Persistence to Basal Insulin: Association With Health Outcomes in a Population With Type 2 Diabetes Free

This retrospective analysis was conducted using the Truven (now part of IBM Watson Health) MarketScan Commercial and Medicare Supplemental Databases. This combined database includes prescription drug claims and medical claims for more than 250 million insured individuals. Data were extracted from 1 January 2011 to 31 December 2015, the most recent data available at the time of the study.

Medical claims data used in the study included all claims related to outpatient and inpatient care to capture health care utilization and expenditures for both outpatient and inpatient services. The prescription drug claims data included medication filled (National Drug Code and General Product Identifier codes), prescription fill date, quantity dispensed, and days’ supply, and laboratory test results data included LOINC (Logical Observation Identifiers Names and Codes) to identify laboratory tests and result values.

For each subject, an index date was identified that corresponded to the date of the first claim for basal insulin (i.e., glargine or detemir) during the identification period (1 July 2011 to 31 December 2014). Baseline variables were calculated for the 6-month period before each subject’s index date. Post-index variables were calculated for the 12-month period after each subject’s index date.

Individuals were included in the study if they 1) were ≥18 years of age on the index date, 2) were continuously enrolled in medical and pharmacy benefits for a commercial or Medicare supplemental insurance plan during the study period, 3) had a diagnosis of type 2 diabetes (International Classification of Diseases, 9th revision, clinical modification [ICD-9-CM] codes 250.x0 or 250.x2) any time during the study period, 4) had at least one claim for a basal insulin during the identification period, 5) were basal insulin–naive (i.e., newly initiated on basal insulin) during the pre-index period (i.e., the 6 months before the index date), and 6) had at least one claim for an oral antidiabetic drug (OAD) or glucagon-like peptide 1 (GLP-1) receptor agonist during the 6-month pre-index period. The requirements for type 2 diabetes diagnosis codes and OAD/GLP-1 receptor agonist claims were to exclude individuals with type 1 diabetes and ensure that the study was focused on individuals with type 2 diabetes.

This study did not include individuals using NPH insulin to align with the basal insulin products used in the Wei study. However, given its common use, NPH insulin was added to the measure development and is included in the current health plan performance measure specifications.

Persistence to basal insulin was measured from prescription drug claims data using the Wei method, which compares a patient’s refill interval to the expected refill interval based on the population distribution of initial refill intervals (15,16). The population distribution and 90th percentile were calculated for each basal insulin quantity. Subjects were classified as persistent if their refill intervals were less than the 90th percentile of the population distribution for subjects who used the same basal insulins and quantities (Supplementary Table S1).

Baseline demographics, including age, sex, and type of insurance (commercial vs. Medicare supplemental) were derived from the enrollment data for each member. Baseline clinical, utilization, and cost characteristics, including A1C laboratory test results (for the subset of the study population with A1C values), all-cause and diabetes-related hospital admissions, all-cause emergency department (ED) and outpatient physician visits, and all-cause and diabetes-related health care costs, were calculated. Medication characteristics were also assessed, including the number of unique prescription drug classes and the type of basal insulin used (pen vs. vial). In addition, the Deyo-Charlson Comorbidity Index (DCCI) was constructed from ICD-9-CM codes based on the methodology of Deyo et al. (17). All baseline factors were identified in the 6 months prior to the index date.

Outcomes of interest included health care utilization, expenditures, and clinical variables measured in the 12-month post-index period. Two clinical variables were measured to assess disease state control and medication safety. The first was the categorical variable of whether the subject had an A1C ≥8% in the last 3 months of the post-index period for subjects who had at least one A1C value in that period. For patients with more than one A1C value in the 3-month period, the last A1C value was used. The A1C threshold of 8% was selected based on the HEDIS (Healthcare Effectiveness Data and Information Set) comprehensive diabetes care measure, which defines A1C <8% as “good control” (18). The second clinical measure was safety- oriented and measured hypoglycemia-related events. Hypoglycemia events were identified from claims with a diagnosis code for hypoglycemia (ICD-9-CM 250.8x, 251.0x, 251.1x, or 251.2x). This study examined the occurrence of any hypoglycemia-related events, on average, and the number of patients who had at least one hypoglycemia-related event during the study period. Health care utilization variables included all-cause hospital admissions and ED visits identified using medical claims. All-cause total health care expenditures, including costs resulting from comorbidities or complications and those not directly related to treatment of the disease, and diabetes-related medical expenditures (claims with a diagnosis code of ICD-9-CM 250.x0 or 250.2) were calculated and adjusted to 2016 U.S. dollars using the medical component of the Consumer Price Index. Expenditures included the patient’s cost-share (i.e., coinsurance, copayment, deductible, and coordination of benefits) and the amount paid by the health plan.

Multivariable generalized linear models with a log link and γ distribution were used to assess the relationship between insulin persistence and health care expenditures while controlling for baseline variables. Beta coefficients were used to compute cost ratios (CRs) and their corresponding 95% CIs. Multivariable logistic regression models were used to determine the relationships of persistence with health care resource utilization outcomes and clinical outcomes while controlling for baseline variables. Beta coefficients were used to compute odds ratios (ORs) and corresponding 95% CIs. Multicollinearity of variables within the model was also assessed through the variable inflation factor (VIF), and no substantial collinearity was found (i.e., VIF <10 for all tests).

Subject characteristics were assessed using t tests or Wilcoxon rank sum tests for continuous variables and χ² tests for categorical variables. Fully adjusted model results are reported. All statistical analyses were performed using SAS, v. 9.4, statistical software. All hypothesis tests were two-sided with an a priori significance level of 0.05.

After applying the inclusion criteria, the final sample size was 14,126 (Figure 1). The distribution of basal insulin products among the subjects was detemir pens, 3,921 (27.8%); detemir vials, 417 (2.9%); glargine pens, 7,700 (54.5%); and glargine vials, 2,088 (14.8%). The mean age for the study population was 56.8 ± 11.1 years, 53.4% of the population was male, and a majority (79.6%) were enrolled in Medicare supplemental insurance. In addition, the mean A1C at baseline for the subset of the population with at least one A1C value (n = 4,121) was 9.5 ± 2.1%, and the mean number of physician visits was 6.9 ± 6.9. Of the 14,126 total subjects, 9,821 (69.5%) were categorized as persistent to basal insulin. A comparison of baseline characteristics between persistent and nonpersistent groups is shown in Table 1.

The bivariate comparison of post-index outcomes between persistent and nonpersistent groups is presented in Table 2. On average, persistent patients had lower A1C values (8.3 vs. 8.6%, P <0.001), fewer hospital admissions (0.2 vs. 0.3, P <0.001), fewer ED visits (0.4 vs. 0.6, P <0.001), lower diabetes-related medical costs ($8,062 vs. $11,884, P <0.001), and lower all-cause total health care costs ($23,311 vs. $25,412, P = 0.010) compared with nonpersistent patients. Conversely, there were no differences in the total number of hypoglycemia events (P = 0.415) across groups; however, there was a difference between the groups in having at least one hypoglycemia-related event (P = 0.039).

Multivariable regression model results are presented in Figure 2. Persistence to basal insulin was associated with lower odds of experiencing an ED visit (OR 0.780, 95% CI 0.717–0.848) or an all-cause hospital admission (OR 0.739, 95% CI 0.671–0.848). Persistence was associated with lower diabetes-related medical costs (CR 0.710, 95% CI 0.671–0.750) but not all-cause total health care costs (P = 0.078). Persistence also was not significantly associated with glycemic control, defined as A1C values <8 versus ≥8% (P = 0.454) or having at least one hypoglycemia-related event (P = 0.076).

This study examined the association between persistence to basal insulin therapy and clinical and economic outcomes among individuals with type 2 diabetes. Findings from this study showed that, compared with subjects who were nonpersistent to basal insulin, those who were persistent to basal insulin had 22% lower odds of ED visits, 26% lower odds of hospital admissions, and 29% lower cost of diabetes-related medical expenditures.

The associations of medication persistence with health care resource utilization and economic outcomes are consistent with other studies in subjects with diabetes. A study by Perez-Nieves et al. (19) showed that persistence to basal insulin was associated with lower medical costs, fewer ED visits, and shorter hospital stays, and these findings were corroborated by another study in the literature (20). Differences in all-cause total health care costs approached statistical significance (P = 0.078), and their lack of significance may have resulted from the impact of potential social risk factors that were not identifiable in administrative claims. Previous research has shown that social risk factors such as food, financial, housing, and utility insecurity can adversely affect diabetes management and expenditures (21).

Similar to a study by Wei et al. (15) examining persistence to basal insulin, the current study found statistically significant differences in mean A1C but no association between persistence and hypoglycemic events. However, the difference in A1C was small (8.3 vs. 8.6%). Although this difference reached statistical significance, it may not be clinically significant. The study only looked at outcomes for 12 months post-index. A larger, more clinically significant difference may have been identified after a longer duration of insulin persistence.

The findings regarding hypoglycemia are important because one of the factors cited by individuals with diabetes regarding their hesitance to use insulin is the risk of hypoglycemia (22,23). The findings from this current study, corroborated by those of Wei et al. (15), demonstrate hypoglycemia should not be a barrier to basal insulin persistence.

Given the significant consequences of nonpersistence and the relationship between medication use and health outcomes, it is appropriate to evaluate the complete array of diabetes medications. The current Medicare Part D Star Ratings diabetes medication measure excludes individuals using insulin, who represent 30% of the population with type 2 diabetes. This exclusion leaves a significant measurement gap for persistence with insulin therapy. Based on the results of this study, as well as review of other literature, PQA developed and endorsed a health plan measure evaluating persistence to basal insulin using the Wei methodology (24). CMS added the measure for persistence to basal insulin to the publicly reported Medicare Part D display page for 2024 and 2025 (25).

An important consideration with the Wei method of estimating persistence is that the determination of persistence is based on a comparison of the individual to the population distribution of refill intervals (15). Using this method, if a substantial portion of the population has discontinued use of insulin, then an individual with “partial persistence” would be considered persistent if they refilled more frequently than the subjects with the worst level of persistence. Therefore, this method of applying different thresholds for persistence may estimate the proportion of patients with optimal medication-use behaviors differently among populations and potentially only identify as nonpersistent those patients with minimal levels of medication use.

The potential for the Wei method to apply different thresholds for persistence may explain a finding in the current study, which found no association between persistence to basal insulin and glycemic (A1C) control. The Wei methodology allows us to determine whether a patient is as persistent to basal insulin as other patients in the population with the same insulin type and similar dosages (15). As a result, the estimated refill periods used to determine persistence may differ from one population to another based on the patient composition, basal insulin type, and dosages available within the study. Therefore, it is possible that individuals who were potentially nonpersistent in the Wei study population were included in the persistent group in the current study population because of these differences, which may have reduced the impact of persistence on A1C control, resulting in findings that were not statistically significant. However, this possibility also means that, as the population persistence increases, the 90th percentile also increases, thereby increasing the threshold for an individual to be considered persistent. Thus, the use of the Wei method in a quality measure may raise the threshold for persistence over time and ensure that, ultimately, those who are truly consistent with their insulin regimen are included in the persistent group. Future studies should explore basal insulin persistence trends over time using the Wei method and the subsequent impact of those trends on A1C control. Although no measure of medication persistence may be a perfect reflection of an individual’s medication use pattern, the Wei method is a reasonable approach to track population trends in the appropriate use of basal insulin.

Limitations common to studies using administrative claims data apply to this study. For example, filling a prescription does not mean the individual took the medication as prescribed. For this study, the effect of this limitation may be even more pronounced given that it is not possible to ascertain the specific dosage adjustments that a patient could make, resulting in an imperfect estimate of the ideal refill date for the insulin prescription. However, the use of the 90th percentile within the study population helps to mitigate this issue. Claims data also have limited information on social and functional risk factors such as income, race/ethnicity, and functional status, which may have impacts on the study outcomes. In addition, clinical factors such as hypoglycemia may be under-coded in claims. For these reasons, the results of the current study may underestimate the occurrence of hypoglycemia within the study population. Despite these limitations, administrative data are a reliable source for estimating medication persistence for chronic-use medications and identifying patients at risk for therapeutic failure when it is not possible to directly measure patient persistence to medications (26).

Although the MarketScan data include important information on laboratory test values not usually found in administrative claims databases, these values are only provided for a subset of the population, as was the case with the A1C analysis in this study. Future studies should examine to what extent, if any, subgroups with laboratory values differ from the broader MarketScan population and whether any identified differences affect A1C analyses.

Another limitation is the increase in prices of basal insulins over the past several years, which may increase the likelihood of cost-related nonadherence compared with the time period of this study. Furthermore, this study used MarketScan data on commercial and Medicare supplemental claims, which are limited to beneficiaries of employer-sponsored plans and Medicare-eligible retirees with employer-sponsored Medicare supplemental plans. Consequently, this study is not generalizable to insured populations covered by other health plans. Further research of this method in other insured populations may be necessary to strengthen the supporting evidence and further assess the usability of this method for health plan performance measurement.

The data on which these analyses were conducted are now several years old; this is a common challenge in research where a data lag of several years is common in observational datasets. As a result, certain changes in the clinical and policy environments since these analyses took place warrant discussion. For example, since 2015, typical insulin prices continued to increase before experiencing a substantial drop for Medicare recipients and beyond in 2023 resulting from out-of-pocket cost caps enacted with the Inflation Reduction Act (27,28). However, these changes are unlikely to fundamentally alter the relationship between persistence to basal insulin using the Wei method and the outcomes discussed. Although changing prices may shift patterns in insulin prescription fills over time (e.g., higher costs may reduce access, which in turn may reduce persistence) and trigger shifts in the 90th percentile, the use of such a high percentile to designate patients as persistent or nonpersistent largely insulates this measurement from external shocks. However, follow-up studies to reexamine the relationships over time in response to a changing policy environment may be useful. Similarly, new basal insulins have come to market since this study. Although there is not an a priori rationale to believe that patients using new basal insulins will demonstrate a different relationship between persistence and outcomes, this is an important area for continuing research.

Patients with type 2 diabetes who were persistent with basal insulin therapy had lower hospitalization and ED utilization rates and lower expenditures on diabetes-related medical care compared with those who were nonpersistent with basal insulin. The results of this study show that an insulin persistence quality measure based on the Wei methodology may be appropriate for evaluating, at the population level, the medication-use patterns of people who are prescribed basal insulin.

This research was supported by Eli Lilly, Novo Nordisk, and Sanofi. The companies were not involved in the design, analysis, or interpretation of the results, nor were they involved in writing or revising the manuscript.

No potential conflicts of interest relevant to this article were reported.

I.N. researched the data, conducted the analysis, contributed to interpretation of study results, and reviewed and edited the manuscript. P.J.C. and M.A.P. contributed to interpreting the study results and to writing, reviewing, and editing the manuscript. L.E.H. and M.P. contributed to conceptualizing the study, interpreting the study results, and reviewing and editing the manuscript. D.P.N. contributed to conceptualizing the study, interpreting the study results, and writing the manuscript. I.N. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.