Diabetes Ther (2018) 9:1099–1111 https://doi.org/10.1007/s13300-018-0421-5 ORIGINAL RESEARCH Adherence to Basal Insulin Therapy Among People with Type 2 Diabetes: A Retrospective Cohort Study of Costs and Patient Outcomes . . . Magaly Perez-Nieves Kristina S. Boye Jacek Kiljanski Dachung Cao Maureen J. Lage Received: January 8, 2018 / Published online: April 11, 2018 The Author(s) 2018 multivariable analyses were used to examine the ABSTRACT associations between adherence to basal insulin therapy and costs, resource utilization, and Introduction: This research compares costs, acute complications. resource utilization, and complications between Results: A total of 21,363 individuals were adherent and nonadherent patients over the included in the study. Three years after initiat- 3-year period post initiation on basal insulin ing therapy on basal insulin, patients who were therapy. adherent over time to basal insulin treatment Methods: The study utilized the US-based Tru- therapy (33.8% of patients) had signiﬁcantly ven Health MarketScan Research Databases higher diabetes-related drug costs. However, from 2011 through 2015. Adults aged 18 years patients’ adherence was associated with signiﬁ- or older and identiﬁed with type 2 diabetes cantly lower diabetes-related outpatient, acute (T2D) who initiated therapy on basal insulin in care, and total costs. Results for all-cause costs 2012 were included. Patients were excluded if were similar. Adherent patients also had signif- they were pregnant, ﬁlled their index basal icantly fewer all-cause and diabetes-related insulin prescription via mail order, or were not hospitalizations and emergency room visits and continuously insured from 1 year before were signiﬁcantly less likely to be diagnosed through 3 years following initiation of treat- with an acute complication. ment with basal insulin. Instrumental variables Conclusions: Results of this study illustrate that were used to control for selection bias, and despite higher drug costs, there are disease- speciﬁc and all-cause cost offsets and improved Enhanced content To view enhanced content for this article go to https://doi.org/10.6084/m9.ﬁgshare. patient outcomes associated with adherence to basal insulin therapy for people with T2D. Funding: Eli Lilly and Company. Electronic supplementary material The online version of this article (https://doi.org/10.1007/s13300- 018-0421-5) contains supplementary material, which is Keywords: Adherence; Basal insulin; Costs; available to authorized users. Proportion of days covered; Resource utilization M. Perez-Nieves K. S. Boye J. Kiljanski D. Cao Eli Lilly and Company, Indianapolis, IN, USA INTRODUCTION M. J. Lage (&) HealthMetrics Outcomes Research, LLC, Bonita Approximately 30 million American adults have Springs, FL, USA diabetes mellitus (diabetes) , which in the e-mail: email@example.com 1100 Diabetes Ther (2018) 9:1099–1111 USA is the leading cause of kidney failure, non- such, the research will add to the body of evi- trauma related lower-limb amputations, and dence speciﬁcally related to the use of basal incident blindness . Diabetes has also been insulin and will examine adherence over an found to be one of the major causes of heart extended time horizon. disease, stroke, and death . In 2012, the costs associated with diabetes totaled an estimated METHODS $245 billion in the USA, and such costs are forecasted to rise as the population ages . Data covering the period from 1 January 2011 Insulin is one of the suggested treatments for through 21 December 2015 were obtained from patients with type 2 diabetes (T2D) whose dia- the Truven Health MarketScan Research betes is not controlled with oral agents alone Databases. Both the Commercial Claims and , although research has shown that over time Encounters (CCAE) and Medicare Supplemental most patients need multiple therapies to main- and Coordination of Beneﬁts (MDCR) Databases tain glycemic control . As it is long-acting, were used in this study. The CCAE database basal insulin is used to keep blood glucose level consists of health insurance claims and patient stable . The American Diabetes Association enrollment data from large employers and states that basal insulin is the most convenient health plans across the USA. The MDCR dataset initial insulin regimen and is usually prescribed represents health services received by employ- in combination with metformin and/or another ees, dependents, and retirees in the USA with noninsulin agent . primary or Medicare supplemental coverage. Previous research among patients with T2D The CCAE and MDCR databases are generally has shown that basal insulin therapy is associ- representative of the population in the USA ated with a decrease in glycated hemoglobin , and these databases provide detailed data (A1c) and fasting blood glucose as well as with on costs, medical resource use, and outcomes improved treatment satisfaction , and treat- for healthcare services performed in both inpa- ment guidelines list basal insulin as the antidi- tient and outpatient settings. In addition to abetic treatment with the highest efﬁcacy . providing information about inpatient and However, physicians often delay initiation of outpatient services, medical claims are linked to insulin therapy among their patients with T2D outpatient prescription drug clams and person- . Frequently cited reasons for this delay level enrollment information. The data are fully include patient concerns about the pain asso- de-identiﬁed and compliant with the Health ciated with blood testing and insulin injections, Insurance Portability and Accountability Act hypoglycemia, and weight gain associated with (HIPAA). This article does not contain any new the use of insulin . While patients have been studies with human or animal subjects per- reported to prefer simpler injection regimens formed by any of the authors. , previous research has indicated that a large For entry into this study, patients were proportion of T2D patients discontinue their required to be identiﬁed as having T2D during basal insulin therapy . the calendar year 2012. Such patients were For patients with T2D, reduced adherence to identiﬁed if they received more diagnoses for drug therapy, in general, has been associated T2D than for type 1 diabetes (T1D) and/or if with poorer patient outcomes. Speciﬁcally, they received a diagnosis for T2D and ﬁlled a reduced adherence has been shown to be asso- prescription for an oral glucose-lowering agent ciated with increases in hospitalizations and (GLA) other than metformin or a sodium-glu- emergency room (ER) visits [12–15], higher rates cose linked transport 2 inhibitor, since the latter of complications [12, 16], and greater medical agents may be used in the treatment of T1D costs . The goal of the present research was [4, 19, 20]. Patients were furthermore required to compare costs, resource use, and complica- to have ﬁlled a prescription for basal insulin in tions between patients with T2D who were 2012, with the date of the ﬁrst ﬁll identiﬁed as either adherent or nonadherent to basal insulin the index date. For each patient, data were therapy within 3 years of initiating therapy. As Diabetes Ther (2018) 9:1099–1111 1101 included from 1 year before the index date (e.g., prescription via mail order. The ﬁnal sample the pre-period) through 3 years after the index consisted of 21,363 individuals. Figure 1 illus- date (e.g., the post-period). Patients were trates how each of the inclusion and exclusion excluded from the study if they: (1) ﬁlled a criteria affected sample size. prescription for basal insulin during the pre- The aim of the study was to examine the period, (2) were identiﬁed as pregnant at any relationship between patient adherence to basal time from the start of the pre-period through insulin as a class of therapeutics and outcomes, the end of the post-period, (3) were less than where adherence was proxied by the proportion 18 years old as of the index date, (4) did not of days covered (PDC). PDC was constructed as have continuous insurance coverage from the the percentage of days during the post-period start of the pre-period through the end of the that an individual had a supply of basal insulin, post-period, (5) did not have valid demographic with adjustment for the possibility that insulin data, or (6) received their index basal insulin may not be used in a method consistent with the days’ supply ﬁeld in a claims database . A patient was considered to be adherent if the Identified as Having T2D in 2012 PDC reached the 80% threshold, and a patient N=3,969,447 was considered to be nonadherent if the PDC was \ 80% [12, 15]. One advantage of using PDC as the measure of adherence is that it is Receipt of Rx for Basal Insulin in 2012, with used by both the Pharmacy Quality Alliance first date identified as index date (PQA) and the Centers for Medicare and Medi- N=408,784 caid Services (CMS) as a measure for examining the treatment of patients with diabetes [22, 23]. No Receipt of Rx for Basal Insulin in The patient outcomes of interest included Pre-Period medical costs, resource utilization, and acute N=157,333 complications. Costs were constructed using gross payments to a provider for a service, where No Diagnosis of Pregnancy Over Study Period payments were equal to the amount eligible for N=153,593 payment under the medical plan terms after applying rules such as discounts, but before applying coordination of beneﬁts, copayments, Age 18 or Older at Index Date and deductibles. All costs were converted to 2015 N=151,193 amounts using the medical component of the consumer price index. Both all-cause and dia- betes-related costs were examined, and each of Continuous Insurance Coverage over Study these cost categories was subcategorized to Period examine acute care (costs associated with hospi- N=25,640 talizations or ER visits), outpatient costs, and prescription drug costs. All-cause costs consisted Valid Demographic Data of all medical costs associated with inpatient, N=25,617 outpatient, and prescription drugs, and diabetes- related costs were constructed as the sum of all costs where (1) there was an accompanying pri- Index Rx Not Filled via Mail Order mary or secondary diagnosis of diabetes; (2) there N=21,363 was receipt of a prescription for a GLA or diabetic supplies; or (3) the patient was identiﬁed as Study period includes a 1 year pre-period and 3 year post-period. having hypoglycemia based upon a previously T2D – type 2 diabetes; Rx - prescription published and validated algorithm . In addition to examining costs, the analyses Fig. 1 Flow diagram of inclusion–exclusion criteria and also examined medical resource utilization and sample size 1102 Diabetes Ther (2018) 9:1099–1111 acute complications. The resource utilization classes of GLAs prescribed in the pre-period and outcomes examined included the probability of the number of non-GLA medications prescribed a hospitalization, the probability of an ER visit, in the pre-period. Finally, the analyses con- the number of hospitalizations, the number of trolled for the number of A1c tests the patient ER visits, and hospital length of stay (LOS). As received in the pre-period. Consistent with with costs, resource utilization was categorized previous research, this measure was used as a as either all-cause or diabetes-related. The study proxy for glycemic control , since laboratory also examined the probability of being diag- results were unavailable in the database. nosed with an acute complication over the All multivariable models used the covariates 3-year post-period. An acute complication was discussed above, while the speciﬁc functional identiﬁed based upon receipt of a primary or form depended upon the dependent variable secondary diagnosis of hyperglycemia or dia- being examined. In all cases, the estimated betic coma or identiﬁcation of hypoglycemia outcome was constructed from instrumental based upon the same algorithm used to identify variables models, which are designed to allow the condition when constructing disease-speci- for the role of unmeasured confounding. The ﬁc costs . use of retrospective data does not allow for When examining the relationship between patients to be randomized to treatment, adherence to basal insulin therapy and out- potentially leading to sample selection bias. comes, the multivariable analyses controlled for Instrumental variables are used to adjust for factors that may potentially inﬂuence patient unmeasured confounding and the sample outcomes. Given the factors available in the selection bias . Consistent with previous database, the analyses controlled for patient research , the following variables were con- demographic characteristics, general health and structed over the ﬁrst 30 days of the post-period comorbidities, type of provider visited, medi- and used as instruments: (1) the percentage of cation use, and A1c tests. Patient demographic total payments for basal insulin prescriptions characteristics that were measured at the index that were paid by coinsurance; and (2) the dol- date consisted of age, sex, region of residence, lar (US) amount of copayments that were paid and insurance plan type. Patient overall general for basal insulin prescriptions. It is hypothe- health was measured over the pre-period and sized that these variables will directly affect assessed using the Charlson comorbidity index patient adherence while not being directly (CCI) , while, the Diabetes Complications related to patient outcomes. Severity Index (DCSI) was utilized to proxy the General linear models with gamma distribu- severity of diabetes complications . In tion and log link were used to estimate all costs addition to these index scores, the models also models except those for acute care. Such general included pre-period comorbid diagnoses of linear models have been shown to account for anxiety, depression, and hyperlipidemia, since the skewed nature of cost data and to behave these diagnoses have been shown to be com- well in the estimation of population means of mon in patients with diabetes or linked to healthcare costs . Acute care costs were patient nonadherence but were not captured in estimated using a two-part model where, in the either of the two index scores [27, 28]. The ﬁrst part, a binary regression model was used to analyses also controlled for visits to specialists estimate the probability of having an acute care in the pre-period with indicator variables for visit. In the second stage, a general linear model patient visits to a cardiologist, endocrinologist, with gamma distribution and log link was used ophthalmologist, or nephrologist, and they to estimate acute care costs for patients who had controlled for the type of basal insulin pre- at least one acute care visit. The predicted scribed at the index date, the number of GLAs, probability of an acute care visit, estimated from and overall medication use. These factors were the ﬁrst part of the model, was then multiplied captured by an indicator variable if the index by costs estimates from the second part of the prescription was for U-500 basal insulin. As model to obtain unconditional average costs well, indicator variables captured the number of estimates. Diabetes Ther (2018) 9:1099–1111 1103 Negative binomial models were used to likely to reside in the South (40.96 vs 44.81%; examine the number of hospitalizations, the P \ 0.0001), compared to nonadherent number of ER visits, and hospital LOS. Logistic patients. Adherent patients were also in better models were used to examine the probability of general health than nonadherent patients, as being hospitalized or visiting the ER and the evidenced by lower rates of anxiety (3.74 vs. probability of being diagnosed with an acute 5.13%; P \ 0.0001) and depression (6.96 vs. complication. 8.98%; P \ 0.0001), as well as by a signiﬁcantly Descriptive statistics were examined using lower mean CCI score (2.27 [SD 1.84] vs 2.50 Chi-square statistics for categorical variables and [SD 2.08]; P \ 0.0001). A smaller percentage of Kruskal–Wallis tests for continuous variables. For adherent patients had visited a cardiologist logistic models the odds ratios (ORs) and 95% (29.10 vs. 31.75%; P \ 0.0001) in the pre-period conﬁdence intervals (CIs) associated with being compared to the nonadherent. In contrast, adherent (vs. nonadherent) were reported. For all adherent patients were more likely to have been other outcomes, the multivariable analyses were diagnosed with comorbid hyperlipidemia used to predict adjusted mean outcomes. Robust (57.07 vs. 51.82%; P \ 0.0001) and to have vis- standard errors for each of these adjusted means ited an ophthalmologist (24.94 vs. 22.39%; were used to construct 95% conﬁdence intervals P \ 0.0001). Adherent patients may have had for the adjusted means. Complete results of all less severe diabetes compared to nonadherent multivariable analyses are presented in the sup- patients, as evidenced by the signiﬁcantly lower plementary material. All analyses were con- DCSI score (1.22 [SD 1.69] vs. 1.45 [SD 1.87]; ducted using the SAS version 9.4 statistical P \ 0.0001) and suggested by the larger number software (SAS Institute, Cary, NC, USA). P values of A1c tests ordered for adherent patients (1.74 of \ 0.05 were considered, a priori, to be statis- [SD 1.45] vs. 1.43 [SD 1.39]; P \ 0.0001). The tically signiﬁcant. adherent patients were also prescribed more classes of GLAs in the pre-period (2.19 [SD 1.24] vs. 1.69 [SD 1.22]; P \ 0.0001). RESULTS Figure 2 shows the relation between adher- ence to basal insulin and 3-year diabetes-related Table 1 presents the descriptive statistics for the and all-cause medical costs. As shown in Fig. 2a, 21,363 T2D patients included in the study. The adherence to insulin therapy was associated entire cohort’s mean age was 57 (standard devia- with signiﬁcantly lower diabetes-related outpa- tion [SD] 12.67) years, and the majority were tient costs ($6319 vs. $6759 [all costs shown in male (54.71%). Patients most commonly resided U.S. dollars]; P \ 0.0001) and acute care costs in the South (43.51%) or North Central (27.10%) ($8290 vs. $9789; P \ 0.0001), but signiﬁcantly regions of the USA, and most were covered by higher diabetes-related drug costs ($11,898 vs. preferred provider organizations (56.15%) or had $10,316; P \ 0.0001). Overall, diabetes-related comprehensive insurance (15.02%). On average, total costs were signiﬁcantly lower for adherent patients ﬁlled prescriptions for 1.86 (SD 1.25) patients than for nonadherent patients classes of GLAs in the pre-period and 9.32 ($26,488 vs. $26,800; P = 0.0227). All-cause (SD 6.30) prescriptions for non-GLA medications costs generally follow the same pattern. Specif- over the same period. Almost one-third of ically, despite signiﬁcantly higher drug costs patients visited a cardiologist (30.85%), nearly over the post-period for the adherent patients one-quarter of patients visited an ophthalmolo- relative to the nonadherent patients($22,267 vs. gist (23.25%), and almost one-ﬁfth visited an $21,030; P \ 0.0001), the adherent cohort had endocrinologist (17.27%) in the pre-period. signiﬁcantly lower outpatient costs ($29,057 vs. In addition to characterizing the entire $32,042; P \ 0.0001), acute care costs ($22,112 population, Table 1 also presents descriptive vs. $25,458; P \ 0.0001), and all-cause total statistics comparing adherent to nonadherent costs ($73,687 vs. $78,778; P \ 0.0001). patients. Adherent patients were more likely to The associations between adherence and be male (57.36 vs 53.35%; P \ 0.0001) and less medical resource use are examined in Table 2 1104 Diabetes Ther (2018) 9:1099–1111 Table 1 Patient characteristics—overall and by adherence status Variable All patients Nonadherent patients Adherent P value a a (N = 21,363) (N = 14,149) (N = 7214) N or mean %or N or mean %or N or mean %or [median] SD [median] SD [median] SD Patient characteristic Age (in years) (mean [median]; 57.09  12.67 57.06  13.41 57.17  11.07 0.0024 SD) Sex (n;%) \0.0001 Male 11,687 54.71% 7549 53.35% 4138 57.36% Female 9676 45.29% 6600 46.65% 3076 42.64% Region (n;%) \0.0001 Northeast 3843 17.99% 2435 17.21% 1408 19.52% North Central 5789 27.10% 3774 26.67% 2015 27.93% South 9295 43.51% 6340 44.81% 2955 40.96% West 2436 11.40% 1600 11.31% 836 11.59% Insurance plan type (n;%) \0.0001 Comprehensive insurance 3208 15.02% 2261 15.98% 947 13.13% Health maintenance 2679 12.54% 1769 12.50% 910 12.61% organization Point of service 1842 8.62% 1216 8.59% 626 8.68% Preferred provider 11,995 56.15% 7819 55.26% 4176 57.89% organization Unknown 1639 7.67% 1084 7.66% 555 7.69% General health and comorbidities in pre-period CCI (mean [median]; SD) 2.42  2.01 2.50  2.08 2.27  1.84 \0.0001 DCSI (mean [median]; SD) 1.37  1.82 1.45  1.87 1.22  1.69 \0.0001 Anxiety (n; %) 996 4.66% 726 5.13% 270 3.74% \0.0001 Depression (n; %) 1772 8.29% 1270 8.98% 502 6.96% \0.0001 Hyperlipidemia (n; %) 11,449 53.59% 7332 51.82% 4117 57.07% \0.0001 Number of A1c tests (mean 1.53  1.42 1.43  1.39 1.74  1.45 \0.0001 [median]; SD) Number of GLA classes (mean 1.86  1.25 1.69  1.22 2.19  1.24 \0.0001 [median]; SD) Number of Non-GLA 9.32 6.30 9.27 6.42 9.42 6.07 0.0849 medications (mean; SD) Diabetes Ther (2018) 9:1099–1111 1105 Table 1 continued Variable All patients Nonadherent patients Adherent P value a a (N = 21,363) (N = 14,149) (N = 7214) N or mean %or N or mean %or N or mean %or [median] SD [median] SD [median] SD Provider and prescriber visited in pre-period (n;%) Cardiologist visit 6591 30.85% 4492 31.75% 2099 29.10% 0.0001 Ophthalmologist 4967 23.25% 3168 22.39% 1799 24.94% \0.0001 Nephrologist 1285 6.02% 875 6.18% 410 5.68% 0.1454 Endocrinologist 3690 17.27% 2439 17.24% 1251 17.34% 0.8502 Chi-square tests were used to compare differences in categorical variables. Kruskal–Wallis tests were used to compare differences in the distribution of continuous variables. P values of \ 0.05 were considered, to be statistically signiﬁcant SD Standard deviation, CCI Charlson comorbidity index, DCSI diabetes complications severity index, A1c glycated hemoglobin, GLA glucose-lowering agent A patient was considered to be adherent if the proportion of days covered (PDC) reached the 80% threshold, and a patient was considered to be nonadherent if the PDC was \ 80% a Diabetes-Related Costs b All-Cause Costs $30,000 $90,000 $26,800 $26,488 $78,778 $80,000 $73,687 $25,000 $70,000 $20,000 $60,000 $50,000 $15,000 $11,898 $40,000 $32,042 $10,316 $9,789 $29,057 $10,000 $30,000 $25,458 $8,290 $22,267 $22,112 $6,759 $21,030 $6,319 $20,000 $5,000 $10,000 $0 $0 Total Outpatient Acute Drug Total Outpatient Acute Care Drug Non-Adherent Adherent Non-Adherent Adherent Results of t-tests which examine estimated mean costs (with bootstrapped 95% confidence intervals) from instrumental variables multivariable regression model that controls for patient characteristics, general health, comorbidities, number of A1c tests, visits to specialists, medication use and patient adherence. * T-tests which compare differences in costs between non-adherence and adherent patients were statistically significant (P<0.05). Adherence defined as PDC ≥ 80% nonadherence defined as PDC < 80% Fig. 2 Diabetes-related and all-cause medical costs: adherent versus nonadherent patients. Ac1 Gycated hemoglobin, PDC Proportion of days covered Costs ($) Costs ($) 1106 Diabetes Ther (2018) 9:1099–1111 Table 2 All-cause and diabetes-related resource utilization: adherent versus nonadherent patients Resource utilization Nonadherent (PDC < 80%) Adherent (PDC ‡ 80%) Mean 95% Conﬁdence interval Mean 95% Conﬁdence interval Diabetes-related resource utilization Number of hospitalizations 0.64 0.63–0.65 0.53 0.52–0.54 Hospital length of stay 3.68 3.61–3.76 2.91 2.83–2.99 Number of ER visits 1.21 1.19–1.22 1.00 0.98–1.02 All-cause resource utilization Number of hospitalizations 0.89 0.87–0.90 0.73 0.71–0.74 Hospital length of stay 5.40 5.29–5.52 4.33 4.20–4.45 Number of ER visits 2.46 2.43–2.50 2.05 2.01–2.09 ER Emergency room Statistically signiﬁcant results (P \ 0.0001) from multivariable regression models that controlled for patient character- istics, general health, comorbidities, number of A1c tests performed, visits to specialists, medication use and patient adherence 0.90 0.82 0.80 0.77 0.77 0.76 0.75 0.74 0.72 0.71 0.71 0.70 0.70 0.70 0.68 0.67 0.66 0.65 0.60 Diabetes-Related Diabetes-Related ER All-Cause All-Cause ER Visit Acute Complication Hospitalization Visit Hospitalization Results from logistic regression that controls for patient characteristics, general health, comorbidities, number of A1c tests, visits to specialists, medication use and patient adherence. Odds ratio is for adherent patient compared to non-adherent patients; 95 confidence intervals. Odds ratio represents the odds the event will occur for an adherent patient relative to a non-adherent patient. Adherence defined as PDC ≥ 80%; non-adherence defined as PDC < 80%. Fig. 3 Likelihood of resource utilization or acute complication: adherent vs nonadherent patients. ER Emergency room and Fig. 3. In general, these results indicate that nonadherence. As Fig. 3 illustrates, adherent adherence was associated with signiﬁcantly patients were 29% less likely to have a diabetes- better patient outcomes relative to related hospitalization (OR 0.71; 95% CI Odds Ratio and 95% Confidence Interval Diabetes Ther (2018) 9:1099–1111 1107 0.67–0.76) and 30% less likely to have a dia- adherence in general and improved patient betes-related ER visit (OR 0.70; 95% CI outcomes [12, 13, 15, 30, 32] as well as studies 0.66–0.75). Similarly, adherent patients were showing an association between persistence 28% less likely to be hospitalized for any cause with basal insulin therapy and better outcomes (OR 0.72; 95% CI 0.68–0.77), and 30% less likely [33, 34]. The present investigation furthers this to have an all-cause ER visit (OR 0.70; 95% CI research by focusing on basal insulin therapy 0.65–0.74). Consistent with the results shown over an extended time horizon while control- in Fig. 3, Table 2 illustrates that the adherent ling for a wide range of factors that may affect patients had signiﬁcantly fewer diabetes-related patient outcomes. Additionally, this study hospitalizations (0.53 [SD 0.45] vs. 0.64 examined the robustness of results to alterna- [SD 0.58]; P \ 0.0001) and ER visits (1.00 tive measures of adherence. [SD 0.86] vs. 1.21 [SD 1.09]; P \ 0.0001), as well Consistent with previous research [33, 34], as signiﬁcantly fewer all-cause hospitalizations adherence to basal insulin in the present study (0.73 [SD 0.69] vs. 0.89 [SD 0.85]; P \ 0.0001) was associated with higher drug costs but lower and all-cause ER visits (2.05 [SD 1.84] vs. 2.46 acute care costs. Adherence was furthermore [SD 2.29]; P \ 0.0001). Diabetes-related hospital associated with lower total costs, a result that is LOS, all-cause hospital LOS, and the probability in agreement with those of a number of studies of an acute complication were also signiﬁcantly that have looked at GLA adherence in general lower for patients who were adherent to basal . The present ﬁndings suggest that the insulin therapy. Speciﬁcally, adherence to basal increased medication costs associated with insulin therapy was associated with a 23% improved adherence were offset by decreases in reduction in the probability of an acute com- acute care costs. It should be noted that dia- plication being diagnosed over the post-period betes-related total costs were not statistically (OR 0.77; 95% CI 0.71–0.82). signiﬁcantly different between the two cohorts Two sensitivity analyses were conducted to when alternative measures of adherence were test the robustness of the results. First, all anal- examined. yses were re-estimated using the medication In concert with earlier literature [12, 35–37], possession ratio (MPR) rather than the PDC as the adherent patients in the present study had a the measure of adherence. In contrast to the signiﬁcantly lower use of hospital and ER PDC, the MPR is calculated using the sum of resources relative to the nonadherent patients. days’ supply for all prescriptions in a period of This reduction in acute care costs represents interest. As a result, this measure may overesti- potentially large costs savings for payers, given mate adherence for patients who reﬁll their that hospitalization comprised 43% of all direct prescriptions early. Second, PDC was recalcu- diabetes spending in the USA in 2012 . lated without the adjustment to account for the Decreased use of acute care due to better fact that the days’ supply ﬁeld ﬁlled out by the adherence also has humanistic beneﬁts, with pharmacist may not be an accurate measure of less hospitalization associated with improved how insulin is actually used by the patient. The long-term outcomes, increased health-related ﬁndings were generally not sensitive to these quality of life , as well as reductions in alternative speciﬁcations, except that in both of indirect costs for patients and caregivers . the alternative models there were no statisti- Hospitalization furthermore carries intrinsic cally signiﬁcant differences between the dia- health risks, including healthcare-associated betes-related total costs of the adherent relative infections [39, 40] and medical errors [41, 42]. to the nonadherent patients. Thus, spending on drug therapy that is associ- ated with decreased use of acute care may be considered worthwhile in and of itself, let alone DISCUSSION when offset by other cost reductions. Relative to those who were nonadherent, the The ﬁndings support a large body of previous adherent cohort was signiﬁcantly less likely to research that has revealed a link between GLA have an acute complication. This ﬁnding is 1108 Diabetes Ther (2018) 9:1099–1111 consistent with previous research [12, 16]as with a wide range of previous research which well as with the ﬁnding in this study that has examined the association between the use adherence was associated with lower acute care of insulin and patient outcomes [46–49]. costs. Research has also shown that prolonged In addition, the analyses were based upon exposure to hyperglycemia is associated with observational health insurance claims data that the development of diabetic complications , described a population of commercially insured and among patients with one or more cardio- T2D patients who may or may not be represen- vascular risk factor, severe hypoglycemia is tative of the majority of Americans with T2D. For associated with macrovascular events and car- example, the population does not include diovascular mortality . Furthermore, for patients who were uninsured or insured via older patients with T2D, research has found Medicaid or through the Veterans Health that hypoglycemic episodes accompanied with Administration. Researchhas shownthatboth hospitalization or ER visits are associated with insurance type and the lack of insurance affect an increased risk of dementia . A lower rate patient adherence [50, 51]. The data also over- of acute complications over a 3-year time hori- represent patients from the Southern and North zon may, therefore, have implications for Central regions of the USA , although the longer term complications and outcomes. multivariable analyses controlled for regional The present study is similar in design to a differences. The use of diagnostic codes was not as previous one which examined adherence to rigorous as formal assessments and may under- GLAs in general over a 3-year time period . represent certain conditions, such as hypo- The results are generally consistent between the glycemia. Factors such as A1c, race, duration of two studies, although both all-cause and dia- diabetes, or socioeconomic class, may be associ- betes-related costs are higher in the present ated with patient outcomes but could not be study and resource utilization in the post-period controlled for as they were not observable in our was higher. One key difference between the two database. However, the instrumental variables studies is that this study included patients age models were used to control for unobservable 65 years or older, and these patients generally confounders. Finally, the study focused on sta- have more comorbidities and higher medical tistical signiﬁcance and was unable to determine expenses. Furthermore, patients in the present whether differences in outcomes represented study may have had T2D for a longer duration minimal clinically important changes. or more severe T2D, given that basal insulin therapy is not typically prescribed at the initia- CONCLUSIONS tion of therapy but further along the path of disease progression. In conclusion, these analyses showed that The results of this study must be interpreted patients who adhered to their basal insulin within the context of its limitations. First, the therapy over a 3-year time horizon had signiﬁ- use of claims data does not allow for an exam- cantly lower total costs (diabetes-related and all- ination of whether the medication is being cause) relative to their nonadherent counter- taken as prescribed, or even taken at all. This parts, despite having higher drug costs. In this limitation may be especially important when study, basal insulin adherence was furthermore examining the use of insulin, which is more associated with lower diabetes-related and all- challenging to administer than oral drugs. Also, cause outpatient costs and acute care costs, all patients who stopped taking insulin in the lower odds of an acute complication, lower post-period were identiﬁed as having reduced odds or a hospitalization or ER visits, fewer adherence, since the reason for discontinuation hospitalizations and ER visits, and shorter hos- of therapy is not known. As a result, patients pital LOS. The ﬁndings of this study suggest that who switched or discontinued therapy due to for adults with T2D adherence to basal insulin adverse events were categorized as nonadher- may lead to signiﬁcant patient beneﬁts while ent. However, it should be noted that this reducing costs for payers. measurement and terminology are consistent Diabetes Ther (2018) 9:1099–1111 1109 License (http://creativecommons.org/licenses/ ACKNOWLEDGEMENTS by-nc/4.0/), which permits any noncommercial use, distribution, and reproduction in any medium, provided you give appropriate credit Funding. Sponsorship for this study and to the original author(s) and the source, provide article processing charges was funded by Eli a link to the Creative Commons license, and Lilly and Company. indicate if changes were made. Editorial Assistance. The authors would like to thank Patricia Platt and Michael Treglia of HealthMetrics Outcomes Research for their REFERENCES assistance in manuscript preparation. Authorship. All named authors meet the 1. American Diabetes Association. Statistics about diabetes [Internet]. 2017. http://www.diabetes.org/ International Committee of Medical Journal diabetes-basics/statistics/. Accessed December 2017. Editors (ICMJE) criteria for authorship for this manuscript. All authors had full access to all of 2. Centers for Disease Control and Prevention. the data in this study and take complete National diabetes fact sheet: national estimates and general information on diabetes and prediabetes in responsibility for the integrity of the data and the United States, 2011 [Internet]. Atlanta, GA: U.S. accuracy of the data analysis. Department of Health and Human Services, Center for Disease Control and Prevention 2011. https:// Disclosures. Magaly Perez-Nieves is an www.cdc.gov/diabetes/pubs/pdf/ndfs_2011.pdf. employee and shareholder of Eli Lilly and Accessed December 2017. Company. Kristina S. Boye is an employee and 3. American Diabetes Association. Economic costs of shareholder of Eli Lilly and Company. Jacek diabetes in the US in 2012. Diabetes Care. Kiljanki is an employee and shareholder of Eli 2013;36(4):1033–46. Lilly and Company. Dachung Cao is an 4. American Diabetes Association. 8. Pharmacologic employee and shareholder of Eli Lilly and approaches to glycemic treatment. Standards of Company. Maureen J Lage is an employee of medical care in diabetes—2018. Diabetes Care. HealthMetrics Outcomes Research, which 2018;41[Suppl 1]:S73–85. received consulting fees from Eli Lilly and 5. Turner RC, Cull CA, Frighi V, Holman RR. Glycemic Company for her work on this project. control with diet, sulfonylurea, metformin, or insulin in patients with type 2 diabetes mellitus: Compliance with Ethics Guidelines. This progressive requirement for multiple therapies article does not contain any new studies with (UKPDS 49). UK Prospective Diabetes Study (UKPDS) Group. JAMA. 1999;281(21):2005–12. human or animal subjects performed by any of the authors. All patients were de-identiﬁed and 6. Kuroda A, Kaneto H, Yasuda T, et al. Basal insulin complied with the patient conﬁdentiality requirement is 30% o. f the total daily insulin dose requirements of the Health Insurance Portabil- in type 1 diabetic patients who use the insulin pump. Diabetes Care. 2011;34(5):1089–90. ity and Accountability Act. 7. Chien M, Chen Y, Hung Y, et al. Glycemic control Data Availability. The datasets analyzed and adherence to basal insulin therapy in Tai- during the current study are not publicly avail- wanese patients with type 2 diabetes mellitus. J Di- able due to licensing agreements between Tru- abetes Investig. 2016;7(6):881–8. ven and Eli Lilly and Company but are available 8. Home P, Riddle M, Cefalu WT, et al. Insulin therapy from the corresponding author on reasonable in people with type 2 diabetes: opportunities and request. challenges? Diabetes Care. 2014;37(6):1499–508. 9. Lau ANC, Tang T, Halapy H, Thorpe K, Yu CHY. Open Access. This article is distributed Initiating insulin in patients with type 2 diabetes. under the terms of the Creative Commons CMAJ. 2012;184(7):767–76. Attribution-NonCommercial 4.0 International 1110 Diabetes Ther (2018) 9:1099–1111 10. Peyrot M, Rubin RR, Kruger DF, Travis LB. Corre- 22. Pharmacy Quality Alliance. PQA medication qual- lates of insulin injection omission. Diabetes Care. ity measures in the health insurance marketplace 2010;33(2):240–5. [Internet]. 2017. http://pqaalliance.org/measures/ qrs.asp. Accessed December 2017. 11. Westerbacka J, Mihailov H, Valle T, Ja¨a¨skela¨inen S, Kaukua J. Adherence to initiated basal insulin ana- 23. Academy of Managed Care Pharmacy. Medicare log treatment in type 1 and 2 diabetes. Value part D quality measures [Internet]. 2014. http:// Health. 2015;18(7):A612. www.amcp.org/WorkArea/DownloadAsset.aspx?id= 18900. Accessed December 2017 12. Gibson TB, Song X, Alemayehu B, et al. Cost shar- ing, adherence, and health outcomes in patients 24. Ginde A, Blanc P, Lieberman R, Camargo C. Vali- with diabetes. Am J Manag Care. dation of ICD-9-CM coding algorithm for improved 2010;16(8):589–600. identiﬁcation of hypoglycemia visits. BMC Endocr Disord. 2008;8(1):4. 13. Jha AK, Aubert RE, Yao J, Teagarden JR, Epstein RS. Greater adherence to diabetes drugs is linked to less 25. Quan H, Sundararajan V, Halfon P, et al. Coding hospital use and could save nearly $5 billion algorithms for deﬁning comorbidities in ICD-9-CM annually. Health Aff (Millwood). and ICD-10 administrative data. Med Care. 2012;31(8):1836–46. 2005;43(11):1130–9. 14. Curtis SE, Boye KS, Lage MJ, Garcia-Perez L-E. 26. Chang H-Y, Weiner JP, Richards TM, Bleich SN, Medication adherence and improved outcomes Segal JB. Validating the adapted Diabetes Compli- among patients with type 2 diabetes. Am J Manag cations Severity Index in claims data. Am J Manag Care. 2017;23(7):e208–14. Care. 2012;18(11):721–6. 15. Juarez DT, Tan C, Davis J, Mau M. Factors affecting 27. Ayyagari R, Wei W, Cheng D, Pan C, Signorovitch J, sustained medication adherence and its impact on Wu EQ. Effect of adherence and insulin delivery health care utilization in patients with diabetes. system on clinical and economic outcomes among J Pharm Health Serv Res. 2013;4(2):89–94. patients with type 2 diabetes initiating insulin treatment. Value Health. 2015;18(2):198–205. 16. Yu AP, Yu YF, Nichol MB. Estimating the effect of medication adherence on health outcomes among 28. Palizgir M, Bakhtiari M, Esteghamati A. Association patients with type 2 diabetes—an application of of depression and anxiety with diabetes mellitus marginal structural models. Value Health. type 2 concerning some sociological factors. Iran 2010;13(8):1038–45. Red Crescent Med J. 2013;15(8):644–8. 17. Kennedy-Martin T, Boye KS, Peng X. Cost of med- 29. Basu A. Estimating decision-relevant comparative ication adherence and persistence in type 2 diabetes effects using instrumental variables. Stat Biosci. mellitus: a literature review. Patient Prefer Adher- 2011;3(1):6–27. ence. 2017;11:1103–17. 30. Encinosa WE, Bernard D, Dor A. Does prescription 18. Hansen L. The Truven Health MarketScan Data- drug adherence reduce hospitalizations and costs? bases for life sciences researchers[Internet]. 2017. The case of diabetes. Adv Health Econ Health Serv https://truvenhealth.com/Portals/0/Assets/2017- Res. 2010;22:151–73. MarketScan-Databases-Life-Sciences-Researchers- WP.pdf. Accessed March 2018 31. Malehi AS, Pourmotahari F, Angali KA. Statistical models for the analysis of skewed healthcare cost 19. Livingstone R, Boyle JG, Petrie JR. REMOVAL study data: a simulation study. Health Econ Rev. team. A new perspective on metformin therapy in 2015;5:11. type 1 diabetes. Diabetologia. 2017;60(9):1594–600. 32. Cheng S-H, Chen C-C, Tseng C-H. Does medication 20. El Masri D, Ghosh S, Jaber LA. Safety and efﬁcacy of adherence lead to lower healthcare expenses for sodium-glucose cotransporter 2 (SGLT2) inhibitors patients with diabetes? Am J Manag Care. in type 1 diabetes: a systematic review and meta- 2013;19(8):662–70. analysis. Diabetes Res Clin Pract. 2018;137:83–92. 33. Hadjiyianni I, Desai U, Suzuki S, et al. Basal insulin 21. Buysman E, Conner C, Aagren M, Bouchard J, Liu F. persistence, associated factors, and outcomes after Adherence and persistence to a regimen of basal treatment initiation: a retrospective database study insulin in a pre-ﬁlled pen compared to vial/syringe among people with Type 2 diabetes mellitus in in insulin-naıve patients with type 2 diabetes. Curr Japan. Diabetes Ther. 2017;8(1):149–66. Med Res Opin. 2011;27(9):1709–17. Diabetes Ther (2018) 9:1099–1111 1111 34. Perez-Nieves M, Kabul S, Desai U, et al. Basal insulin 44. Haukka J, Hoti F, Era¨sto¨ P, Saukkonen T, Ma¨kimat- persistence, associated factors, and outcomes after tila S, Korhonen P. Evaluation of the incidence and treatment initiation among people with type 2 risk of hypoglycemic coma associated with selec- diabetes mellitus in the US. Curr Med Res Opin. tion of basal insulin in the treatment of diabetes: a 2016;32(4):669–80. Finnish register linkage study. Pharmacoepidemiol Drug Saf. 2013;22(12):1326–35. 35. Wong ES, Bryson CL, Hebert PL, Liu C-F. Estimating the impact of oral diabetes medication adherence 45. Whitmer RA, Karter AJ, Yaffe K, Quesenberry CP, on medical costs in VA. Ann Pharmacother. Selby JV. Hypoglycemic episodes and risk of 2014;48(8):978–85. dementia in older patients with type 2 diabetes mellitus. JAMA. 2009;301(15):1565–72. 36. Asche C, LaFleur J, Conner C. A review of diabetes treatment adherence and the association with 46. Chandran A, Bonafede MK, Nigam S, Saltiel-Berzin clinical and economic outcomes. Clin Ther. R, Hirsch LJ, Lahue BJ. Adherence to insulin pen 2011;33(1):74–109. therapy is associated with reduction in healthcare costs among patients with type 2 diabetes mellitus. Am Health Drug Beneﬁts. 2015;8(3):148–58. 37. Han E, Suh D-C, Lee S-M, Jang S. The impact of medication adherence on health outcomes for chronic metabolic diseases: a retrospective cohort 47. Cobden D, Lee WC, Balu A, Joshi V, Pashos CL. study. Res Soc Adm Pharm. 2014;10(6):e87–98. Health outcomes and economic impact of therapy conversion to a biphasic insulin analog pen among 38. Courtney M, Edwards H, Chang A, Parker A, Fin- privately insured patients with type 2 diabetes layson K, Hamilton K. Fewer emergency readmis- mellitus. Pharmacotherapy. 2007;27(7):948–62. sions and better quality of life for older adults at risk of hospital readmission: a randomized controlled 48. Kleinman NL, Schaneman JL, Lynch WD. The trial to determine the effectiveness of a 24-week association of insulin medication possession ratio, exercise and telephone follow-up program. J Am use of insulin glargine, and health beneﬁt costs in Geriatr Soc. 2009;57(3):395–402. employees and spouses with type 2 diabetes. J Oc- cup Environ Med. 2008;50(12):1386–93. 39. Sydnor ERM, Perl TM. Hospital epidemiology and infection control in acute-care settings. Clin 49. Egede LE, Gebreqziabher M, Dismuke CE, et al. Microbiol Rev. 2011;24(1):141–73. Medication nonadherence in diabetes: longitudinal effects on costs and potential cost savings from 40. Quach C, McArthur M, McGeer A, et al. Risk of improvement. Diabetes Care. 2012;35(12):2533–9. infection following a visit to the emergency department: a cohort study. CMAJ. 50. Iuga AO, McGuire MJ. Adherence and health care 2012;184(4):E232–9. costs. Risk Manag Healthc Policy. 2014;7(35–44):3. 41. Gleason KM, McDaniel MR, Feinglass J, et al. 51. Cohen MJ, Shaykevich S, Cawthon C, Kripalani S, Results of the medications at transitions and clini- Paasche-Orlow MK, Schnipper JL. Predictors of cal handoffs (MATCH) study: an analysis of medi- medication adherence postdischarge: the impact of cation reconciliation errors and risk factors at patient age, insurance status, and prior adherence. hospital admission. J Gen Intern Med. J Hosp Med. 2012;7(6):470–5. 2010;25(5):441–7. 52. US Census Bureau. United States population growth 42. Bleetman A, Sanusi S, Dale T, Brace S. Human fac- by region [Internet]. 2017. https://www.census.gov/ tors and error prevention in emergency medicine. popclock/data_tables.php?component=growth. Emerg Med J. 2012;29(5):389–93. Accessed March 2018. 43. Aronson D. Hyperglycemia and the pathobiology of diabetic complications. Adv Cardiol. 2008;45:1–16.
– Springer Journals
Published: Apr 11, 2018