Real-time Monitoring using a breathalyzer-based eHealth system can identify lapse/relapse patterns in alcohol use disorder Patients

Real-time Monitoring using a breathalyzer-based eHealth system can identify lapse/relapse... Abstract Aim We introduce a new remote real-time breathalyzer-based method for monitoring and early identification of lapse/relapse patterns for alcohol use disorder (AUD) patients using a composite measure of sobriety, the Addiction Monitoring Index (AMI). Methods We constructed AMI from (a) obtained test results and (b) the pattern of ignored tests using data from the first 30 patients starting in the treatment arms of two on-going clinical trials. The patients performed 2–4 scheduled breath alcohol content (BrAC)-tests per day presented as blood alcohol content (BAC) data. In total, 10,973 tests were scheduled, 7743 were performed and 3230 were ignored during 3982 patient days. Results AMI-time profiles could be used to monitor the daily trends of alcohol consumption and detect early signs of lapse and relapses. The pattern of ignored tests correlates with the onset of drinking. AMI correlated with phosphatidyl ethanol (n = 61, F-ratio = 34.6, P < 0.0001, R = −0.61). The recognition of secret drinking could further be improved using a low alcohol detection threshold (BrAC = 0.025 mg/l, BACSwe = 0.05‰ or US = 0.0053g/dl), in addition to the legal Swedish traffic limit (BrAC = 0.1 mg/l, BACSwe = 0.2‰ or US = 0.021 g/dl). Nine out of 10 patients who dropped out from the study showed early risk signs as reflected in the level and pattern in AMI before the actual dropout. Conclusions AMI-time profiles from an eHealth system are useful for monitoring the recovery process and for early identification of lapse/relapse patterns. High-resolution monitoring of sobriety enables new measurement-based treatment methods for proactive personalized long-term relapse prevention and treatment of AUD patients. Clinical Trial Registration The data used for construction of AMI was from two clinical trials approved by the Regional Ethics Committee of Uppsala, Sweden and performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participating subjects. The study was registered at ClinicalTrials.gov (NCT03195894). INTRODUCTION Relapse rate varies between 2 and 90% depending on the post-remittance time interval and the measure used to define occurrence and severity (Miller, 1996; Miller et al., 2001). Even if a lapse/relapse occurs the overall improvement (e.g. fewer heavy drinking days) after treatment is positive (Miller and Sanchez-Craig, 1996) and the view of alcohol use disorder (AUD) as a dichotomous disorder has been heavily criticized (Miller and Sanchez-Craig, 1996). Recovery is usually a continuous process rather than a discrete event and the occurrence of lapses and relapses should be seen as a natural part of the disease (Miller and Sanchez-Craig, 1996; Brandon et al., 2007). Therefore, AUD should be managed as a chronic condition with a proactive aftercare (Dennis and Scott, 2007) to maintain a behavioural change (Brandon et al., 2007). Despite this, at discharge, AUD patients are typically not offered long-term frequent aftercare. Self-reporting is classically used to monitor the recovery of patients with AUD. Due to recall and honesty issues, self-reported data can be of questionable quality (Midanik, 1988; Searles et al., 2002; Del Boca and Darkes, 2003; Bajunirwe et al., 2014). To improve treatment outcome measures, self-reporting has been combined with the detection of various biomarkers for alcohol. Frequent (3–5/week) verification of sobriety performed at the caregiver’s facility is a common but inconvenient method. The typically conducted measurement is a breath alcohol content (BrAC) test, recalculated and interpreted as blood alcohol content (BAC). A no-show-up for testing is typically classified as a sign of a relapse. Other indirect and direct alcohol biomarkers used in clinical practice have recently been reviewed (Neumann and Spies, 2003; Nanau and Neuman, 2015). The highly selective and sensitive phosphatidyl ethanol (PEth) test is currently in extensive clinical use in Sweden. The half-life of PEth is 4.5–12 days, and the high sensitivity enables the detection of daily moderate drinking (Kechagias et al., 2015). Nevertheless, PEth as a biomarker might be insufficient for the detection of a lapse, in particular when the time intervals between laboratory testing and a drinking incidence are long. Combined with other biomarkers with shorter half-lives (e.g. ethyl glucuronide), an objective and good measure of alcohol use and abstinence can be obtained. Thus, biomarkers can be used to determine whether alcohol has been consumed with high selectivity and specificity, but frequent laboratory testing makes the approach unpractical and unrealistic, especially if total abstinence is desired. Because of the above-mentioned accuracy and practical limitations, alternative methods for monitoring of sobriety and drinking patterns are needed. The rapid development in mobile devices (e.g. smart phones) opens new possibilities to monitor alcohol intake (Rose et al., 2012; Gustafson et al., 2014; Quanbeck et al., 2014; Gurvich et al., 2013). Most of the new digital systems address the problems with recall using more frequent sampling of data on alcohol consumption. Still, patients who want to conceal drinking can falsify their self-reports. Remote monitoring of real-alcohol consumption using transdermal sensors (Gurvich et al., 2013; Barnett et al., 2011; Leffingwell et al., 2013) and cellular phone digital breathalyzers (Skipper et al., 2014; Gordon et al., 2017) give objective measures of alcohol use. These methods have only been briefly tested (Barnett et al., 2011; Skipper et al., 2014; Alessi and Petry, 2013). The transdermal sensors have problems with discomfort, stigma, integrity, reliability, false positives and negatives and lack of sensitivity (Gurvich et al., 2013). The fuel cell-based cellular phone systems have higher sensitivity and good integrity, as well as a potential to provide efficient tools for relapse prevention and remote therapy (Quanbeck et al., 2014). The practical limitations with all cellular phone based systems are the relatively long night rest without data. In this paper, we analyse data from an eHealth system where a pocket-size digital fuel cell-based breathalyzer is connected to a cellular phone and a cloud database. The number of scheduled BrAC tests varied between 2 and 4. When developing methods to analyse the raw data from the breathalyzer tests in order to give the best possible indication of the patient’s status, treatment progress and drinking patterns, we found ourselves faced with three important questions: How to interpret an ignored scheduled test? Is this a mistake or an indication of drinking? Can ignored tests and results from performed tests be combined into one single useful measure of the patient’s current sobriety status? How can a breathalyzer-based eHealth system improve treatment and aftercare of patients with AUD? We discuss one way to combine the breathalyzer test results and ignored/missed tests into an Addiction Monitoring Index (AMI). We present how AMI is constructed, and how the eHealth system can be used for monitoring and early identification of secret drinking and lapse/relapse patterns. We compare AMI levels to PEth values and the frequency of dropout from care. METHODS Ethical aspects Data were collected from the first 30 patients enroled in two clinical trials aimed at testing the potential of an eHealth system (TripleA®, Kontigo Care AB, Uppsala, Sweden) as an add-on to monitor and enhance the diagnosis, care and aftercare of alcohol dependent patients. These clinical trials were conducted in line with the ISO 14,155:2011 standard and were approved by the Regional Ethical Review Board of Uppsala, Sweden and all study participants have provided informed consent to participate in the research project prior to inclusion. The trials have been registered at clinicaltrials.gov (NCT03195894). The present article focuses on the new type of information provided by the daily monitoring of alcohol in the applied eHealth system. Therefore, the patients randomized to the control group are not used in this context. The outcome of the clinical trial, as defined in the study protocol, including the effect of the TripleA treatment and the difference between the tested treatments, will be analysed and reported at a later stage. Patient group The group of patients included 19 males (age 38–69, mean 54, SD 9) and 11 females (age 37–62, mean 50, SD 8) who were recruited among: (a) the regular group of patients attending the Department of Addiction Psychiatry at Uppsala University Hospital in Uppsala (N = 19) and (b) the aftercare patients at three different geographical sites of Nämndemansgården (N = 11). Nämndemansgården is a privately held care provider, practising a therapy based on the twelve-step programme. All of Nämndemansgården’s patients had received in-patient therapy at a recovery centre for at least 4 weeks before enroling in the clinical study. Treatment and monitoring Patients in the clinical trials were randomized either to receive conventional care or conventional care plus daily monitoring via the eHealth system. The conventional form of care was diverse, reflecting current practice in the field, ranging from motivation enhancement therapy, cognitive behavioural therapy, medical treatment (acamprosate, disulfiram) and combinations thereof. Most of the patients from the 12-step aftercare programme participated in an aftercare relapse prevention programme (about 4 h per week). Patient participation in the daily monitoring programme ranged from 7 to 325 days by the cut-off date (27 August 2016) for inclusion in the analysis. For the dropout patients, the number of days without data at the end of treatment varied depending on when the caregiver considered the patient as a dropout. In cases with more than 10 consecutive days without measurement data, Day 11 and onwards were excluded. The participants were provided with free of charge dedicated cellphones with a SIM-card connected to the mobile phone net provider Tele2. The system works also with WiFi. The phones and breathalyzer devices were recovered from all participants after leaving the study. The results from the breathalyzer test were reviewed at least once a week. If a test was positive or tests were ignored more than 2 days in a row the caregiver tried to contact the patient by phone. Patients received notifications to their cellular phone before each requested test according to a schedule. We recommended a test schedule where the first/last test was performed as early/late as possible in the morning/evening. The additional test(s) were evenly distributed between the first and last test of the day. The time window for a sobriety test (start time—deadline) should be practical and short (1−2 h). The caregiver could change the time window from a minimum 1 h to a maximum window length which always left one hour between the deadline and the start time of the next test. If the patient did not perform a test before its scheduled deadline a message (SMS and/or e-mail) was sent to the caregiver. The identity was verified with a photo taken during the test. The photos were validated manually by the caregiver and deleted directly after validation. No defined action of an invalid (e.g. too dark) photo was included in the study. The caregiver had the option to reclassify a breathalyzer test result from negative to positive if they judged that the person on the picture was not the right patient. Such a reclassification would decrease the AMI value. No such reclassifications were done in this study. The participants were encouraged to perform a test even if they had consumed alcohol. Immediately after performing a test, the information is displayed for the care provider in the care portal. Data collection TripleA® (Kontigo Care AB) is a CE-marked medical device class I eHealth system consisting of a fuel cell-based pocket-size breathalyzer connected to a cellular phone via Bluetooth, with an app for the patient and a cloud-based care portal for the care provider (Hämäläinen and Andersson, 2016). The breathalyzer was calibrated at 0.1 mg/l. The BrAC-data measured in mg/l was converted into blood alcohol content in permille (‰) BAC by multiplication with 2.0. This corresponds to the US BAC 0.021 g/dl. The care provider used the system to schedule between two and four requested alcohol breath tests per day. The default number of tests was three, but this could be adapted to accommodate the patient’s needs. Each patient was also asked to state an individual goal, either to stop drinking (total sobriety, n = 22) or to reduce alcohol consumption (risk reduction, n = 8). A BAC threshold of 0.2‰ was used for the identification of alcohol consumption. In cases when the BAC value exceeded 0.2‰, the app prompted the patient to conduct another test after 15 min, a B sample. The patient did not get any detailed information of the measured BAC level, only a green (below threshold) or red (above threshold) indicator. A second threshold (>0.05‰) was used to detect secret drinking. The highest BAC-test value each day was defined as the ‘patient-day’ value or MaxBAC. Phosphatidyl ethanol—PEth—(16:0/18:1, in μmol/l) in blood was measured with LC-MS/MS. Twenty-one patients had 1–6 PEth measurements (total n = 61). AMI value the day before sampling of PEth was used for the correlation study. Addiction Monitoring Index The AMI presented here combines the two important components of the patient’s BrAC testing behaviour into one parameter: Test result (BAC value) Ignored tests (i.e. the patient did not perform the test). The AMI value was calculated for each day. First, each scheduled test was analysed according to the following algorithm: If the test was performed and below the threshold it was given the value 100. If an ignored test was preceded by a continuous array of ignored tests exceeding 48 h, it was given a zero value. If an ignored test was preceded by a continuous array of ignored tests exceeding 24 h, it was given the value 33. If a test was ignored while at least one test had been conducted during the past 24 h, it was given value 67. If the test result exceeded the threshold BAC the test was given a zero value. If a test was ignored and the next test exceeded the BAC threshold, the test was given a value of 33. Thus, there is a retrospective scoring of omitted test and a historic AMI must be updated if this situation occurs. Per-test values were then converted to per-day values by selecting the lowest per-test value for that day. AMI is the exponentially smoothed values calculated from the daily values, using simple exponential smoothing with a smoothing factor of 0.21. s0=x0 st=axt+(1−a)st−1,t>0 where a is the smoothing factor, and 0 < a < 1. The AMI value for the current day has been implemented in subsequent versions of the TripleA system, the information was displayed to the caregiver in the web portal as soon as all the day’s tests have passed their deadline. RESULTS The total number of 10,973 tests was scheduled, of which 7743 were performed and 3230 were ignored (70.1% compliance). It was found that 293 BAC tests (3.8%) had a value between 0.05‰ and 0.2‰, while 314 (4.1%) had scores greater than 0.2‰. Ignored tests were 5–10 times more frequent than tests where alcohol was detected, depending on whether 0.05‰ or 0.2‰ is considered as verified alcohol in the breath. The highest alcohol test value each day was selected as the ‘patient-day’ value, resulting in 3276 BAC-test patient-day values. The number of days with all tests ignored was 706 (including the 10 end days for some of the dropouts), giving a total of 3982 patient days and a daily performance rate of 82.1%. We created a composite measure of sobriety where we combined the BAC-test results with the pattern of ignored BAC tests into an exponentially smoothed AMI. AMI varies between 0 and 100 and a high value (~>80) indicates that no alcohol above the defined threshold is detected and that the patient performs the majority of the scheduled tests (Figs 1 and 2). The cause of variation in AMI levels over time can be seen in the shape and colour of symbols used in the graphs. If AMI decreases and the symbol is a green circle, one or more of the measurements during the day have been ignored. Patients can be grouped based on AMI pattern (Figs 1 and 2): patients with no alcohol consumption and high BAC-test compliance (Patients 5, 18); non-detectable alcohol consumption but low compliance (Patients 8, 12, 14, 17, 19–22, 25); high alcohol consumption and fair compliance (Patients 2, 6); and patients with many days of verified elevated BAC and low compliance (Patients 23, 30). Two thirds of the patients have one or more period of 5 consecutive days where all requested tests were ignored (>5 black diamonds in a row). Fig. 1. View largeDownload slide AMI vs. treatment day number. Four patients in (a) and 16 patients in (b). Symbols: green circle, maximal daily BAC (MaxBAC) ≤0.05‰, (0.0053 g/dl); blue triangle, MaxBAC > 0.05‰ but ≤0.2‰ (0.021 g/dl); red square, MaxBAC > 0.2‰ (0.021 g/dl). A black diamond indicates a day with all tests ignored. Fig. 1. View largeDownload slide AMI vs. treatment day number. Four patients in (a) and 16 patients in (b). Symbols: green circle, maximal daily BAC (MaxBAC) ≤0.05‰, (0.0053 g/dl); blue triangle, MaxBAC > 0.05‰ but ≤0.2‰ (0.021 g/dl); red square, MaxBAC > 0.2‰ (0.021 g/dl). A black diamond indicates a day with all tests ignored. Fig. 2. View largeDownload slide AMI vs. treatment day for the 10 dropouts from the study (Symbols as in Fig. 1). Fig. 2. View largeDownload slide AMI vs. treatment day for the 10 dropouts from the study (Symbols as in Fig. 1). Dropout patients often (9 out of 10) had average AMI levels below 80 (Table 1). The only dropout (24) with high AMI (87.8) had 27 days (18.6 % of total) with BAC´s in the 0.05–0.2 ‰ interval (Table 1). The general trend for dropouts were that AMI decreases with time (Fig. 2), except for the two patients (23, 30) who showed extremely high alcohol consumption and with AMI values mainly below 20. Some patients (26–29) showed a rapid decrease in AMI due to a high number of ignored tests. For patient 29, the low AMI valleys are related to verified alcohol consumption. Many patients showed a high degree of periodicity in their ignoring of tests (dropouts, 21, 22 and 25, as well as patients 7, 8, 9, 10, 12 and 15). For the dropouts, already the first few weeks show a negative trend of declining AMI values due to many ignored tests and BAC’s above 0.2 ‰. Table 1. Gender and age of the patients, addiction monitoring index, test performance statistics and blood alcohol content statistics Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 aPatient: gender: M = male, F = Female; age in years. bAMI, mean and standard deviation. cTest rate in % test performed out of scheduled. dTest (no of days): total number of days in treatment, ignored days (number of days), number of days within different blood alcohol content bins. eBAC content in permille: mean, standard deviation, median and IQR, inter quartile range. Table 1. Gender and age of the patients, addiction monitoring index, test performance statistics and blood alcohol content statistics Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 aPatient: gender: M = male, F = Female; age in years. bAMI, mean and standard deviation. cTest rate in % test performed out of scheduled. dTest (no of days): total number of days in treatment, ignored days (number of days), number of days within different blood alcohol content bins. eBAC content in permille: mean, standard deviation, median and IQR, inter quartile range. AMI showed a negative correlation with phosphatidyl ethanol (n = 61, F-ratio 34.6, P < 0.0001, R = −0.61). To depict how AMI can be used for detailed monitoring of the recovery process, we display data of a 6 months’ period for a patient with a pronounced weekend and holiday relapse pattern (Fig. 3). The treatment started with 14 days with verified sobriety (AMI of 100) (Fig. 3 period A). This was followed by three weekend relapses (B) with three positive BACs and three whole days of ignored tests, producing valleys of around AMI 60. PEth was 1.3 μmol/l after the second weekend. The AMI increased to 80 during sober weekdays but declined to 60 due to weekend relapses. A sober 6-week recovery process (C, PEth 0.42 μmol/l) was followed by one light and then multiple heavy weekend relapses (D). PEth was below detection limit (<0.05 μmol/l) in the beginning of period D. During Period D, the test compliance deteriorated and there were 6 consecutive days with ignored tests coinciding with the Swedish midsummer holiday, resulting in an AMI of ~20. In Period E, the ~4 week long summer vacation is characterized by mainly ignored tests and the few positive tests were all in the mornings. The AMI values declined close to zero and alcohol consumption was also verified during weekdays (Mon, Thu). PEth was 4.1 μmol/l 2 days before the patient was hospitalized (F). After detox AMI was in the 30–60 range and drinking now established also during weekdays (G). Fig. 3. View largeDownload slide Time series of AMI, number of BACs performed and weekday drinking pattern for a patient (9) with a pronounced weekend and holiday relapse pattern. The different areas indicate the following: (A), High AMI due to high test performance rate and negative BAC-test results; (B), three consecutive weekends with three verified drinking days (red squares) and 3 days with ignored tests (black diamonds); (C), long period with good test performance rate; (D), a period with shorter and longer periods with ignored tests followed by vacations (E) and hospitalization (F) to break the relapse. In (G), drinking is now also verified during weekdays. BAC´s/day is number of test performed per day out of the three scheduled. The numbers in brackets correspond to four phosphatidyl ethanol values in μmol/l. (Symbols as in Fig. 1). Fig. 3. View largeDownload slide Time series of AMI, number of BACs performed and weekday drinking pattern for a patient (9) with a pronounced weekend and holiday relapse pattern. The different areas indicate the following: (A), High AMI due to high test performance rate and negative BAC-test results; (B), three consecutive weekends with three verified drinking days (red squares) and 3 days with ignored tests (black diamonds); (C), long period with good test performance rate; (D), a period with shorter and longer periods with ignored tests followed by vacations (E) and hospitalization (F) to break the relapse. In (G), drinking is now also verified during weekdays. BAC´s/day is number of test performed per day out of the three scheduled. The numbers in brackets correspond to four phosphatidyl ethanol values in μmol/l. (Symbols as in Fig. 1). DISCUSSION This article describes the use of an eHealth system for monitoring AUD applied in a live setting for the purpose of understanding and defining suitable indicators for lapse and for long-term management of the actual disease. Patient, family and society would benefit tremendously from an eHealth system capable of comprehensibly monitoring the recovery process and alerting health care providers and family when the patient is at elevated risk for relapse. The relatively small cost of using systems like TripleA is likely to be paid by health care providers and social services aiming to reduce the large financial and social burden of AUD. In Sweden, the municipalities are responsible for long-term care and aftercare of AUD patients including the social implications to the family, notably children, and have started to procure the TripleA system for their citizens. Addiction monitoring index, as defined here, is a sobriety indicator based on high-resolution monitoring of alcohol addiction, combining the measurement result and test compliance. To simplify interpretation, AMI was compiled to range between 0 and 100. By relying on a real physical measurement unit (e.g. breathalyzer) and including lack of test compliance in the AMI, intentional deceiving behaviour is tracked. Therefore, AMI appears as a robust indicator of recent confirmed or suspected alcohol use. This was confirmed by the negative correlation between AMI and PEth. Previous studies have shown a strong correlation between PEth and interlock BAC profiles (Marques et al., 2010). The interpretation of a positive BAC-test is straightforward; alcohol has been consumed. The ignored tests can both be a mistake and/or an intentional omission to hide that a lapse/relapse has occurred. It is therefore logical to give a higher weight/influence of a verified alcohol intake on the AMI than for ignored tests. The construction of AMI relates to (a) a balanced use of measured data and ignored tests and (b) the degree of history to weigh into new data points. We hypothesize that patients are predominantly ignoring tests to conceal drinking, and that the impact of ignored tests is tuned to result in the equality of a positive BAC test in 48 h. The smoothening is tailored to identify potential dropouts. It was rare that patients with an average AMI above 80 dropped out. One isolated single-day lapse (from 100 level) results in an AMI of 79 (due to a smoothing factor of 0.21). After a sober day, the level increased to 83.4%. Similarly, a weekend lapse ending in an AMI of ~60 returned to good compliance (AMI > 80) after four sober days (e.g. Period B in Fig. 3). Thus, a quick return to sobriety resulted in a rapid increase in AMI. Weekend drinking combined with ignored test during the week, gave an AMI that does not climb above 80 (Fig. 3, day 87 + 3 weeks). Note that in this case AMI was not influenced by positive BAC´s (all < 0.2 ‰). AMI was high when only single tests are ignored and none exceeded the BAC threshold (Period C in Fig. 3). Overall, in the current cohort, an AMI exceeding 80 at most times appeared as a strong indicator that the patient will remain compliant with the eHealth monitoring system and will remain reasonably sober. In our clinical trial, we used an alcohol detection threshold based on the Swedish legal value of 0.2 ‰, which is used to judge sobriety in traffic. Similar type of frequently collected BAC data has been obtained with the ignition interlock devices and found to strongly correlate with biomarkers and recidivism to driving under influence of alcohol (Marques et al., 2010). To create a second and more sensitive indicator of secret drinking we used a BAC threshold of 0.05 ‰, enabling identification of an additional large number of drinking occasions. For example, the only dropout (patient 24) with high AMI values (87.8) had 27 days with elevated alcohol levels (>0.05 ‰ to ≤0.02 ‰). A recalculation using a 0.05 ‰ BAC threshold gave an AMI value of 72.1. Thus, it will be important to further analyse and adapt the threshold value used for creating the AMI score, and maybe even tailor it to both the individual BAC background and the purpose of the enrolment in treatment. One key question is whether time-resolved AMI profiles can be used for early identification of patterns of lapse and relapse. We showed that analysis of AMI data gives an opportunity to identify lapse and secret drinking before heavy relapses and a final dropout from care. Not surprisingly, positive BACs in the morning and/or weekends are frequent, indicating that social drinking continues even if the patient’s goal is to be fully sober. The same seems true for ignored BACs in the evening, morning and weekends. At the very least, we can conclude that the present AMI is useful for objective follow-up and has the potential to be further developed into an early response system. The main limitation of using breath alcohol measurements for the identification of secret drinking is the long period during night without measurement data. For ethical reasons, the minimum night time interval is around 8 h, but can be 10–14 h depending on the schedule. This means that it is possible to drink after the last test in the evening and still be able to perform a negative test in the morning. However, this type of ‘strategically planned’ drinking often ended up in the omission of first test in the morning and/or the last test in the evening or a positive BAC-test (>0.05 ‰). This limitation and a new method to identify secret drinking and optimal scheduling of eHealth systems will be discussed in an upcoming paper. When needed, AMI can be combined with the biomarker PEth giving a good continuous monitoring of alcohol use. PEth can also be used to distinguish a decline in test performance rate—which might occur after a longer treatment period—from secret drinking. A unique aspect of the continuous AMI values is the ability for the therapist to monitor the patient situation and base therapy meetings on facts. Individual patterns are often easily identified (weekend drinking, secret drinking, sobriety patterns overlapping with custody of children in shared children custody and similar) and support the therapist in understanding the individual challenges for each patient. With higher accuracy than self-reporting and immediate access to recent data, the therapist has a chance to intervene before or shortly after onset of a lapse and thus preventing a relapse. Hence, the immediate access to AMI trending has the potential to alter current clinical practice for AUD care and aftercare. The patients in this study used the eHealth system as an add-on to conventional therapy. Future studies should include and test different methods to predict, prevent and interrupt relapses (Brandon et al., 2007; Dennis and Scott, 2007; Gustafson et al., 2011; Hendershot et al., 2011; Beckjord and Shiffman, 2014; Chih et al., 2014) when early signs are seen in AMI and related data. It is also possible to use the monitoring capabilities of the eHealth system to study the effect of different therapeutic actions (Alessi and Petry, 2013) and/or medications. Future studies may also extend the use of AMI to other addictions (e.g. gambling). CONCLUSIONS The eHealth system was shown to be effective in obtaining detailed, real-time knowledge of the drinking patterns for AUD patients. AMI, which combines test result values and patterns of ignored tests into one single measure, is a useful tool to monitor and display the patient’s progress over time. It was shown that the AMI value, independent from whether AMI was altered as a consequence of positive BrAC tests or from ignored tests, correlate with results from PEth analysis. Healthcare providers will be able to use AMI to detect and prevent lapse and relapse at an early stage. High-resolution monitoring enables development of a new type of efficient, proactive and personalized long-term measurement-based care and aftercare. ACKNOWLEDGEMENTS We thank Gunilla Svedström, Tobias Eriksson, Ingrid Lundberg, Karin Stenborg, Tobias Sahlin and Liselotte Nordness for their help with the clinical study. FUNDING This work was supported by the Swedish governmental agency for innovation systems (Vinnova) (Grant no 2014−03659). CONFLICT OF INTEREST STATEMENT Markku D. Hämäläinen, Andreas Zetterström, Maria Winkvist, Marcus Söderquist are all employees of Kontigo Care AB. Fred Nyberg is member of the scientific advisory committee of Kontigo Care AB. Markku D. Hämäläinen and Karl Andersson are inventors of a patent related to this manuscript. REFERENCES Alessi SM , Petry NM . ( 2013 ) A randomized study of cellphone technology to reinforce alcohol abstinence in the natural environment . Addiction 108 : 900 – 9 . Google Scholar CrossRef Search ADS PubMed Bajunirwe F , Haberer JE , Boum Y , et al. . ( 2014 ) Comparison of self-reported alcohol consumption to phosphatidylethanol measurement among HIV-infected patients initiating antiretroviral treatment in southwestern Uganda . PLoS One 9 : e113152 . Google Scholar CrossRef Search ADS PubMed Barnett NP , Tidey J , Murphy JG , et al. . ( 2011 ) Contingency management for alcohol use reduction: a pilot study using a transdermal alcohol sensor . Drug Alcohol Depend 118 : 391 – 9 . Google Scholar CrossRef Search ADS PubMed Beckjord E , Shiffman S . ( 2014 ) Background for real-time monitoring and intervention related to alcohol use . Alcohol Res 36 : 9 – 18 . Google Scholar PubMed Brandon TH , Virine JI , Litvin EB . ( 2007 ) Relapse and Relapse prevention . Annu Rev Clin Psychol 3 : 257 – 84 . Google Scholar CrossRef Search ADS PubMed Chih M-Y , Patton T , McTavish FM , et al. . ( 2014 ) Predictive modelling of addiction lapses in a mobile health application . J Subst Abuse Treat 46 : 9 – 35 . Google Scholar CrossRef Search ADS Del Boca FK , Darkes J . ( 2003 ) The validity of self-reports of alcohol consumption: state of the science and challenges for research . Addiction 98 : 1 – 12 . Google Scholar CrossRef Search ADS PubMed Dennis M , Scott CK . ( 2007 ) Managing addiction as a chronic condition . Addict Sci Clin Pract 4 : 45 – 55 . Google Scholar CrossRef Search ADS PubMed Gordon A , Jaffe A , McLellan AT , et al. . ( 2017 ) How should remote clinical monitoring be used to treat alcohol use disorders?: initial findings from an expert round table discussion . J Addict Med 11 : 145 – 53 . Google Scholar CrossRef Search ADS PubMed Gurvich EM , Kenna GA , Leggio L . ( 2013 ) Use of novel technology-based techniques to improve alcohol-related outcomes in clinical trials . Alcohol Alcohol 48 : 712 – 9 . Google Scholar CrossRef Search ADS PubMed Gustafson DH , McTavish FM , Chih MY , et al. . ( 2014 ) A smartphone application to support recovery from alcoholism: a randomized clinical trial . JAMA Psychiatry 71 : 566 – 72 . Google Scholar CrossRef Search ADS PubMed Gustafson DH , Shaw BR , Isham A , et al. . ( 2011 ) Explicating an evidence-based, theoretically informed, mobile technology-based system to improve outcomes for people in recovery for alcohol dependence . Subst Use Misuse 46 : 96 – 111 . Google Scholar CrossRef Search ADS PubMed Hendershot CS , Witkiewitz K , George WH , et al. . ( 2011 ) Relapse prevention for addictive behaviors . Subst Abuse Treat Prev Policy 6 : 1 – 17 . Google Scholar CrossRef Search ADS PubMed Hämäläinen MD , Andersson K ( 2016 ) Method and device for estimating a risk of relapse of addictive behaviour. WO2016178617. Kechagias S , Dernroth DN , Blomgren A , et al. . ( 2015 ) Phosphatidylethanol compared with other blood tests as a biomarker of moderate alcohol consumption in healthy volunteers: a prospective randomized study . Alcohol Alcohol 50 : 399 – 406 . Google Scholar CrossRef Search ADS PubMed Leffingwell TR , Cooney NJ , Murphy JG , et al. . ( 2013 ) Continuous objective monitoring of alcohol use: twenty-first century measurement using transdermal sensors . Alcohol Clin Exp Res 37 : 16 – 22 . Google Scholar CrossRef Search ADS PubMed Marques P , Tippetts S , Allen J , et al. . ( 2010 ) Estimating driver risk using alcohol biomarkers, interlock blood alcohol concentration tests and psychometric assessments: initial descriptives . Addiction 105 : 226 – 39 . Google Scholar CrossRef Search ADS PubMed Midanik LT . ( 1988 ) Validy of self-reported alcohol use: a literature review and assessment . Br J Addict 83 : 1019 – 29 . Google Scholar CrossRef Search ADS PubMed Miller WR . ( 1996 ) What is relapse? Fifty ways to leave the wagon . Addiction 91 : S15 – 27 . Google Scholar CrossRef Search ADS PubMed Miller WR , Sanchez-Craig M . ( 1996 ) How to have a high success rate in treatment: advice for evaluators of alcohol programs . Addiction 91 : 779 – 85 . Google Scholar CrossRef Search ADS PubMed Miller WR , Walters ST , Bennett ME . ( 2001 ) How effective is alcoholism treatment in the United States? J Stud Alcohol 62 : 211 – 20 . Google Scholar CrossRef Search ADS PubMed Nanau RM , Neuman MG . ( 2015 ) Biomolecules and biomarkers used in diagnosis of alcohol drinking and in monitoring therapeutic interventions . Biomolecules 5 : 1339 – 85 . Google Scholar CrossRef Search ADS PubMed Neumann T , Spies C . ( 2003 ) Use of biomarkers for alcohol use disorders in clinical practice . Addiction 98 : 81 – 91 . Google Scholar CrossRef Search ADS PubMed Quanbeck A , Chih M-Y , Isham A , et al. . ( 2014 ) Mobile delivery of treatment for alcohol use disorders: a review of the literature . Alcohol Res 36 : 111 – 22 . Google Scholar PubMed Rose GL , Skelly JM , Badger GJ , et al. . ( 2012 ) Interactive voice response for relapse prevention following cognitive-behavioural therapy for alcohol use disorder: a pilot study . Psychology Serv 9 : 174 – 84 . Google Scholar CrossRef Search ADS Searles JS , Helzer JE , Rose GL , et al. . ( 2002 ) Concurrent and retrospective reports of alcohol consumption across 30, 90 and 366 days: interactive voice response compared with the timeline follow back . J Stud Alcohol 63 : 352 – 62 . Google Scholar CrossRef Search ADS PubMed Skipper GE , Thon N , DuPont RL , et al. . ( 2014 ) Cellular photo digital breathalyzer for monitoring alcohol use: a pilot study . Eur Addict Res 20 : 137 – 42 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Alcohol and Alcoholism Oxford University Press

Real-time Monitoring using a breathalyzer-based eHealth system can identify lapse/relapse patterns in alcohol use disorder Patients

Loading next page...
 
/lp/ou_press/real-time-monitoring-using-a-breathalyzer-based-ehealth-system-can-0rJ4NO0po2
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved.
ISSN
0735-0414
eISSN
1464-3502
D.O.I.
10.1093/alcalc/agy011
Publisher site
See Article on Publisher Site

Abstract

Abstract Aim We introduce a new remote real-time breathalyzer-based method for monitoring and early identification of lapse/relapse patterns for alcohol use disorder (AUD) patients using a composite measure of sobriety, the Addiction Monitoring Index (AMI). Methods We constructed AMI from (a) obtained test results and (b) the pattern of ignored tests using data from the first 30 patients starting in the treatment arms of two on-going clinical trials. The patients performed 2–4 scheduled breath alcohol content (BrAC)-tests per day presented as blood alcohol content (BAC) data. In total, 10,973 tests were scheduled, 7743 were performed and 3230 were ignored during 3982 patient days. Results AMI-time profiles could be used to monitor the daily trends of alcohol consumption and detect early signs of lapse and relapses. The pattern of ignored tests correlates with the onset of drinking. AMI correlated with phosphatidyl ethanol (n = 61, F-ratio = 34.6, P < 0.0001, R = −0.61). The recognition of secret drinking could further be improved using a low alcohol detection threshold (BrAC = 0.025 mg/l, BACSwe = 0.05‰ or US = 0.0053g/dl), in addition to the legal Swedish traffic limit (BrAC = 0.1 mg/l, BACSwe = 0.2‰ or US = 0.021 g/dl). Nine out of 10 patients who dropped out from the study showed early risk signs as reflected in the level and pattern in AMI before the actual dropout. Conclusions AMI-time profiles from an eHealth system are useful for monitoring the recovery process and for early identification of lapse/relapse patterns. High-resolution monitoring of sobriety enables new measurement-based treatment methods for proactive personalized long-term relapse prevention and treatment of AUD patients. Clinical Trial Registration The data used for construction of AMI was from two clinical trials approved by the Regional Ethics Committee of Uppsala, Sweden and performed in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participating subjects. The study was registered at ClinicalTrials.gov (NCT03195894). INTRODUCTION Relapse rate varies between 2 and 90% depending on the post-remittance time interval and the measure used to define occurrence and severity (Miller, 1996; Miller et al., 2001). Even if a lapse/relapse occurs the overall improvement (e.g. fewer heavy drinking days) after treatment is positive (Miller and Sanchez-Craig, 1996) and the view of alcohol use disorder (AUD) as a dichotomous disorder has been heavily criticized (Miller and Sanchez-Craig, 1996). Recovery is usually a continuous process rather than a discrete event and the occurrence of lapses and relapses should be seen as a natural part of the disease (Miller and Sanchez-Craig, 1996; Brandon et al., 2007). Therefore, AUD should be managed as a chronic condition with a proactive aftercare (Dennis and Scott, 2007) to maintain a behavioural change (Brandon et al., 2007). Despite this, at discharge, AUD patients are typically not offered long-term frequent aftercare. Self-reporting is classically used to monitor the recovery of patients with AUD. Due to recall and honesty issues, self-reported data can be of questionable quality (Midanik, 1988; Searles et al., 2002; Del Boca and Darkes, 2003; Bajunirwe et al., 2014). To improve treatment outcome measures, self-reporting has been combined with the detection of various biomarkers for alcohol. Frequent (3–5/week) verification of sobriety performed at the caregiver’s facility is a common but inconvenient method. The typically conducted measurement is a breath alcohol content (BrAC) test, recalculated and interpreted as blood alcohol content (BAC). A no-show-up for testing is typically classified as a sign of a relapse. Other indirect and direct alcohol biomarkers used in clinical practice have recently been reviewed (Neumann and Spies, 2003; Nanau and Neuman, 2015). The highly selective and sensitive phosphatidyl ethanol (PEth) test is currently in extensive clinical use in Sweden. The half-life of PEth is 4.5–12 days, and the high sensitivity enables the detection of daily moderate drinking (Kechagias et al., 2015). Nevertheless, PEth as a biomarker might be insufficient for the detection of a lapse, in particular when the time intervals between laboratory testing and a drinking incidence are long. Combined with other biomarkers with shorter half-lives (e.g. ethyl glucuronide), an objective and good measure of alcohol use and abstinence can be obtained. Thus, biomarkers can be used to determine whether alcohol has been consumed with high selectivity and specificity, but frequent laboratory testing makes the approach unpractical and unrealistic, especially if total abstinence is desired. Because of the above-mentioned accuracy and practical limitations, alternative methods for monitoring of sobriety and drinking patterns are needed. The rapid development in mobile devices (e.g. smart phones) opens new possibilities to monitor alcohol intake (Rose et al., 2012; Gustafson et al., 2014; Quanbeck et al., 2014; Gurvich et al., 2013). Most of the new digital systems address the problems with recall using more frequent sampling of data on alcohol consumption. Still, patients who want to conceal drinking can falsify their self-reports. Remote monitoring of real-alcohol consumption using transdermal sensors (Gurvich et al., 2013; Barnett et al., 2011; Leffingwell et al., 2013) and cellular phone digital breathalyzers (Skipper et al., 2014; Gordon et al., 2017) give objective measures of alcohol use. These methods have only been briefly tested (Barnett et al., 2011; Skipper et al., 2014; Alessi and Petry, 2013). The transdermal sensors have problems with discomfort, stigma, integrity, reliability, false positives and negatives and lack of sensitivity (Gurvich et al., 2013). The fuel cell-based cellular phone systems have higher sensitivity and good integrity, as well as a potential to provide efficient tools for relapse prevention and remote therapy (Quanbeck et al., 2014). The practical limitations with all cellular phone based systems are the relatively long night rest without data. In this paper, we analyse data from an eHealth system where a pocket-size digital fuel cell-based breathalyzer is connected to a cellular phone and a cloud database. The number of scheduled BrAC tests varied between 2 and 4. When developing methods to analyse the raw data from the breathalyzer tests in order to give the best possible indication of the patient’s status, treatment progress and drinking patterns, we found ourselves faced with three important questions: How to interpret an ignored scheduled test? Is this a mistake or an indication of drinking? Can ignored tests and results from performed tests be combined into one single useful measure of the patient’s current sobriety status? How can a breathalyzer-based eHealth system improve treatment and aftercare of patients with AUD? We discuss one way to combine the breathalyzer test results and ignored/missed tests into an Addiction Monitoring Index (AMI). We present how AMI is constructed, and how the eHealth system can be used for monitoring and early identification of secret drinking and lapse/relapse patterns. We compare AMI levels to PEth values and the frequency of dropout from care. METHODS Ethical aspects Data were collected from the first 30 patients enroled in two clinical trials aimed at testing the potential of an eHealth system (TripleA®, Kontigo Care AB, Uppsala, Sweden) as an add-on to monitor and enhance the diagnosis, care and aftercare of alcohol dependent patients. These clinical trials were conducted in line with the ISO 14,155:2011 standard and were approved by the Regional Ethical Review Board of Uppsala, Sweden and all study participants have provided informed consent to participate in the research project prior to inclusion. The trials have been registered at clinicaltrials.gov (NCT03195894). The present article focuses on the new type of information provided by the daily monitoring of alcohol in the applied eHealth system. Therefore, the patients randomized to the control group are not used in this context. The outcome of the clinical trial, as defined in the study protocol, including the effect of the TripleA treatment and the difference between the tested treatments, will be analysed and reported at a later stage. Patient group The group of patients included 19 males (age 38–69, mean 54, SD 9) and 11 females (age 37–62, mean 50, SD 8) who were recruited among: (a) the regular group of patients attending the Department of Addiction Psychiatry at Uppsala University Hospital in Uppsala (N = 19) and (b) the aftercare patients at three different geographical sites of Nämndemansgården (N = 11). Nämndemansgården is a privately held care provider, practising a therapy based on the twelve-step programme. All of Nämndemansgården’s patients had received in-patient therapy at a recovery centre for at least 4 weeks before enroling in the clinical study. Treatment and monitoring Patients in the clinical trials were randomized either to receive conventional care or conventional care plus daily monitoring via the eHealth system. The conventional form of care was diverse, reflecting current practice in the field, ranging from motivation enhancement therapy, cognitive behavioural therapy, medical treatment (acamprosate, disulfiram) and combinations thereof. Most of the patients from the 12-step aftercare programme participated in an aftercare relapse prevention programme (about 4 h per week). Patient participation in the daily monitoring programme ranged from 7 to 325 days by the cut-off date (27 August 2016) for inclusion in the analysis. For the dropout patients, the number of days without data at the end of treatment varied depending on when the caregiver considered the patient as a dropout. In cases with more than 10 consecutive days without measurement data, Day 11 and onwards were excluded. The participants were provided with free of charge dedicated cellphones with a SIM-card connected to the mobile phone net provider Tele2. The system works also with WiFi. The phones and breathalyzer devices were recovered from all participants after leaving the study. The results from the breathalyzer test were reviewed at least once a week. If a test was positive or tests were ignored more than 2 days in a row the caregiver tried to contact the patient by phone. Patients received notifications to their cellular phone before each requested test according to a schedule. We recommended a test schedule where the first/last test was performed as early/late as possible in the morning/evening. The additional test(s) were evenly distributed between the first and last test of the day. The time window for a sobriety test (start time—deadline) should be practical and short (1−2 h). The caregiver could change the time window from a minimum 1 h to a maximum window length which always left one hour between the deadline and the start time of the next test. If the patient did not perform a test before its scheduled deadline a message (SMS and/or e-mail) was sent to the caregiver. The identity was verified with a photo taken during the test. The photos were validated manually by the caregiver and deleted directly after validation. No defined action of an invalid (e.g. too dark) photo was included in the study. The caregiver had the option to reclassify a breathalyzer test result from negative to positive if they judged that the person on the picture was not the right patient. Such a reclassification would decrease the AMI value. No such reclassifications were done in this study. The participants were encouraged to perform a test even if they had consumed alcohol. Immediately after performing a test, the information is displayed for the care provider in the care portal. Data collection TripleA® (Kontigo Care AB) is a CE-marked medical device class I eHealth system consisting of a fuel cell-based pocket-size breathalyzer connected to a cellular phone via Bluetooth, with an app for the patient and a cloud-based care portal for the care provider (Hämäläinen and Andersson, 2016). The breathalyzer was calibrated at 0.1 mg/l. The BrAC-data measured in mg/l was converted into blood alcohol content in permille (‰) BAC by multiplication with 2.0. This corresponds to the US BAC 0.021 g/dl. The care provider used the system to schedule between two and four requested alcohol breath tests per day. The default number of tests was three, but this could be adapted to accommodate the patient’s needs. Each patient was also asked to state an individual goal, either to stop drinking (total sobriety, n = 22) or to reduce alcohol consumption (risk reduction, n = 8). A BAC threshold of 0.2‰ was used for the identification of alcohol consumption. In cases when the BAC value exceeded 0.2‰, the app prompted the patient to conduct another test after 15 min, a B sample. The patient did not get any detailed information of the measured BAC level, only a green (below threshold) or red (above threshold) indicator. A second threshold (>0.05‰) was used to detect secret drinking. The highest BAC-test value each day was defined as the ‘patient-day’ value or MaxBAC. Phosphatidyl ethanol—PEth—(16:0/18:1, in μmol/l) in blood was measured with LC-MS/MS. Twenty-one patients had 1–6 PEth measurements (total n = 61). AMI value the day before sampling of PEth was used for the correlation study. Addiction Monitoring Index The AMI presented here combines the two important components of the patient’s BrAC testing behaviour into one parameter: Test result (BAC value) Ignored tests (i.e. the patient did not perform the test). The AMI value was calculated for each day. First, each scheduled test was analysed according to the following algorithm: If the test was performed and below the threshold it was given the value 100. If an ignored test was preceded by a continuous array of ignored tests exceeding 48 h, it was given a zero value. If an ignored test was preceded by a continuous array of ignored tests exceeding 24 h, it was given the value 33. If a test was ignored while at least one test had been conducted during the past 24 h, it was given value 67. If the test result exceeded the threshold BAC the test was given a zero value. If a test was ignored and the next test exceeded the BAC threshold, the test was given a value of 33. Thus, there is a retrospective scoring of omitted test and a historic AMI must be updated if this situation occurs. Per-test values were then converted to per-day values by selecting the lowest per-test value for that day. AMI is the exponentially smoothed values calculated from the daily values, using simple exponential smoothing with a smoothing factor of 0.21. s0=x0 st=axt+(1−a)st−1,t>0 where a is the smoothing factor, and 0 < a < 1. The AMI value for the current day has been implemented in subsequent versions of the TripleA system, the information was displayed to the caregiver in the web portal as soon as all the day’s tests have passed their deadline. RESULTS The total number of 10,973 tests was scheduled, of which 7743 were performed and 3230 were ignored (70.1% compliance). It was found that 293 BAC tests (3.8%) had a value between 0.05‰ and 0.2‰, while 314 (4.1%) had scores greater than 0.2‰. Ignored tests were 5–10 times more frequent than tests where alcohol was detected, depending on whether 0.05‰ or 0.2‰ is considered as verified alcohol in the breath. The highest alcohol test value each day was selected as the ‘patient-day’ value, resulting in 3276 BAC-test patient-day values. The number of days with all tests ignored was 706 (including the 10 end days for some of the dropouts), giving a total of 3982 patient days and a daily performance rate of 82.1%. We created a composite measure of sobriety where we combined the BAC-test results with the pattern of ignored BAC tests into an exponentially smoothed AMI. AMI varies between 0 and 100 and a high value (~>80) indicates that no alcohol above the defined threshold is detected and that the patient performs the majority of the scheduled tests (Figs 1 and 2). The cause of variation in AMI levels over time can be seen in the shape and colour of symbols used in the graphs. If AMI decreases and the symbol is a green circle, one or more of the measurements during the day have been ignored. Patients can be grouped based on AMI pattern (Figs 1 and 2): patients with no alcohol consumption and high BAC-test compliance (Patients 5, 18); non-detectable alcohol consumption but low compliance (Patients 8, 12, 14, 17, 19–22, 25); high alcohol consumption and fair compliance (Patients 2, 6); and patients with many days of verified elevated BAC and low compliance (Patients 23, 30). Two thirds of the patients have one or more period of 5 consecutive days where all requested tests were ignored (>5 black diamonds in a row). Fig. 1. View largeDownload slide AMI vs. treatment day number. Four patients in (a) and 16 patients in (b). Symbols: green circle, maximal daily BAC (MaxBAC) ≤0.05‰, (0.0053 g/dl); blue triangle, MaxBAC > 0.05‰ but ≤0.2‰ (0.021 g/dl); red square, MaxBAC > 0.2‰ (0.021 g/dl). A black diamond indicates a day with all tests ignored. Fig. 1. View largeDownload slide AMI vs. treatment day number. Four patients in (a) and 16 patients in (b). Symbols: green circle, maximal daily BAC (MaxBAC) ≤0.05‰, (0.0053 g/dl); blue triangle, MaxBAC > 0.05‰ but ≤0.2‰ (0.021 g/dl); red square, MaxBAC > 0.2‰ (0.021 g/dl). A black diamond indicates a day with all tests ignored. Fig. 2. View largeDownload slide AMI vs. treatment day for the 10 dropouts from the study (Symbols as in Fig. 1). Fig. 2. View largeDownload slide AMI vs. treatment day for the 10 dropouts from the study (Symbols as in Fig. 1). Dropout patients often (9 out of 10) had average AMI levels below 80 (Table 1). The only dropout (24) with high AMI (87.8) had 27 days (18.6 % of total) with BAC´s in the 0.05–0.2 ‰ interval (Table 1). The general trend for dropouts were that AMI decreases with time (Fig. 2), except for the two patients (23, 30) who showed extremely high alcohol consumption and with AMI values mainly below 20. Some patients (26–29) showed a rapid decrease in AMI due to a high number of ignored tests. For patient 29, the low AMI valleys are related to verified alcohol consumption. Many patients showed a high degree of periodicity in their ignoring of tests (dropouts, 21, 22 and 25, as well as patients 7, 8, 9, 10, 12 and 15). For the dropouts, already the first few weeks show a negative trend of declining AMI values due to many ignored tests and BAC’s above 0.2 ‰. Table 1. Gender and age of the patients, addiction monitoring index, test performance statistics and blood alcohol content statistics Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 aPatient: gender: M = male, F = Female; age in years. bAMI, mean and standard deviation. cTest rate in % test performed out of scheduled. dTest (no of days): total number of days in treatment, ignored days (number of days), number of days within different blood alcohol content bins. eBAC content in permille: mean, standard deviation, median and IQR, inter quartile range. Table 1. Gender and age of the patients, addiction monitoring index, test performance statistics and blood alcohol content statistics Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 Patienta AMIb Testc Test (no of days)d BAC (‰)e ID (number) Gender Age Mean SD Performed (%) Total number of days Ignored days BAC ≤ 0.05 ‰ BAC > 0.05. ≤ 0.2 ‰ BAC > 0.2 ‰ Mean SD Median IQR 1 M 50 90.7 10 89.5 325 8 225 92 0 0.018 0.029 0.006 0.011 2 M 69 51.2 14 79.1 306 14 141 40 111 0.126 0.346 0.004 0.030 3 F 61 75.6 21 69.8 270 28 233 9 0 0.008 0.011 0.006 0.002 4 M 45 46.8 28 39.5 267 110 149 2 6 0.028 0.090 0.008 0.004 5 M 47 99 3 99.2 191 0 190 1 0 0.004 0.004 0.004 0.000 6 F 62 69.6 14 85.8 190 0 134 19 37 0.072 0.261 0.006 0.004 7 M 67 93.9 8 95.8 190 3 174 10 3 0.011 0.041 0.004 0.002 8 M 62 50.3 28 44.7 177 72 99 6 0 0.013 0.018 0.008 0.006 9 M 56 64.9 30 64.7 170 44 111 5 10 0.039 0.098 0.011 0.006 10 M 52 75.8 23 75.6 149 25 113 5 6 0.019 0.068 0.008 0.006 11 F 57 86.7 20 84.6 145 16 128 1 0 0.006 0.005 0.006 0.004 12 F 45 82 17 78.9 128 16 111 1 0 0.006 0.005 0.006 0.002 13 M 44 34.3 15 31.2 102 48 51 2 1 0.011 0.045 0.006 0.004 14 M 47 70.3 13 60.9 110 11 97 2 0 0.010 0.012 0.008 0.004 15 F 51 35.5 23 30.9 109 63 42 3 1 0.017 0.042 0.008 0.004 16 M 64 85.9 19 85.3 102 12 83 5 2 0.026 0.088 0.008 0.006 17 F 57 25.8 29 25.6 88 59 26 3 0 0.008 0.018 0.004 0.000 18 M 60 98.1 3 98.6 72 0 72 0 0 0.006 0.002 0.006 0.002 19 F 44 49.2 31 49.2 61 25 36 0 0 0.005 0.005 0.004 0.002 20 M 65 83.6 13 85.6 59 3 56 0 0 0.005 0.002 0.004 0.002 21 F 37 67.2 23 60 128 34 93 1 0 0.007 0.010 0.006 0.002 22 F 46 76.9 22 74.8 157 25 130 2 0 0.005 0.008 0.004 0.000 23 M 48 13.5 17 58.2 47 8 6 0 33 1.033 0.933 0.783 1.691 24 M 50 87.8 9 89.2 145 2 114 27 2 0.023 0.110 0.008 0.011 25 F 43 66 12 53.5 53 16 34 3 0 0.013 0.016 0.008 0.004 26 M 38 64.4 30 55.7 73 21 52 0 0 0.006 0.005 0.004 0.002 27 F 52 64.2 25 56.8 27 8 19 0 0 0.008 0.005 0.006 0.004 28 M 45 47.1 36 41.2 17 10 7 0 0 0.005 0.002 0.006 0.002 29 M 55 65.3 24 68.1 72 11 52 3 6 0.061 0.163 0.008 0.006 30 M 56 6.4 7 59.6 52 14 2 1 35 0.673 0.680 0.548 1.042 aPatient: gender: M = male, F = Female; age in years. bAMI, mean and standard deviation. cTest rate in % test performed out of scheduled. dTest (no of days): total number of days in treatment, ignored days (number of days), number of days within different blood alcohol content bins. eBAC content in permille: mean, standard deviation, median and IQR, inter quartile range. AMI showed a negative correlation with phosphatidyl ethanol (n = 61, F-ratio 34.6, P < 0.0001, R = −0.61). To depict how AMI can be used for detailed monitoring of the recovery process, we display data of a 6 months’ period for a patient with a pronounced weekend and holiday relapse pattern (Fig. 3). The treatment started with 14 days with verified sobriety (AMI of 100) (Fig. 3 period A). This was followed by three weekend relapses (B) with three positive BACs and three whole days of ignored tests, producing valleys of around AMI 60. PEth was 1.3 μmol/l after the second weekend. The AMI increased to 80 during sober weekdays but declined to 60 due to weekend relapses. A sober 6-week recovery process (C, PEth 0.42 μmol/l) was followed by one light and then multiple heavy weekend relapses (D). PEth was below detection limit (<0.05 μmol/l) in the beginning of period D. During Period D, the test compliance deteriorated and there were 6 consecutive days with ignored tests coinciding with the Swedish midsummer holiday, resulting in an AMI of ~20. In Period E, the ~4 week long summer vacation is characterized by mainly ignored tests and the few positive tests were all in the mornings. The AMI values declined close to zero and alcohol consumption was also verified during weekdays (Mon, Thu). PEth was 4.1 μmol/l 2 days before the patient was hospitalized (F). After detox AMI was in the 30–60 range and drinking now established also during weekdays (G). Fig. 3. View largeDownload slide Time series of AMI, number of BACs performed and weekday drinking pattern for a patient (9) with a pronounced weekend and holiday relapse pattern. The different areas indicate the following: (A), High AMI due to high test performance rate and negative BAC-test results; (B), three consecutive weekends with three verified drinking days (red squares) and 3 days with ignored tests (black diamonds); (C), long period with good test performance rate; (D), a period with shorter and longer periods with ignored tests followed by vacations (E) and hospitalization (F) to break the relapse. In (G), drinking is now also verified during weekdays. BAC´s/day is number of test performed per day out of the three scheduled. The numbers in brackets correspond to four phosphatidyl ethanol values in μmol/l. (Symbols as in Fig. 1). Fig. 3. View largeDownload slide Time series of AMI, number of BACs performed and weekday drinking pattern for a patient (9) with a pronounced weekend and holiday relapse pattern. The different areas indicate the following: (A), High AMI due to high test performance rate and negative BAC-test results; (B), three consecutive weekends with three verified drinking days (red squares) and 3 days with ignored tests (black diamonds); (C), long period with good test performance rate; (D), a period with shorter and longer periods with ignored tests followed by vacations (E) and hospitalization (F) to break the relapse. In (G), drinking is now also verified during weekdays. BAC´s/day is number of test performed per day out of the three scheduled. The numbers in brackets correspond to four phosphatidyl ethanol values in μmol/l. (Symbols as in Fig. 1). DISCUSSION This article describes the use of an eHealth system for monitoring AUD applied in a live setting for the purpose of understanding and defining suitable indicators for lapse and for long-term management of the actual disease. Patient, family and society would benefit tremendously from an eHealth system capable of comprehensibly monitoring the recovery process and alerting health care providers and family when the patient is at elevated risk for relapse. The relatively small cost of using systems like TripleA is likely to be paid by health care providers and social services aiming to reduce the large financial and social burden of AUD. In Sweden, the municipalities are responsible for long-term care and aftercare of AUD patients including the social implications to the family, notably children, and have started to procure the TripleA system for their citizens. Addiction monitoring index, as defined here, is a sobriety indicator based on high-resolution monitoring of alcohol addiction, combining the measurement result and test compliance. To simplify interpretation, AMI was compiled to range between 0 and 100. By relying on a real physical measurement unit (e.g. breathalyzer) and including lack of test compliance in the AMI, intentional deceiving behaviour is tracked. Therefore, AMI appears as a robust indicator of recent confirmed or suspected alcohol use. This was confirmed by the negative correlation between AMI and PEth. Previous studies have shown a strong correlation between PEth and interlock BAC profiles (Marques et al., 2010). The interpretation of a positive BAC-test is straightforward; alcohol has been consumed. The ignored tests can both be a mistake and/or an intentional omission to hide that a lapse/relapse has occurred. It is therefore logical to give a higher weight/influence of a verified alcohol intake on the AMI than for ignored tests. The construction of AMI relates to (a) a balanced use of measured data and ignored tests and (b) the degree of history to weigh into new data points. We hypothesize that patients are predominantly ignoring tests to conceal drinking, and that the impact of ignored tests is tuned to result in the equality of a positive BAC test in 48 h. The smoothening is tailored to identify potential dropouts. It was rare that patients with an average AMI above 80 dropped out. One isolated single-day lapse (from 100 level) results in an AMI of 79 (due to a smoothing factor of 0.21). After a sober day, the level increased to 83.4%. Similarly, a weekend lapse ending in an AMI of ~60 returned to good compliance (AMI > 80) after four sober days (e.g. Period B in Fig. 3). Thus, a quick return to sobriety resulted in a rapid increase in AMI. Weekend drinking combined with ignored test during the week, gave an AMI that does not climb above 80 (Fig. 3, day 87 + 3 weeks). Note that in this case AMI was not influenced by positive BAC´s (all < 0.2 ‰). AMI was high when only single tests are ignored and none exceeded the BAC threshold (Period C in Fig. 3). Overall, in the current cohort, an AMI exceeding 80 at most times appeared as a strong indicator that the patient will remain compliant with the eHealth monitoring system and will remain reasonably sober. In our clinical trial, we used an alcohol detection threshold based on the Swedish legal value of 0.2 ‰, which is used to judge sobriety in traffic. Similar type of frequently collected BAC data has been obtained with the ignition interlock devices and found to strongly correlate with biomarkers and recidivism to driving under influence of alcohol (Marques et al., 2010). To create a second and more sensitive indicator of secret drinking we used a BAC threshold of 0.05 ‰, enabling identification of an additional large number of drinking occasions. For example, the only dropout (patient 24) with high AMI values (87.8) had 27 days with elevated alcohol levels (>0.05 ‰ to ≤0.02 ‰). A recalculation using a 0.05 ‰ BAC threshold gave an AMI value of 72.1. Thus, it will be important to further analyse and adapt the threshold value used for creating the AMI score, and maybe even tailor it to both the individual BAC background and the purpose of the enrolment in treatment. One key question is whether time-resolved AMI profiles can be used for early identification of patterns of lapse and relapse. We showed that analysis of AMI data gives an opportunity to identify lapse and secret drinking before heavy relapses and a final dropout from care. Not surprisingly, positive BACs in the morning and/or weekends are frequent, indicating that social drinking continues even if the patient’s goal is to be fully sober. The same seems true for ignored BACs in the evening, morning and weekends. At the very least, we can conclude that the present AMI is useful for objective follow-up and has the potential to be further developed into an early response system. The main limitation of using breath alcohol measurements for the identification of secret drinking is the long period during night without measurement data. For ethical reasons, the minimum night time interval is around 8 h, but can be 10–14 h depending on the schedule. This means that it is possible to drink after the last test in the evening and still be able to perform a negative test in the morning. However, this type of ‘strategically planned’ drinking often ended up in the omission of first test in the morning and/or the last test in the evening or a positive BAC-test (>0.05 ‰). This limitation and a new method to identify secret drinking and optimal scheduling of eHealth systems will be discussed in an upcoming paper. When needed, AMI can be combined with the biomarker PEth giving a good continuous monitoring of alcohol use. PEth can also be used to distinguish a decline in test performance rate—which might occur after a longer treatment period—from secret drinking. A unique aspect of the continuous AMI values is the ability for the therapist to monitor the patient situation and base therapy meetings on facts. Individual patterns are often easily identified (weekend drinking, secret drinking, sobriety patterns overlapping with custody of children in shared children custody and similar) and support the therapist in understanding the individual challenges for each patient. With higher accuracy than self-reporting and immediate access to recent data, the therapist has a chance to intervene before or shortly after onset of a lapse and thus preventing a relapse. Hence, the immediate access to AMI trending has the potential to alter current clinical practice for AUD care and aftercare. The patients in this study used the eHealth system as an add-on to conventional therapy. Future studies should include and test different methods to predict, prevent and interrupt relapses (Brandon et al., 2007; Dennis and Scott, 2007; Gustafson et al., 2011; Hendershot et al., 2011; Beckjord and Shiffman, 2014; Chih et al., 2014) when early signs are seen in AMI and related data. It is also possible to use the monitoring capabilities of the eHealth system to study the effect of different therapeutic actions (Alessi and Petry, 2013) and/or medications. Future studies may also extend the use of AMI to other addictions (e.g. gambling). CONCLUSIONS The eHealth system was shown to be effective in obtaining detailed, real-time knowledge of the drinking patterns for AUD patients. AMI, which combines test result values and patterns of ignored tests into one single measure, is a useful tool to monitor and display the patient’s progress over time. It was shown that the AMI value, independent from whether AMI was altered as a consequence of positive BrAC tests or from ignored tests, correlate with results from PEth analysis. Healthcare providers will be able to use AMI to detect and prevent lapse and relapse at an early stage. High-resolution monitoring enables development of a new type of efficient, proactive and personalized long-term measurement-based care and aftercare. ACKNOWLEDGEMENTS We thank Gunilla Svedström, Tobias Eriksson, Ingrid Lundberg, Karin Stenborg, Tobias Sahlin and Liselotte Nordness for their help with the clinical study. FUNDING This work was supported by the Swedish governmental agency for innovation systems (Vinnova) (Grant no 2014−03659). CONFLICT OF INTEREST STATEMENT Markku D. Hämäläinen, Andreas Zetterström, Maria Winkvist, Marcus Söderquist are all employees of Kontigo Care AB. Fred Nyberg is member of the scientific advisory committee of Kontigo Care AB. Markku D. Hämäläinen and Karl Andersson are inventors of a patent related to this manuscript. REFERENCES Alessi SM , Petry NM . ( 2013 ) A randomized study of cellphone technology to reinforce alcohol abstinence in the natural environment . Addiction 108 : 900 – 9 . Google Scholar CrossRef Search ADS PubMed Bajunirwe F , Haberer JE , Boum Y , et al. . ( 2014 ) Comparison of self-reported alcohol consumption to phosphatidylethanol measurement among HIV-infected patients initiating antiretroviral treatment in southwestern Uganda . PLoS One 9 : e113152 . Google Scholar CrossRef Search ADS PubMed Barnett NP , Tidey J , Murphy JG , et al. . ( 2011 ) Contingency management for alcohol use reduction: a pilot study using a transdermal alcohol sensor . Drug Alcohol Depend 118 : 391 – 9 . Google Scholar CrossRef Search ADS PubMed Beckjord E , Shiffman S . ( 2014 ) Background for real-time monitoring and intervention related to alcohol use . Alcohol Res 36 : 9 – 18 . Google Scholar PubMed Brandon TH , Virine JI , Litvin EB . ( 2007 ) Relapse and Relapse prevention . Annu Rev Clin Psychol 3 : 257 – 84 . Google Scholar CrossRef Search ADS PubMed Chih M-Y , Patton T , McTavish FM , et al. . ( 2014 ) Predictive modelling of addiction lapses in a mobile health application . J Subst Abuse Treat 46 : 9 – 35 . Google Scholar CrossRef Search ADS Del Boca FK , Darkes J . ( 2003 ) The validity of self-reports of alcohol consumption: state of the science and challenges for research . Addiction 98 : 1 – 12 . Google Scholar CrossRef Search ADS PubMed Dennis M , Scott CK . ( 2007 ) Managing addiction as a chronic condition . Addict Sci Clin Pract 4 : 45 – 55 . Google Scholar CrossRef Search ADS PubMed Gordon A , Jaffe A , McLellan AT , et al. . ( 2017 ) How should remote clinical monitoring be used to treat alcohol use disorders?: initial findings from an expert round table discussion . J Addict Med 11 : 145 – 53 . Google Scholar CrossRef Search ADS PubMed Gurvich EM , Kenna GA , Leggio L . ( 2013 ) Use of novel technology-based techniques to improve alcohol-related outcomes in clinical trials . Alcohol Alcohol 48 : 712 – 9 . Google Scholar CrossRef Search ADS PubMed Gustafson DH , McTavish FM , Chih MY , et al. . ( 2014 ) A smartphone application to support recovery from alcoholism: a randomized clinical trial . JAMA Psychiatry 71 : 566 – 72 . Google Scholar CrossRef Search ADS PubMed Gustafson DH , Shaw BR , Isham A , et al. . ( 2011 ) Explicating an evidence-based, theoretically informed, mobile technology-based system to improve outcomes for people in recovery for alcohol dependence . Subst Use Misuse 46 : 96 – 111 . Google Scholar CrossRef Search ADS PubMed Hendershot CS , Witkiewitz K , George WH , et al. . ( 2011 ) Relapse prevention for addictive behaviors . Subst Abuse Treat Prev Policy 6 : 1 – 17 . Google Scholar CrossRef Search ADS PubMed Hämäläinen MD , Andersson K ( 2016 ) Method and device for estimating a risk of relapse of addictive behaviour. WO2016178617. Kechagias S , Dernroth DN , Blomgren A , et al. . ( 2015 ) Phosphatidylethanol compared with other blood tests as a biomarker of moderate alcohol consumption in healthy volunteers: a prospective randomized study . Alcohol Alcohol 50 : 399 – 406 . Google Scholar CrossRef Search ADS PubMed Leffingwell TR , Cooney NJ , Murphy JG , et al. . ( 2013 ) Continuous objective monitoring of alcohol use: twenty-first century measurement using transdermal sensors . Alcohol Clin Exp Res 37 : 16 – 22 . Google Scholar CrossRef Search ADS PubMed Marques P , Tippetts S , Allen J , et al. . ( 2010 ) Estimating driver risk using alcohol biomarkers, interlock blood alcohol concentration tests and psychometric assessments: initial descriptives . Addiction 105 : 226 – 39 . Google Scholar CrossRef Search ADS PubMed Midanik LT . ( 1988 ) Validy of self-reported alcohol use: a literature review and assessment . Br J Addict 83 : 1019 – 29 . Google Scholar CrossRef Search ADS PubMed Miller WR . ( 1996 ) What is relapse? Fifty ways to leave the wagon . Addiction 91 : S15 – 27 . Google Scholar CrossRef Search ADS PubMed Miller WR , Sanchez-Craig M . ( 1996 ) How to have a high success rate in treatment: advice for evaluators of alcohol programs . Addiction 91 : 779 – 85 . Google Scholar CrossRef Search ADS PubMed Miller WR , Walters ST , Bennett ME . ( 2001 ) How effective is alcoholism treatment in the United States? J Stud Alcohol 62 : 211 – 20 . Google Scholar CrossRef Search ADS PubMed Nanau RM , Neuman MG . ( 2015 ) Biomolecules and biomarkers used in diagnosis of alcohol drinking and in monitoring therapeutic interventions . Biomolecules 5 : 1339 – 85 . Google Scholar CrossRef Search ADS PubMed Neumann T , Spies C . ( 2003 ) Use of biomarkers for alcohol use disorders in clinical practice . Addiction 98 : 81 – 91 . Google Scholar CrossRef Search ADS PubMed Quanbeck A , Chih M-Y , Isham A , et al. . ( 2014 ) Mobile delivery of treatment for alcohol use disorders: a review of the literature . Alcohol Res 36 : 111 – 22 . Google Scholar PubMed Rose GL , Skelly JM , Badger GJ , et al. . ( 2012 ) Interactive voice response for relapse prevention following cognitive-behavioural therapy for alcohol use disorder: a pilot study . Psychology Serv 9 : 174 – 84 . Google Scholar CrossRef Search ADS Searles JS , Helzer JE , Rose GL , et al. . ( 2002 ) Concurrent and retrospective reports of alcohol consumption across 30, 90 and 366 days: interactive voice response compared with the timeline follow back . J Stud Alcohol 63 : 352 – 62 . Google Scholar CrossRef Search ADS PubMed Skipper GE , Thon N , DuPont RL , et al. . ( 2014 ) Cellular photo digital breathalyzer for monitoring alcohol use: a pilot study . Eur Addict Res 20 : 137 – 42 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Medical Council on Alcohol and Oxford University Press. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Alcohol and AlcoholismOxford University Press

Published: Mar 24, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off