A Systematic Review of Insulin Adherence Measures in Patients with Diabetes

BACKGROUND: Diabetes care is associated with a considerable burden to the health care system in the United States, and measuring the quality of health care is an important development goal of the Department of Health and Human Services and the Centers for Medicare & Medicaid Services. Diabetes is a priority disease within the National Quality Strategy and should therefore remain a focus in the measurement of health care quality. Despite the importance of measuring quality in diabetes care management, no quality measure is currently associated with adherence to insulin treatment, and measuring adherence to insulin is known to be complicated. OBJECTIVES: To (a) identify methods to measure insulin adherence in patients with diabetes and (b) evaluate whether identified methods could be considered for testing as a quality measure. METHODS: Systematic searches were conducted in the online electronic databases Embase, MEDLINE, and the Cochrane Library, supplemented with additional manual searches to identify publications on insulin adherence from the year 2000 onward. Identified citations were screened for relevance against predefined eligibility criteria, and methods to measure adherence to insulin were extracted from relevant studies into data extraction tables. Methods were critiqued on the feasibility for consideration as a quality measure. RESULTS: Seventy-eight publications met the inclusion criteria and were reviewed. Included studies reported various indirect methods to measure adherence to insulin, using prescription claims or self-report questionnaires. Commonly reported methods included the (adjusted) medication possession ratio, proportion of days covered, persistence, daily average consumption, and the Morisky Medication Adherence Scale. All types of identified methods were associated with measuring challenges varying from accuracy of estimated adherence, complexity of data collection, absence of validated threshold for good adherence, and reliability of adherence outcomes. CONCLUSIONS: Without additional research, none of the identified methods are appropriate for use as a quality measure for insulin adherence. We suggest patient involvement in future research and additional quality measure development.

D iabetes is a chronic metabolic disorder characterized by elevated blood glucose levels (hyperglycemia) and is associated with considerable morbidity and mortality. Type 1 diabetes mellitus (T1DM) is an autoimmune disease in which there is failure by pancreatic islet β-cells to produce insulin. The American Diabetes Association recommends that T1DM be treated with multiple-dose insulin injections (3 to 4 injections per day of basal and bolus insulin), or continuous subcutaneous insulin infusion. 1 Type 2 diabetes mellitus (T2DM) is characterized by insulin resistance with progressive loss of β-cell function and subsequent insulin deficiency. Risk factors for T2DM include obesity, diet, sedentary living, family history of diabetes, and race/ethnicity, to name a few. T2DM is a progressive disease and treatment is intensified with disease progression. Many patients with T2DM will eventually require insulin to control their blood glucose levels, and many patients will need to intensify their treatment regimens over time. 2 Treatment adherence is critical to effective management of T1DM and T2DM. Adherence to insulin treatment has been associated with improved levels of glycemic control, 3 which reduces the risk of developing micro-and macro-vascular complications 4 and all-cause mortality. 5 However, poor adherence to insulin regimens is common in people with diabetes. 6 The 2 most commonly reported difficulties with insulin treatment, as reported by both patients and physicians, are the requirements for multiple injections and adherence to prescribed dosing times. 7 Fear of hypoglycemia, 8,9 weight gain, 9,10 and treatment complexity 6-9 are additional barriers to good adherence to

Data Collection and Analysis
Citations identified through the searches were assessed by a reviewer (author Webb) based on title and abstract using the predefined eligibility criteria. Full publications of potentially relevant citations were obtained and examined by 2 reviewers (Webb and author Kroes). Disputes were resolved via discussion with third parties (authors Wisniewski and Stolpe). Relevant data from eligible publications were extracted into a data extraction table by a reviewer (Webb) and verified by a second reviewer (Kroes).
Included studies were evaluated for risk of bias using the Newcastle-Ottawa scale (NOS) 15 : a "star system" to assess the quality of nonrandomized studies on 3 perspectives: selection of study groups; comparability of groups; and ascertainment of exposure or outcome of interest for case-control or cohort studies, respectively.

Study Selection
A diagram illustrating the flow of citations through the SR process is provided in Figure 1. A total of 4,845 citations were identified in the full search. Seventy-eight publications reporting insulin adherence measures in adults (aged ≥ 18 years) with T1DM or T2DM were included in the final dataset. insulin treatment. In 2010, Cooke et al. reported that persistence with injectable insulin 12 months after initiation was as low as 28.7% in a U.S. population of patients with T2DM. 11 With an estimated prevalence of 14.3% in 2011-2012, diabetes costs the United States approximately $245 billion annually. 12 The combination of high prevalence and cost of diabetes has resulted in increased attention to appropriate care and control of patients with diabetes among both public and private payers. This increased attention has manifested in the development and implementation of several quality measures intended to ensure adequate disease control at the population level. Table 1 describes current diabetes quality control measures endorsed by the National Quality Forum (NQF). A Pharmacy Quality Alliance (PQA) measure that addresses adherence to diabetes medications is currently being used within the Centers for Medicare & Medicaid Services (CMS) Star Ratings for Medicare Parts C and D and in the Health Insurance Exchanges Quality Rating System. Through a consensus-based process, PQA endorsed the proportion of days covered (PDC) method, which is currently considered the measurement standard for adherence to oral medications. 13 The PQA diabetes medication adherence measure does not include insulin due to feasibility challenges associated with prescription claims data and injectable medication days supply. This is an important challenge to address because approximately 30% of patients with diabetes, and those with the most advanced disease, are treated with insulin. 14 Following the passage of the Patient Protection and Affordable Care Act in 2010, which aimed to reduce health care costs while preserving or enhancing the quality of health care delivered, it has become increasing apparent that there is a need for a standardized approach to measure insulin adherence.
Given the known complexity of measuring adherence to injectable treatments and the lack of a quality measure that assesses adherence to insulin in federal quality measurement programs, we conducted a systematic review (SR) to understand the evidence base associated with methods to measure adherence to insulin in patients with diabetes. Our purpose was to identify potential methods that could be considered for further investigation as a quality measure suitable for public-facing performance programs and value-based payment structures.

■■ Methods
An SR was conducted to identify studies reporting methods to measure adherence to self-injectable treatments in patients with diabetes, rheumatoid arthritis, multiple sclerosis, or human immunodeficiency virus, or in those requiring human growth hormone. Here we present the subset of findings relating to methods of measuring adherence to insulin in patients with diabetes.   Most studies used retrospective claims cohorts, were conducted in the United States (61 studies), and were industry sponsored. The majority (58 studies) included patients with T2DM, 2 included T1DM patients, 10 included both, and the remaining did not report diabetes type. The mean population age ranged between 37.4 and 74.6 years. Most studies investigated any insulin treatment, with insulin type not specified, and the time period of most analyses was 1 year. The overall quality of the cohort studies was generally high, with many achieving 7 stars when using the NOS. Table 3 lists the adherence measures most commonly reported, their definitions, and areas of divergence between studies. The main areas of divergence were the denominator used when calculating the medication possession ratio (MPR), the definition of the refill gap while measuring persistence, and the additional assumptions used when calculating the PDC (e.g., imputation of missing days supply). Although not generally considered a direct measure of adherence, daily average consumption (DACON) was reported in 12 studies alongside the methods described above, and usually only for those patients established as adherent. All identified methods to measure adherence were indirect, with no studies identified that measured adherence directly through observation of medication taking, return of device used to inject the medication, or biological fluid samples.

Medication Possession Ratio.
Twenty-two studies measured adherence using MPR.  The numerator used to calculate MPR was consistent in 21 studies, 16-19,21-37 whereas the denominator varied. Denominators included number of days between first fill and last fill in 2 studies 17,32 ; between first and last fill plus the days supply of the last fill in 3 studies 19,27,29 ; from first fill date to either the end of the last day of supplied medication during the study period or the end of the study period (whichever came first) in 1 study 33 ; and the analysis period in 15 of the studies. 16,18,21-26,28,30,31,34-37 Table 3 shows a summary of the denominators used. In 1 study, the method used to calculate the MPR was not reported. 20 Adjusted Medication Possession Ratio. An adjusted MPR was used in 11 studies to account for potential difference in pack sizes across pens and vials. 30,31,35-43 Adjusted MPRs were calculated by analyzing the distribution of data by product and method of administration. The observed MPR was multiplied by the ratio of either the mean (N = 10 studies) or median (N = 1 study) time between claims divided by the mean or median pharmacy reported days supply. Five of the studies reported adherence rates using both the traditional and adjusted MPR method. [35][36][37][38]42 Comparison of both methods showed that the adherence increased when measured using the adjusted MPR. [35][36][37][38]42 In Grabner et al. (2013), adherence was similar between pen and vial cohorts when measured using the traditional MPR method; however, when measured according to adjusted MPR, adherence was statistically significantly different between groups (P < 0.001). 42 Proportion of Days Covered. Ten studies measured adherence using the PDC. 44-53 PDC was measured using the traditional method in 8 studies, 44,46-49,51-53 and in 1 study PDC methodology was not reported. 50 In Donnelly et al. (2007), the authors used a method resembling PDC, although it was not specifically termed PDC. 45 Assumptions were reported in several PDC studies. Consistent with the PDC approach, which prohibits the possibility of having a PDC > 1, 4 studies reported appending days of overlap to the previous prescription's end, where overlaps existed. 49,51,53,54 In Lee et al. (2011), data on the days supply were missing and therefore, the average days supply was imputed. 49 In Gibson et al. (2010), if a patient switched medication within a therapeutic class, it was assumed the remainder of the previous prescription was discarded and coverage commenced with the supply of the new medication. It was also assumed that days in hospital were considered adherent days. To account for the potential that the days supply of insulin might not indicate the full extent of coverage, a multiplier to reflect the average time between insulin prescriptions was applied. 48

Other Adherence Measures
In 20 studies, methods other than those described in Table 3 were used to measure adherence (data not shown). 7,76-94

■■ Discussion
This SR was designed to identify studies reporting methods to measure adherence to insulin treatment in patients with diabetes. Previous SRs have studied adherence to diabetes medication, but to our knowledge this is the first to focus specifically on methods to measure adherence to insulin in adults with diabetes for the purpose of consideration for a quality performance measure.

Limitations
The SR was conducted using predefined eligibility criteria and conformed to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, 107

which contributes
In Buysman et al. (2011), a similar method of using the expected time of medication coverage was used to measure persistence. However, in this study, an adjustment factor was used when measuring persistence to account for variations in insulin pack size and dosing schedules. 40 An adjustment factor was calculated as the ratio of the 80th percentile of time between insulin claims divided by the 80th percentile of the pharmacy-reported days supply. The adjustment factor was then multiplied by the actual days supply reported on the pharmacy claim to calculate an adjusted days supply, and patients with a gap of 1 or more days in insulin therapy based on the adjusted days supply were defined as discontinuing therapy. 40 In the remaining 7 studies, patients were considered nonpersistent with therapy if the time without treatment was of a predefined length. 11,17,24,44,55,57,62 In Rashid et al. (2012), this measure was defined as the time to all-cause discontinuation, and a predefined gap of 90 days without any drug supply available was considered discontinuation. 62 The following gaps were used in other included studies: 60 and 90 days, 24 60 days, 44   to the robustness of the conclusions. There were some limitations in the SR methodology. The searches were restricted to identify only those studies published after the year 2000. Searches were limited to 2000 onward, as outcomes research and quality measures for adherence would not have been studied in great depth or available to patients before this date. 108 A factor influencing the generalizability of the data is the inclusion of studies that used databases, such as the PharMetrics database, the Thomson Reuters MarketScan database, and the Innovus IMPACT national managed care database. These studies used specific patient populations, and the data from these studies may not be generalizable to other populations.

Challenges with Measuring Insulin Adherence
There are advantages and disadvantages associated with the different methods to measure insulin adherence, which should be considered when assessing their applicability as a quality measure.
The MPR and PDC methods allow different ways to calculate the supply and possession of medication based on pharmacy claims data. The key strength of the adjusted MPR and adjusted PDC over the traditional MPR and PDC is the ability to adjust for differences in insulin pack sizes, and to correct for the fact that almost all insulin prescriptions are dispensed with a 30-day supply period even though patients may have more insulin than ordered due to variations in their daily dosing. This adjustment allows for a more accurate representation of how a patient adhered to the prescription as written by the prescriber. However, insulin adherence calculations from prescription claims data are particularly challenging, as such calculations are performed based on the days supply entered by the pharmacist during claim submission and adjudication. The days supply used for billing often does not match the clinically appropriate days supply because of issues such as irregular insulin dosing, sliding scale usage, and unit-based dispensing. While the reliability of the days supply primarily persists with pens, assuming a 28-day supply for vials because they should not be used beyond that due to shelf life limitations, is equally problematic from a quality measure perspective, as it is an additional assumption that would need to be made in any analysis.
Persistence can also be measured with pharmacy claims data. As described in the results, similar adjustment factors have been applied to persistence and adherence. The challenge with the use of any adjustment factor, whether associated with MPR, PDC, or persistence, is that the method is not entirely standardized because the adjustment factor is contingent on the distribution of each product and formulation within the population of interest. While this is feasible from the research perspective, it may be problematic when financial incentives or penalties are applied if a given insurance plan does or does not meet the metric.
We suggest that an adjusted MPR, PDC, or persistence analysis would need to be validated before considering use as a performance measure. Additionally, a cross-validation technique would need to be developed that could be applied randomly when used as part of performance measurement to minimize potential for human error due to the nonstandardized approach. In the interim, researchers should perform and report sensitivity analyses based on the distribution of data that informs any adjustments. In the case of PDC or MPR, analyses that adjust for both mean and median days supply and/or analyses that consider the lower and upper bounds of the 95% confidence interval when the mean is applied may be appropriate. For persistence, sensitivity analyses that consider a minimum 10% variation in the percentile of time period analyzed may be appropriate.
The insulin consumption as measured with DACON accounts for dose consumed; however, a limitation specifically with DACON is the lack of a validated threshold for good adherence. A large population with inter-and intrapersonal variability in insulin types and dose requirements complicates generalization of consumption requirements. Whereas DACON is traditionally measured over a persistent time period, it may be worth investigating whether a redefinition of the time period of interest is warranted in this context. For example, if DACON were calculated over a calendar year of interest rather than a persistent period of interest, it may be possible to identify a minimum threshold that correlates with both persistency and diabetes control (measured by hemoglobin A1c). While this hypothesis would need to be tested before further consideration of a revised DACON as a means to approximate adherence for performance measurement purposes, the objectivity of the DACON value is reassuring.
The measures using pharmacy claims data may be more reliable than those recorded from self-report questionnaires (such as MMAS), as the latter may be subject to reporting bias, and those patients reporting their medication consumption tend to overreport their adherence for various reasons. [109][110][111] There is an added complexity associated with the implementation of selfreport questionnaires for quality measurement, as they require the need to survey either an entire population, which is likely not feasible, or they require appropriate adjustment to ensure plan representativeness if a random subsample is selected. At a plan level, this would require additional resources that may limit the uptake of the measure and will not provide patient adherence outliers for quality improvement interventions.
Claims data provide a practical approach to adherence calculation for quality reporting and are the current basis for the PQA diabetes metric used in the Medicare Star Ratings adherence calculation. The majority of included studies used a 1-year post-index period as follow-up time, and an adherence quality measure could use a calendar year as measuring period. However, using any of the calculation methods identified in this SR, pharmacy claims data may not accurately represent the medication supply and currently cannot reflect the appropriate timing and dosing of insulin administration. For this reason, we suggest that, in the long term, consideration is taken to modify database systems used to adjudicate reimbursement. While it is well understood these systems were borne out of a need to track and reimburse insurance claims, they have evolved to serve as research tools and analytical platforms for quality measurement. Allowing pharmacists to enter accurate clinical days supply for products, like insulin, in which there is variation between the reimbursed supply and actual supply available to a patient would enable tracking and reporting of insulin adherence. This tracking and reporting would therefore align the evolution of systems with that of health care, which can only serve to achieve the goal of managing costs while maintaining superior quality of care for the U.S. population.

Role of Quality Measures in U.S. Government Health Care Reform
Currently, the push for improving quality in health care is accelerating in the United States, and this has numerous implications for measuring adherence in diabetes. Plans from the Department of HHS include quality measure initiatives with a focus on improved outcome measures, patient experience measures, and other metrics such as care coordination, access, and delivery. These plans will consider gaps in quality measurement and applicability of measures across health care settings. Currently, CMS considers measures of adherence intermediate outcome metrics-surrogate measures for disease outcomes metrics. 112

Call for an Insulin Quality Measure
Diabetes is one of the diseases targeted as priority by the National Quality Strategy and as such should continue to be a focus of quality measure development. 113 Currently, several quality measures endorsed by NQF for diabetes include adherence measures, but not insulin adherence (Table 1). This gap is important, as a considerable proportion of patients with diabetes, particularly those with advanced disease, use insulin. A quality measure for appropriate use of insulin should therefore be considered.
Because improved disease outcomes (i.e., glycemic control, A1c levels) are considered of highest priority, and patients are the center of care delivery for insulin treatment, a link between outcome and insulin usage would be a relevant target for a quality measure, as it could serve as a gauge to understand the degree to which the health care team is able to partner with patients in improving their health-related engagement and activation. By addressing patient-centric barriers with adherence, specifically with insulin, costs due to microvascular complications may be reduced in the long term. Therefore, patient engagement and education will be essential to help develop and implement an insulin adherence quality measure. As part of the development process, additional studies confirming the validity of the new measure should be conducted in order to reach consensus.

■■ Conclusions
This SR identifies a range of possible methods to measure adherence to insulin using pharmacy claims or patientreported questionnaires. We conclude that, while there is some potential based on what has previously been developed, without additional research or modification of current databases used to track claims and measure performance, none of these are currently appropriate for use as a quality measure for insulin adherence. The relevance of insulin adherence as an intermediate outcome for diabetes management is of high priority in the context of health care reform plans with measurable goals. Studies confirming the soundness of new measurement methodology should be conducted.

DISCLOSURES
Novo Nordisk paid DRG Abacus to complete the systematic review and manuscript and was involved in the study design, interpretation of data, and decision to publish the findings of the systematic review. Kroes and Webb report personal fees from Novo Nordisk during the conduct of the study and personal fees from DRG Abacus, outside of the submitted work. Webb is employed by DRG Abacus, and Kroes was employed by DRG Abacus at the time of this study. Wisniewski is an employee of Novo Nordisk, which funded the systematic review reported in this article, and also owns stocks in Novo Nordisk. Stolpe has nothing to disclose.
Study concept and design were contributed by Kroes, Webb, and Wisniewski, with assistance from Stolpe. Webb took the lead in data collection, along with Kroes, and data interpretation was performed by all the authors. The manuscript was written by Kroes, Webb, and Wisniewski, with assistance from Stolpe, and revised by Kroes, Stolpe, Wisniewski, and Webb.