and earthquake exposure variables (N = 1492) Variable n (%) Sex  Male 666 (44.6)  Female 826 (55.4) Mean age at 12 months postearthquake, years (SD) 15.01 (1.26) Grade  Junior high students 212 (14.2)  Senior high students 1,280 (85.8) No. of children in the family  1 1,239 (83.0)  ≥2 253 (17.0) Residence location  Urban 836 (56.0)  Rural 656 (44.0) Death, disappearance, and/or injury of family members  No 1,107 (74.2)  Injury 193 (12.9)  Death/disappearance 192 (12.9) House damage  No 461 (30.9)  Moderate 391 (26.2)  Severe 640 (42.9) Property loss other than house damage  No 641 (43.0)  Moderate 524 (35.1)  Severe 327 (21.9) Direct witness of tragic scenes  No 602 (40.3)  Yes 890 (59.7) Variable n (%) Sex  Male 666 (44.6)  Female 826 (55.4) Mean age at 12 months postearthquake, years (SD) 15.01 (1.26) Grade  Junior high students 212 (14.2)  Senior high students 1,280 (85.8) No. of children in the family  1 1,239 (83.0)  ≥2 253 (17.0) Residence location  Urban 836 (56.0)  Rural 656 (44.0) Death, disappearance, and/or injury of family members  No 1,107 (74.2)  Injury 193 (12.9)  Death/disappearance 192 (12.9) House damage  No 461 (30.9)  Moderate 391 (26.2)  Severe 640 (42.9) Property loss other than house damage  No 641 (43.0)  Moderate 524 (35.1)  Severe 327 (21.9) Direct witness of tragic scenes  No 602 (40.3)  Yes 890 (59.7) Open in new tab Table 1. Demographics and earthquake exposure variables (N = 1492) Variable n (%) Sex  Male 666 (44.6)  Female 826 (55.4) Mean age at 12 months postearthquake, years (SD) 15.01 (1.26) Grade  Junior high students 212 (14.2)  Senior high students 1,280 (85.8) No. of children in the family  1 1,239 (83.0)  ≥2 253 (17.0) Residence location  Urban 836 (56.0)  Rural 656 (44.0) Death, disappearance, and/or injury of family members  No 1,107 (74.2)  Injury 193 (12.9)  Death/disappearance 192 (12.9) House damage  No 461 (30.9)  Moderate 391 (26.2)  Severe 640 (42.9) Property loss other than house damage  No 641 (43.0)  Moderate 524 (35.1)  Severe 327 (21.9) Direct witness of tragic scenes  No 602 (40.3)  Yes 890 (59.7) Variable n (%) Sex  Male 666 (44.6)  Female 826 (55.4) Mean age at 12 months postearthquake, years (SD) 15.01 (1.26) Grade  Junior high students 212 (14.2)  Senior high students 1,280 (85.8) No. of children in the family  1 1,239 (83.0)  ≥2 253 (17.0) Residence location  Urban 836 (56.0)  Rural 656 (44.0) Death, disappearance, and/or injury of family members  No 1,107 (74.2)  Injury 193 (12.9)  Death/disappearance 192 (12.9) House damage  No 461 (30.9)  Moderate 391 (26.2)  Severe 640 (42.9) Property loss other than house damage  No 641 (43.0)  Moderate 524 (35.1)  Severe 327 (21.9) Direct witness of tragic scenes  No 602 (40.3)  Yes 890 (59.7) Open in new tab The Human Research Ethics Committee of South China Normal University approved the WEAHC study. Students completed self-administered paper-and-pencil questionnaires in the classroom setting during school days under instructions of trained interviewers who were psychological professionals from South China Normal University. Written informed consent and parental permission were obtained from all participants at each assessment. Measures Demographics and exposure to the earthquake Demographic characteristics included gender (0 = male, 1 = female), grade in school at baseline (0 = 7th, 1 = 10th), number of children in the family (0 = one child, 1 = more than one children), and residence (0 = urban, 1 = rural). The degree of exposure to the earthquake was assessed with the following items: death, disappearance, and/or injury of family members; house damage; property loss other than house damage; and witness or hearing of tragic scenes. For the first item, the choices were: “death of family members,” “disappearance of family members,” “serious injury of family members,” “moderate injury of family members,” or “none of the above”; for house damage and property loss other than house damage, the degree of exposure was measured on a five-point scale; for witness or hearing of tragic scenes, the choices were: “witness a lot,” “witness some,” “hear a lot,” “hear some,” and “none of the above.” When the earthquake occurred, participants were at school. Thus, witness or hearing of tragic scenes might be different from death of family members, house damage, or property loss. Insomnia Insomnia symptoms during the past month were assessed with two particular items extracted from the Pittsburgh Sleep Quality Index (PSQI) [28]: “Difficulty falling asleep: cannot get to sleep within 30 minutes” and “Difficulty maintaining sleep: wake up in the middle of the night or early morning.” Both items were rated on a four-point scale: “never,” “less than once a week,” “once or twice a week,” and “three or more times a week.” Insomnia was defined by the selection of “three or more times a week” on either item (or both). To screen high-risk adolescents quickly and effectively, the two items rather than full version PSQI were used. The method to assess insomnia here was validated in our previous studies of adolescents [29, 30]. PTSD symptoms The severity of PTSD symptoms during the past 6 months was measured by the PTSD Self-Rating Scale (PTSD-SS), which was developed on the basis of PTSD diagnostic criteria in Diagnostic and Statistical Manual of Mental Disorders (4th ed.; DSM-IV) and the Chinese Classification of Mental Disorders (2nd ed., rev.; CCMD-2-R) [31]. The PTSD-SS consists of 24 items, which are all scored on a scale of 1 (none) to 5 (very severe). Two PTSD-SS sleep-related items (nightmares and sleep disorders) were omitted in the present analyses to avoid overlap with the symptom of insomnia, so the PTSD severity score was based on the remaining 22 items. The possible total score range was 24–110, and a cutoff score of 46 was used to screen clinically significant PTSD symptoms [31]. The full version of PTSD-SS was also analyzed to examine whether inflation effect of repeated items existed. The PTSD-SS has been demonstrated to have satisfactory test–retest reliability, internal consistency, and construct validity in Chinese adolescents [31, 32]. Cronbach’s alpha of the PTSD-SS in the present study ranged from .945 to .956. Depressive symptoms Adolescents’ depressive symptoms over the previous week were measured using the Depression Self-Rating Scale (DSRS) [33]. The Chinese version of the DSRS has been validated, showing good reliability and validity among Chinese adolescents [34]. The DSRS consists of 18 items, each of which is scored 0–2. Most of the items are similar to DSM diagnostic criteria for depression. However, the DSRS measures nightmares but not suicide. Two DSRS sleep-related items (nightmares and sleep quality) were omitted as was for the PTSD-SS; thus, the DSRS severity score was based on the remaining 16 items. The total score ranged from 0 to 32. A cutoff score of 14 was used to classify whether an adolescent had clinically significant depressive symptoms or not. Also, the full version of DSRS was also analyzed to examine the inflation effect. Cronbach’s alpha of the DSRS ranged from .798 to .812 in the present study. Statistical analyses Analyses were conducted using SAS version 9.4 for Windows (SAS Inst., Cary, NC). Chi-square tests were conducted to identify possible characteristics of participants who had not completed all the three surveys. Generalized estimating equation (GEE) models with the unstructured correlation structure were used to test the cross-sectional and longitudinal associations between insomnia, PTSD, and depression. GEE is a type of regression analysis that accounts for within-subject correlation across repeated measurements and is appropriate to estimate population-averaged effects over time [35]. In addition, GEE is suitable for both dichotomous and continuous outcome variables; specifically, for dichotomous outcome variables, the GEE models worked out odds ratios (ORs) and 95% confidence intervals (CIs) For cross-sectional analyses in this study, insomnia was the dependent variable, while PTSD and depression were entered simultaneously into the equation as time-varying independent variables. For longitudinal analyses, two sets of lagged GEE models were analyzed: the first set of analysis was conducted to examine the prediction from PTSD and depression to insomnia, in which insomnia at time t + 1 (i.e. T2 and T3) was the dependent variable, while PTSD and depression at time t (T1 and T2) were modeled as the predictors with baseline (T1) insomnia being controlled; the second set was conducted to examine the prediction from insomnia to PTSD or depression, in which PTSD and depression at time t + 1 (T2 and T3) were set as the dependent variable separately, while insomnia at time t (T1 and T2) was modeled as the predictor with baseline (T1) PTSD or depression being controlled. The visualized analyzing methods are displayed in Figure 1. Figure 1. Open in new tabDownload slide Visualization of the longitudinal lagged GEE models to predict (a) insomnia and (b) PTSD and depression, with demographics and earthquake exposure being controlled. Abbreviations: INSOM: insomnia; DEP: depression; PTSD: posttraumatic stress disorder. Figure 1. Open in new tabDownload slide Visualization of the longitudinal lagged GEE models to predict (a) insomnia and (b) PTSD and depression, with demographics and earthquake exposure being controlled. Abbreviations: INSOM: insomnia; DEP: depression; PTSD: posttraumatic stress disorder. In order to examine the unique effect of each independent variable, simple (model 1 and model 2) and multiple (model 3) analyses were conducted. All GEE analyses in this study were performed under adjustments for demographics (sex, grade, number of children in the family, and residence) and earthquake exposure variables (death, disappearance, and/or injury of family members, house damage, property loss other than house damage, and witness or hearing of tragic scenes). In addition, multiple linear regressions with PTSD and depression as continuous variables were conducted to evaluate the stability of the results. The main findings were similar to logistic regressions. To examine the inflation effects of repeated items, both data from full version scales and after deletion of sleep items were analyzed. The two methods showed no difference. Finally, Sobel test was used to examine possible mediation [36]. All p values were two sided and statistical significance was determined at p < .05. Among the 1,573 target participants, 1,492 adolescents who had completed at least one time point constituted the sample of this study. Missing data were replaced with estimates derived by single imputation, which was performed using an expectation maximization algorithm with all T1 variables as predictors. In order to investigate possible bias, we reran all analyses using only participants who had no missing data, and the results of the two methods were similar. Thus, only the results of single imputation are presented. Results In the current sample, the prevalence rates of probable PTSD were 22.5%, 12.7%, and 13.8% at T1, T2, and T3, respectively; the prevalence rates of depressive symptoms were 38.0%, 28.8%, and 36.7% at T1, T2, and T3, respectively; and the prevalence rates of insomnia symptoms were 29.4%, 23.5%, and 23.5% at T1, T2, and T3, respectively. Figure 2 displays proportions of depression and insomnia by PTSD status. Overall, among participants with probable PTSD, approximately 70% had comorbid depressive symptoms (73.2%, T1; 68.2%, T2; 70.9%, T3) and approximately 47% complained of insomnia (48.5%, T1; 48.1%, T2; 44.2%, T3). Figure 2. Open in new tabDownload slide Proportions of probable depression and insomnia by probable PTSD status. Figure 2. Open in new tabDownload slide Proportions of probable depression and insomnia by probable PTSD status. The cross-sectional associations between insomnia, PTSD, and depressive symptoms are presented in Table 2. Insomnia was significantly associated with PTSD symptoms (OR = 2.04, p < .001), after adjustment of demographics, earthquake exposure variables, and depression. When demographics, earthquake exposure variables, and PTSD were controlled, insomnia was also significantly associated with depressive symptoms (OR = 2.10, p < .001). Table 2. Cross-sectional GEE model to predict insomnia Variable Insomnia OR [95% CI] Sex, ref = male 0.97 [0.81, 1.16] Grade at baseline, ref = seventh 1.73 [1.28, 2.33]*** No. of children in the family, ref = 1 1.05 [0.83, 1.33] Location, ref = urban 1.24 [1.03, 1.49]* Death, disappearance, and/or injury of family members, ref = no  Injury 0.93 [0.71, 1.22]  Death/disappearance 0.94 [0.72, 1.24] House damage, ref = no  Moderate 0.92 [0.72, 1.17]  Severe 0.95 [0.75, 1.21] Property loss, ref = no  Moderate 0.87 [0.70, 1.09]  Severe 1.09 [0.83, 1.43] Direct witness of tragic scenes, ref= no 1.21 [1.00, 1.46]* PTSDa, ref = no 2.04 [1.67, 2.49]*** Depressiona, ref = no 2.10 [1.78, 2.47]*** Variable Insomnia OR [95% CI] Sex, ref = male 0.97 [0.81, 1.16] Grade at baseline, ref = seventh 1.73 [1.28, 2.33]*** No. of children in the family, ref = 1 1.05 [0.83, 1.33] Location, ref = urban 1.24 [1.03, 1.49]* Death, disappearance, and/or injury of family members, ref = no  Injury 0.93 [0.71, 1.22]  Death/disappearance 0.94 [0.72, 1.24] House damage, ref = no  Moderate 0.92 [0.72, 1.17]  Severe 0.95 [0.75, 1.21] Property loss, ref = no  Moderate 0.87 [0.70, 1.09]  Severe 1.09 [0.83, 1.43] Direct witness of tragic scenes, ref= no 1.21 [1.00, 1.46]* PTSDa, ref = no 2.04 [1.67, 2.49]*** Depressiona, ref = no 2.10 [1.78, 2.47]*** aTime-varying covariates. *p < .05; **p < .01; ***p < .001. Open in new tab Table 2. Cross-sectional GEE model to predict insomnia Variable Insomnia OR [95% CI] Sex, ref = male 0.97 [0.81, 1.16] Grade at baseline, ref = seventh 1.73 [1.28, 2.33]*** No. of children in the family, ref = 1 1.05 [0.83, 1.33] Location, ref = urban 1.24 [1.03, 1.49]* Death, disappearance, and/or injury of family members, ref = no  Injury 0.93 [0.71, 1.22]  Death/disappearance 0.94 [0.72, 1.24] House damage, ref = no  Moderate 0.92 [0.72, 1.17]  Severe 0.95 [0.75, 1.21] Property loss, ref = no  Moderate 0.87 [0.70, 1.09]  Severe 1.09 [0.83, 1.43] Direct witness of tragic scenes, ref= no 1.21 [1.00, 1.46]* PTSDa, ref = no 2.04 [1.67, 2.49]*** Depressiona, ref = no 2.10 [1.78, 2.47]*** Variable Insomnia OR [95% CI] Sex, ref = male 0.97 [0.81, 1.16] Grade at baseline, ref = seventh 1.73 [1.28, 2.33]*** No. of children in the family, ref = 1 1.05 [0.83, 1.33] Location, ref = urban 1.24 [1.03, 1.49]* Death, disappearance, and/or injury of family members, ref = no  Injury 0.93 [0.71, 1.22]  Death/disappearance 0.94 [0.72, 1.24] House damage, ref = no  Moderate 0.92 [0.72, 1.17]  Severe 0.95 [0.75, 1.21] Property loss, ref = no  Moderate 0.87 [0.70, 1.09]  Severe 1.09 [0.83, 1.43] Direct witness of tragic scenes, ref= no 1.21 [1.00, 1.46]* PTSDa, ref = no 2.04 [1.67, 2.49]*** Depressiona, ref = no 2.10 [1.78, 2.47]*** aTime-varying covariates. *p < .05; **p < .01; ***p < .001. Open in new tab The longitudinal associations from PTSD and depressive symptoms to insomnia are demonstrated in Table 3. After adjustment of demographics, earthquake exposure variables, and baseline insomnia, both PTSD (model 1, OR = 1.69, p < .001) and depressive symptoms (model 2, OR = 1.56, p < .001) significantly predicted insomnia independently. When PTSD and depression were simultaneously added into the regression equation, the associations were weakened but still significant (model 3, OR = 1.50, p < .001 for PTSD; OR = 1.42, p < .001 for depression). Table 3. Longitudinal lagged GEE models to predict insomniaa Insomnia Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] Insomnia at baseline 4.67 [3.75, 5.76]*** 4.64 [3.75, 5.75]*** 4.49 [3.62, 5.56]*** PTSDb 1.69 [1.34, 2.13]*** 1.50 [1.18, 1.92]*** Depressionb 1.56 [1.28, 1.90]*** 1.42 [1.16, 1.75]*** Insomnia Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] Insomnia at baseline 4.67 [3.75, 5.76]*** 4.64 [3.75, 5.75]*** 4.49 [3.62, 5.56]*** PTSDb 1.69 [1.34, 2.13]*** 1.50 [1.18, 1.92]*** Depressionb 1.56 [1.28, 1.90]*** 1.42 [1.16, 1.75]*** aDemographics, earthquake exposure variables, and time were controlled in all these models. bTime-varying covariates. *p < .05; **p < .01; ***p < .001. Open in new tab Table 3. Longitudinal lagged GEE models to predict insomniaa Insomnia Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] Insomnia at baseline 4.67 [3.75, 5.76]*** 4.64 [3.75, 5.75]*** 4.49 [3.62, 5.56]*** PTSDb 1.69 [1.34, 2.13]*** 1.50 [1.18, 1.92]*** Depressionb 1.56 [1.28, 1.90]*** 1.42 [1.16, 1.75]*** Insomnia Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] Insomnia at baseline 4.67 [3.75, 5.76]*** 4.64 [3.75, 5.75]*** 4.49 [3.62, 5.56]*** PTSDb 1.69 [1.34, 2.13]*** 1.50 [1.18, 1.92]*** Depressionb 1.56 [1.28, 1.90]*** 1.42 [1.16, 1.75]*** aDemographics, earthquake exposure variables, and time were controlled in all these models. bTime-varying covariates. *p < .05; **p < .01; ***p < .001. Open in new tab In the opposite direction, when demographics, earthquake exposure variables, and baseline PTSD were controlled, both insomnia (model 1, OR = 1.30, p < .05) and depression (model 2, OR = 3.28, p < .001) longitudinally predicted PTSD symptoms (Table 4). However, when insomnia and depression were simultaneously added in the regression, only depression (model 3, OR = 3.25, p < .001) remained predictable to PTSD symptoms, while the longitudinal association from insomnia to PTSD was no longer significant (model 3, OR = 1.11, p = .445). According to Baron and Kenny [37], results of these three models for PTSD here showed a mediation effect of depressive symptoms between insomnia and PTSD. To directly test the indirect effect of insomnia on PTSD symptoms through depression, we performed the Sobel test using SAS macro with 10,000 bootstrap [36]. The results indicated that the indirect effect was 0.047 (p < .001; 95% CI: 0.038–0.058). Table 4. Longitudinal lagged GEE models to predict PTSD and depressiona Variable PTSD Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] PTSD at baseline 15.75 [11.57, 21.46]*** 12.93 [9.50, 17.60]*** 12.63 [9.22, 17.30]*** Depressionb 3.28 [2.47, 4.34]*** 3.25 [2.44, 4.32]*** Insomniab 1.30 [1.00, 1.70]* 1.11 [0.84, 1.46] Depression Model 1 Model 2 Model 3 Variable OR [95% CI] OR [95% CI] OR [95% CI] Depression at baseline 8.34 [6.77, 10.28]*** 7.65 [6.20, 9.44]*** 7.25 [5.87, 8.96]*** PTSDb 3.19 [2.39, 4.27]*** 3.11 [2.31, 4.18]*** Insomniab 1.69 [1.39, 2.05]*** 1.65 [1.35, 2.01]*** Variable PTSD Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] PTSD at baseline 15.75 [11.57, 21.46]*** 12.93 [9.50, 17.60]*** 12.63 [9.22, 17.30]*** Depressionb 3.28 [2.47, 4.34]*** 3.25 [2.44, 4.32]*** Insomniab 1.30 [1.00, 1.70]* 1.11 [0.84, 1.46] Depression Model 1 Model 2 Model 3 Variable OR [95% CI] OR [95% CI] OR [95% CI] Depression at baseline 8.34 [6.77, 10.28]*** 7.65 [6.20, 9.44]*** 7.25 [5.87, 8.96]*** PTSDb 3.19 [2.39, 4.27]*** 3.11 [2.31, 4.18]*** Insomniab 1.69 [1.39, 2.05]*** 1.65 [1.35, 2.01]*** aDemographics, earthquake exposure variables, and time were controlled in all these models. bTime-varying covariates. *p < .05; ***p < .001. Open in new tab Table 4. Longitudinal lagged GEE models to predict PTSD and depressiona Variable PTSD Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] PTSD at baseline 15.75 [11.57, 21.46]*** 12.93 [9.50, 17.60]*** 12.63 [9.22, 17.30]*** Depressionb 3.28 [2.47, 4.34]*** 3.25 [2.44, 4.32]*** Insomniab 1.30 [1.00, 1.70]* 1.11 [0.84, 1.46] Depression Model 1 Model 2 Model 3 Variable OR [95% CI] OR [95% CI] OR [95% CI] Depression at baseline 8.34 [6.77, 10.28]*** 7.65 [6.20, 9.44]*** 7.25 [5.87, 8.96]*** PTSDb 3.19 [2.39, 4.27]*** 3.11 [2.31, 4.18]*** Insomniab 1.69 [1.39, 2.05]*** 1.65 [1.35, 2.01]*** Variable PTSD Model 1 Model 2 Model 3 OR [95% CI] OR [95% CI] OR [95% CI] PTSD at baseline 15.75 [11.57, 21.46]*** 12.93 [9.50, 17.60]*** 12.63 [9.22, 17.30]*** Depressionb 3.28 [2.47, 4.34]*** 3.25 [2.44, 4.32]*** Insomniab 1.30 [1.00, 1.70]* 1.11 [0.84, 1.46] Depression Model 1 Model 2 Model 3 Variable OR [95% CI] OR [95% CI] OR [95% CI] Depression at baseline 8.34 [6.77, 10.28]*** 7.65 [6.20, 9.44]*** 7.25 [5.87, 8.96]*** PTSDb 3.19 [2.39, 4.27]*** 3.11 [2.31, 4.18]*** Insomniab 1.69 [1.39, 2.05]*** 1.65 [1.35, 2.01]*** aDemographics, earthquake exposure variables, and time were controlled in all these models. bTime-varying covariates. *p < .05; ***p < .001. Open in new tab As for the prediction model for depressive symptoms, both insomnia and PTSD significantly predicted the development of depression no matter whether they were included in the equation separately (model 1, OR = 1.69, p < .001 for insomnia; model 2, OR = 3.19, p < .001 for PTSD) or jointly (model 3, OR = 3.11, p < .001 for PTSD; OR = 1.65, p < .001 for insomnia). Discussion To our best knowledge, this is the first longitudinal study to examine the bidirectional associations of insomnia with PTSD and depressive symptoms among trauma-exposed adolescents. In the current sample, insomnia complaints were common in adolescents with probable PTSD. Our findings demonstrated that insomnia symptoms were significantly associated with PTSD and depressive symptoms cross-sectionally. Furthermore, insomnia could predict and was predicted by PTSD and depressive symptoms longitudinally. Depressive symptoms played an important mediation role between insomnia and PTSD symptoms. During the second year after the deadly earthquake, nearly half of the adolescents with probable PTSD reported difficulty initiating or maintaining sleep, which was similar to that for adult PTSD [4, 5]. In addition, consistent with previous numerous cross-sectional studies [4, 5], our findings showed that insomnia was significantly associated with PTSD and depressive symptoms among adolescent survivors. The consistency among various studies highlights that insomnia is highly comorbid with PTSD for both adolescents and adults. Previous longitudinal studies have indicated that pretrauma and peri-trauma insomnia condition predicts the development and maintenance of PTSD [12, 14–16]. Our findings suggested that insomnia was still longitudinally associated with PTSD symptoms even after a long period from the traumatic event. Taken together, these studies imply that insomnia has an important contribution to PTSD among both adults and children. Multiple plausible mechanistic processes may explain the role of insomnia in the developmental pathology of PTSD, such as underlying neurobiological alterations [38], disruption of sleep-dependent emotional processing [39], and compromised generalization of fear extinction [40]. It is possible that the role of sleep is distinct in different stages of trauma. Future studies can explore specific mechanisms according to specific time frame of trauma. In addition, mediators and moderators between insomnia and PTSD should be further examined. As we will discuss below, depressive symptoms, for example, contribute to the link from insomnia to PTSD. There has previously been only one study which examined the longitudinal prediction from PTSD to insomnia. It was reported that depression and PTSD symptoms at 4 months postdeployment failed to significantly predict changes in insomnia at 12 months postdeployment in a sample of Iraq combat veterans [18]. Contrary to that study, our findings supported that PTSD could increase the risk of insomnia over time. The conflicting findings between the two studies may be due to differences in samples (adolescents vs. young adults), study instruments, and particular stages after the trauma. It should also be noted that sleep-related items were removed from the estimation of PTSD symptoms in our study, so the replicate prediction effect of those sleep-related items was excluded. Recently, an ecological momentary assessment study indicated that both sleep- and PTSD-related factors played a role in maintaining insomnia among subjects with PTSD [41]. Extending previous studies in general population [22], our findings demonstrated a reciprocal relationship between insomnia and depression in adolescent earthquake survivors. The results also suggested that depression played an important mediating role between insomnia and PTSD. Several postdisaster studies have documented depression to be intractable over time among adolescents [42, 43]. These highlight the importance of treatment of depression for the prevention and treatment of other postdisaster psychopathology. Moreover, for the first time, our findings showed that there was a vicious circle among insomnia, PTSD, and depression in trauma-exposed adolescents. The interplay of insomnia, PTSD, and depression has important clinical implications. Randomized controlled trials have provided initial support for the use of cognitive behavioral therapy for insomnia or trauma-focused therapy to address sleep disturbances and PTSD symptoms in PTSD patients [44–46], but insomnia still remains residual for nearly half of the patients after therapy. It is likely that intractable depression obstructs the recovery of insomnia and PTSD symptoms. Future behavioral and pharmacological treatment for posttraumatic insomnia should integrate depression or traumatic grief treatment module into existing frameworks. Strengths and limitations Strengths of this study include a large sample size, multiple waves of longitudinal data collection, and standardized measures used to assess PTSD, depression, and insomnia. However, there are several limitations that need to be noted. First, PTSD, depression, and insomnia were measured with self-reported questionnaires, so there might be report bias. Previous studies have observed inconsistencies between subjective and objective sleep outcomes in PTSD patients [3]. Multiple methods to assess sleep such as actigraphy and PSG are needed in future studies. Second, most participants were high school students. We should be cautious not to generalize the findings to other population or other types of traumatic disaster. Third, males and individuals who had witnessed tragic scenes were more likely to drop out in follow-up. Although expectation maximization algorithm was used to impute the missing data, the SE may be underestimated, which could lead to increased risk of false positive rate. Fourth, there were some potential covariates that we had not assessed, such as exposure to new traumatic events, treatment status, and mental conditions before the earthquake. Finally, this is an observational study. Future studies can use intervention methods to further determine the causal relationships among insomnia, PTSD, and depression. Conclusions Insomnia symptoms were common in adolescents with PTSD. Insomnia, PTSD, and depression could prospectively predict each other among adolescents. These findings may provide important clinical implications for the management of sleep disturbances while treating PTSD and depression in adolescent survivors after a traumatic disaster. Further studies are needed to examine the potential mediators and moderators among relationships of insomnia, PTSD, and depression. Funding The present study was funded by the National Natural Science Foundation of China (grant numbers: 31271096, 31671165, and 31700987); Research on the Processes and Repair of Psychological Trauma in Youth, Project of Key Institute of Humanities and Social Sciences, MOE (grant number: 16JJD190001); Guangdong Province Universities and Colleges Pearl River Scholar Funded Scheme (GDUPS 2016). Conflict of interest statement. None declared. References 1. Pigeon WR , et al. Insomnia as a precipitating factor in new onset mental illness: a systematic review of recent findings . Curr Psychiatry Rep. 2017 ; 19 ( 8 ): 44 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Miller KE , et al. . Sleep and dreaming in posttraumatic stress disorder . Curr Psychiatry Rep. 2017 ; 19 ( 10 ): 71 . Google Scholar Crossref Search ADS PubMed WorldCat 3. Spoormaker VI , et al. . Disturbed sleep in post-traumatic stress disorder: secondary symptom or core feature? Sleep Med Rev. 2008 ; 12 ( 3 ): 169 – 184 . Google Scholar Crossref Search ADS PubMed WorldCat 4. Neylan TC , et al. . Sleep disturbances in the Vietnam generation: findings from a nationally representative sample of male Vietnam veterans . Am J Psychiatry. 1998 ; 155 ( 7 ): 929 – 933 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Ohayon MM , et al. . Sleep disturbances and psychiatric disorders associated with posttraumatic stress disorder in the general population . Compr Psychiatry. 2000 ; 41 ( 6 ): 469 – 478 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Short NA , et al. . The mediating role of sleep disturbances in the relationship between posttraumatic stress disorder and self-injurious behavior . J Anxiety Disord. 2015 ; 35 : 68 – 74 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Belleville G , et al. . Impact of sleep disturbances on PTSD symptoms and perceived health . J Nerv Ment Dis. 2009 ; 197 ( 2 ): 126 – 132 . Google Scholar Crossref Search ADS PubMed WorldCat 8. Brownlow JA , et al. . Influence of sleep disturbance on global functioning after posttraumatic stress disorder treatment . J Trauma Stress. 2016 ; 29 ( 6 ): 515 – 521 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Zayfert C , et al. . Residual insomnia following cognitive behavioral therapy for PTSD . J Trauma Stress. 2004 ; 17 ( 1 ): 69 – 73 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Nappi CM , et al. . Treating nightmares and insomnia in posttraumatic stress disorder: a review of current evidence . Neuropharmacology. 2012 ; 62 ( 2 ): 576 – 585 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Brownlow JA , et al. Treatment of sleep disturbances in post-traumatic stress disorder: a review of the literature . Curr Psychiatry Rep. 2015 ; 17 ( 6 ): 41 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Gehrman P , et al. Predeployment sleep duration and insomnia symptoms as risk factors for new-onset mental health disorders following military deployment . Sleep. 2013 ; 36 ( 7 ): 1009 – 1018 . doi: https://doi.org/10.5665/sleep.2798 . Google Scholar Crossref Search ADS PubMed WorldCat 13. van Liempt S , et al. . Impact of impaired sleep on the development of PTSD symptoms in combat veterans: a prospective longitudinal cohort study . Depress Anxiety. 2013 ; 30 ( 5 ): 469 – 474 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Bryant RA , et al. . Sleep disturbance immediately prior to trauma predicts subsequent psychiatric disorder . Sleep. 2010 ; 33 ( 1 ): 69 – 74 . doi: https://doi.org/10.1093/sleep/33.1.69 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Koren D , et al. . Sleep complaints as early predictors of posttraumatic stress disorder: a 1-year prospective study of injured survivors of motor vehicle accidents . Am J Psychiatry. 2002 ; 159 ( 5 ): 855 – 857 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Garthus-Niegel S , et al. . Maintaining factors of posttraumatic stress symptoms following childbirth: a population-based, two-year follow-up study . J Affect Disord. 2015 ; 172 : 146 – 152 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Geng F , et al. . Sleep problems among adolescent survivors following the 2008 Wenchuan earthquake in China: a cohort study . J Clin Psychiatry. 2013 ; 74 ( 1 ): 67 – 74 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Wright KM , et al. . Insomnia as predictor versus outcome of PTSD and depression among Iraq combat veterans . J Clin Psychol. 2011 ; 67 ( 12 ): 1240 – 1258 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Brady KT , et al. . Comorbidity of psychiatric disorders and posttraumatic stress disorder . J Clin Psychiatry. 2000 ; 61 ( Suppl 7 ): 22 – 32 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Horesh D , et al. . An in-depth look into PTSD-depression comorbidity: a longitudinal study of chronically-exposed Detroit residents . J Affect Disord. 2017 ; 208 : 653 – 661 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Geng F , et al. . Posttraumatic stress disorder and psychiatric comorbidity among adolescent earthquake survivors: a longitudinal cohort study . J Abnorm Child Psychol. 2019 ; 47 ( 4 ): 671 – 681 . Google Scholar Crossref Search ADS PubMed WorldCat 22. Alvaro PK , et al. . A systematic review assessing bidirectionality between sleep disturbances, anxiety, and depression . Sleep. 2013 ; 36 ( 7 ): 1059 – 1068 . doi: https://doi.org/10.5665/sleep.2810 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Breslau N , et al. . Sleep disturbance and psychiatric disorders: a longitudinal epidemiological study of young adults . Biol Psychiatry. 1996 ; 39 ( 6 ): 411 – 418 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Fan F , et al. . Sleep disturbance predicts posttraumatic stress disorder and depressive symptoms: a cohort study of Chinese adolescents . J Clin Psychiatry. 2017 ; 78 ( 7 ): 882 – 888 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Fan F , et al. . ; WEAHC Study Group . Cohort profile: the Wenchuan earthquake adolescent health cohort study . Int J Epidemiol. 2017 ; 46 ( 1 ): 27 – 28 . Google Scholar PubMed WorldCat 26. Geng F , et al. . Prospective associations between sleep problems and subtypes of anxiety symptoms among disaster-exposed adolescents . Sleep Med. 2018 ; 50 : 7 – 13 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Fan F , et al. . Longitudinal trajectories of post-traumatic stress disorder symptoms among adolescents after the Wenchuan earthquake in China . Psychol Med. 2015 ; 45 ( 13 ): 2885 – 2896 . Google Scholar Crossref Search ADS PubMed WorldCat 28. Buysse DJ , et al. . The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research . Psychiatry Res. 1989 ; 28 ( 2 ): 193 – 213 . Google Scholar Crossref Search ADS PubMed WorldCat 29. Liu X , et al. . Insomnia and hypersomnia associated with depressive phenomenology and comorbidity in childhood depression . Sleep. 2007 ; 30 ( 1 ): 83 – 90 . doi: https://doi.org/10.1093/sleep/30.1.83 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Liu X , et al. . Sleep patterns and problems among Chinese adolescents . Pediatrics. 2008 ; 121 ( 6 ): 1165 – 1173 . Google Scholar Crossref Search ADS PubMed WorldCat 31. Liu XC , et al. . Development of the post-traumatic stress disorder self-rating scale and its reliability and validity . Chin J Behav Med Sci . 1998 ; 7 : 93 – 96 . WorldCat 32. Fan F , et al. . Symptoms of posttraumatic stress disorder, depression, and anxiety among adolescents following the 2008 Wenchuan earthquake in China . J Trauma Stress. 2011 ; 24 ( 1 ): 44 – 53 . Google Scholar Crossref Search ADS PubMed WorldCat 33. Birleson P . The validity of depressive disorder in childhood and the development of a self-rating scale: a research report . J Child Psychol Psychiatry. 1981 ; 22 ( 1 ): 73 – 88 . Google Scholar Crossref Search ADS PubMed WorldCat 34. Su LY , et al. Norm of the depression selfrating scale for children in Chinese urban children . Chin Ment Health J . 2003 ; 17 ( 8 ): 547 – 549 . WorldCat 35. Zeger SL , et al. . Longitudinal data analysis for discrete and continuous outcomes . Biometrics. 1986 ; 42 ( 1 ): 121 – 130 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Preacher KJ , et al. . SPSS and SAS procedures for estimating indirect effects in simple mediation models . Behav Res Methods Instrum Comput. 2004 ; 36 ( 4 ): 717 – 731 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Baron RM , et al. . The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations . J Pers Soc Psychol. 1986 ; 51 ( 6 ): 1173 – 1182 . Google Scholar Crossref Search ADS PubMed WorldCat 38. Germain A , et al. . Sleep-specific mechanisms underlying posttraumatic stress disorder: integrative review and neurobiological hypotheses . Sleep Med Rev. 2008 ; 12 ( 3 ): 185 – 195 . Google Scholar Crossref Search ADS PubMed WorldCat 39. Walker MP , et al. . Overnight therapy? The role of sleep in emotional brain processing . Psychol Bull. 2009 ; 135 ( 5 ): 731 – 748 . Google Scholar Crossref Search ADS PubMed WorldCat 40. Pace-Schott EF , et al. . Effects of sleep on memory for conditioned fear and fear extinction . Psychol Bull. 2015 ; 141 ( 4 ): 835 – 857 . Google Scholar Crossref Search ADS PubMed WorldCat 41. Short NA , et al. . Predictors of insomnia symptoms and nightmares among individuals with post-traumatic stress disorder: an ecological momentary assessment study . J Sleep Res. 2018 ; 27 ( 1 ): 64 – 72 . Google Scholar Crossref Search ADS PubMed WorldCat 42. Geng F , et al. . A prospective study of psychiatric symptoms among adolescents after the Wenchuan earthquake . J Trauma Stress. 2018 ; 31 ( 4 ): 499 – 508 . Google Scholar Crossref Search ADS PubMed WorldCat 43. Goenjian AK , et al. . A prospective study of posttraumatic stress and depressive reactions among treated and untreated adolescents 5 years after a catastrophic disaster . Am J Psychiatry. 2005 ; 162 ( 12 ): 2302 – 2308 . Google Scholar Crossref Search ADS PubMed WorldCat 44. Woodward E , et al. . Effects of psychotherapies for posttraumatic stress disorder on sleep disturbances: results from a randomized clinical trial . Behav Res Ther. 2017 ; 97 : 75 – 85 . Google Scholar Crossref Search ADS PubMed WorldCat 45. Talbot LS , et al. Cognitive behavioral therapy for insomnia in posttraumatic stress disorder: a randomized controlled trial . Sleep. 2014 ; 37 ( 2 ): 327 – 341 . doi: https://doi.org/10.5665/sleep.3408 . Google Scholar Crossref Search ADS PubMed WorldCat 46. Ho FY , et al. . Cognitive-behavioral therapy for sleep disturbances in treating posttraumatic stress disorder symptoms: a meta-analysis of randomized controlled trials . Clin Psychol Rev. 2016 ; 43 : 90 – 102 . Google Scholar Crossref Search ADS PubMed WorldCat © Sleep Research Society 2019. Published by Oxford University Press on behalf of the Sleep Research Society. All rights reserved. For permissions, please e-mail [email protected]. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)