Joshua N. Liberman, PhD

The full therapeutic benefit of prescription medications is realized only when the patient follows the healthcare provider’s recommendations, a concept referred to as medication adherence. Premature treatment discontinuation and other forms of poor medication adherence are common and contribute to adverse drug events, worsening disease, increasing healthcare costs, and decreasing quality of life.

 As a result, there is a wealth of scientific evidence documenting risk factors for poor adherence and effectiveness of interventions to improve adherence.

 Yet these studies often rely on secondary data sources, typically administrative pharmacy claims, to identify patients and measure adherence. Though a powerful source of prescription drug information, administrative data can profile only the medication-taking behavior of patients who have already received therapy, and thus have filled at least 1 prescription.

In contrast, relatively little is known about the extent and causes of primary nonadherence, prescriptions given by a healthcare provider but never filled by the patient. Without an electronic record, connecting physicians’ handwritten or oral prescriptions with medication-dispensing records is manual and resource intensive. Thus, studies of primary adherence have relied primarily on patient self-reports or on patient populations receiving prescriptions while hospitalized or receiving emergency care, where computer prescription order entry systems exist. These studies routinely report adherence rates across multiple pharmacotherapies, represent the experience of a single pharmacy, enroll only select patient populations, or report the experience of a non-US patient population. As a result, estimates of primary nonadherence vary widely, with international studies reporting rates from 2% to 30%

 and US studies reporting rates from 2% to 20%.

Electronic prescribing (e-prescribing), a solution formerly available only in hospital settings that had an integrated technology platform connecting physicians with pharmacies, is now gaining acceptance in ambulatory care settings.

 With the increasing use of e-prescribing, identifying and tracking large volumes of unfilled prescriptions have become practical. Horizon Blue Cross Blue Shield of New Jersey (Horizon) in conjunction with its pharmacy benefits manager, CVS Caremark, measured the rate and determinants of initiation failure among individuals receiving initial prescriptions for asthma-controller and dyslipidemia agents.

In 2003, Horizon launched an initiative to equip providers with e-prescribing tools, enabling them to have electronic access to patient formulary and prescription drug history information at the point of care. In this retrospective cohort analysis, prescriptions written via iScribe, the CVS Caremark e-prescribing tool accessed and implemented through a handheld personal digital assistant or via the Internet, were compared with administrative pharmacy claims. Prescriptions qualified for comparison if they were prescribed (1) by 1 of 507 providers participating in this initiative; (2) for a target drug (ie, an asthma-controller or dyslipidemia medication) (Table 1); (3) between January 1, 2005, and October 31, 2006 (the prescribing period); (4) for a patient younger than age 65 years with full pharmacy benefit coverage from 180 days before (the preperiod) to 60 days after (the follow-up period) the prescribing date; and (5) for a patient with no paid claims in the preperiod for a drug in the applicable treatment category.

All qualifying prescriptions (prescribing events) that met the stated criteria were included. Asthma-controller and dyslipidemia therapies were selected as they represent chronic maintenance medication groups with substantial utilization (providing a sufficient sample size for study) that are routinely included in health improvement, disease management, and adherence improvement efforts. Eligibility was restricted to treatment initiators because primary nonadherence in this population would not be captured by currently available adherence measurement studies. Qualified prescriptions were compared against pharmacy claims inclusive of prescriptions dispensed between January 1, 2005, and December 31, 2006, to accommodate follow-up periods for all prescribing dates.

Several data sources were used to define qualified prescribing events and to compile the analytical data set. Electronic prescribing records were the source of the healthcare provider name, patient name, prescribed drug name, prescribing date, and prescription plan benefit identification numbers. Provider reference files were the source of prescriber degree (doctor of medicine [MD] or doctor of osteopathy [DO]) and specialty (internal medicine, family/general practice, pediatrics, pulmonary/allergy/immunology). Paid pharmacy claims were the source for identifying filled prescriptions.

For analysis, patient age was grouped into a pediatric group (0-18 years), younger adult group (19-44 years), and 2 older adult groups (45-54 years and 55-64 years). Household income was estimated from US Census projections by zip code and categorized (≤$41,000; $41,001-$51,000; $51,001-$61,000; $61,001-$71,000; and ≥$71,001). Prescription copayments (<$10, $10 to <$15, $15 to <$20, $20 to <$25, and ≥$25, which were based on the natural peaks observed for this client) were assigned to the prescribed medication using the generic, preferred brand, or nonpreferred brand indication as of the prescribing date. For flat-dollar copayment benefit structures, prescription copayment values were taken directly from the benefit; for percentage-based copayments, copayments were determined by applying the percentage to the estimated drug cost. As an example, a 10% coinsurance rate on a $50 drug would result in a $5 copayment. Drug cost was calculated from the monthly values of cost per quantity by drug name, strength, and quantity. As formulary tier assignment was not captured, it was inferred from copayment differentials. Finally, a count of paid pharmacy claims during the preperiod quantified the patient’s recent drug utilization experience (none, 1-2 paid claims, 3-8 paid claims, and ≥9 paid claims).

To determine whether a written prescription had been filled with a clinically equivalent drug, paid pharmacy claims data from the follow-up period were compared with the prescribing data file. Time-to-fill was measured as the difference between the initial dispensing date and the prescribing date. As generic substitution and therapeutic interchange are common pharmacy practices, lists of clinically equivalent drugs for each class were derived by clinical consensus.

Distributions of key variables by fill status were calculated. Cumulative fill rates were calculated as the sum of all fills divided by the number of applicable prescribing events. Time-to-fill and fill rates by member age were graphed by treatment category. Prescribing event fills were used as the dependent variable in logistic regression models for single and combined class analyses with the groups created for age, sex, prior period claims experience, provider specialty and degrees, household income, copayments, and tier type as the independent variables. The independent variables were included as dummy variables in separate models for each disease and a combined model. Because the members’ age distribution differed between the 2 conditions, the individual disease-specific models tested the effects of different age categories. As the disease-specific models produced similar results (for age and other covariates), we combined age groups and only present the overall model results. The disease-specific models are available in the eAppendix at 

. Analyses were performed using SAS version 9.1.3 (SAS Institute, Inc., Cary, NC). Differences were assessed for significance within the logistic models or using χ2 tests, including tests of filled/unfilled for the variable groups.

The 2243 qualified prescribing events represented 1182 asthma-controller and 1061 dyslipidemia events and were written for 2116 patients (Figure 1). Of the asthma-controller events, a total of 357 (30.2%) were attributed to patients younger than age 19 years and 378 (32.0%) were attributed to patients age 19 to 44 years. For the dyslipidemia prescribing events, only 185 (17.4%) were for patients younger than age 45 years. Among all the prescriptions written, 52.2% were for women (57.3% of asthma-controller prescriptions, 46.5% of dyslipidemia prescriptions). Nearly 30% of dyslipidemia and 18.6% of asthma-controller prescriptions were written for patients with no recent history of prescription medication use. A total of 1733 prescriptions were written by doctors of medicine, 510 were written by doctors of osteopathy, and the remaining 78 were written by nonphysicians. General practitioners, family practitioners, or internal medicine physicians prescribed 77.8% of medications, varying from 61.2% of asthma-controller medications to 86.5% of dyslipidemia medications.

Fill rates varied over time and by several of the characteristics studied. After 60 days, a total of 240 asthmacontroller prescriptions (20.3%) and 362 dyslipidemia prescriptions (34.1%) went unfilled (Table 2, Figure 2). Though 67.9% of asthma-controller and 43.2% of dyslipidemia prescriptions were dispensed on the prescribing date, these prescriptions represented nearly 85.1% and 75.5% of the total dispensed in the follow-up period. Of those ultimately filled, 95.0% of asthma-controller and 93.2% of dyslipidemia medications were filled within 2 weeks of the prescribing date. The unadjusted age-specific fill rates only differed significantly for the older age group receiving dyslipidemia prescriptions, where the fill rate dropped to 59.9% for patients age 55 to 64 years (Table 2). In both cohorts, patients’ sex was not associated with filled prescriptions. For dyslipidemia prescriptions, the 60-day fill rates differed significantly by prescriber degree (DO 72.6% vs MD 64.0%), whereas the degree did not matter for asthma-controller prescriptions, with fill rates around 80%. No significant differences were found for fill rates by specialty.

After adjustment for numerous provider, patient, and plan design characteristics, only patient age, recent history of pharmacy claims, and out-of-pocket cost for the prescription (copayment) remained significant predictors of filling a prescription in the combined class analysis, with similar but unreported results in the single class results. Compared with the youngest age group (age ≤18 years), the probability of filling a prescription decreased with each successive age group, reaching statistical significance among patients age 45 to 54 years (odds ratio [OR] = 0.55) and age 55 to 64 years (OR = 0.45). Recent pharmacy claim history also showed a consistent relationship, with ORs increasing from 0.09 (for individuals with no recent claims) to 0.65 (for individuals with 3-8 recent claims) compared with individuals who had 9 or more claims. Compared with prescriptions with estimated copayments of $25 or more, only the lowest cost prescriptions (ie, those with copayments of <$10) had significantly higher odds of being filled (see eAppendix at 

Our results indicate that about 20% of asthma-controller and 34% of dyslipidemia prescriptions written to initiate therapy went unfilled and that patient out-of-pocket costs, age, and recent use of prescription medications were significant predictors of primary adherence. To our knowledge, no studies have directly assessed failure to initiate prescribed chronic disease therapy by matching electronic prescription records with pharmacy claims in an ambulatory care population. Yet there is a wealth of indirect evidence against which to compare these results. Most of that evidence is derived from studies with significantly different designs, patient populations, follow-up time, or drug mixes.

Rates of first fill adherence with various medications prescribed during visits to or on discharge from an emergency department vary from 65% to 99%.

 Among studies of community-based pharmacies that focused on a broad range of prescription drugs, 1 reported prescription fill rates of 87%,

 whereas several others reported rates in excess of 96%.

 From a population-based survey of 3926 residents in the southeastern United States who had seen a healthcare provider in the past year, Wroth and Pathman

 reported that 894 individuals (21.6%) responded affirmatively to the question, “In the past 12 months, did you delay filling a prescription or not get it at all?” In a comparative study of 9 national surveys, Kirking et al

 reported on medication underuse, which includes primary nonadherence, with estimates that varied from 2.4% to 27%. Finally, Kennedy et al

 reported that 4.4% of Medicare beneficiaries failed to fill or refill 1 or more prescriptions in 2004. Esposito et al

 reported that 8.0% of individuals who used military pharmacies reported failing to claim at least 1 prescription in the previous year.

One study assessed primary adherence with dyslipidemia medications specifically. In a study of 71 veterans identified in a primary care setting, Mann et al

 reported that 10% hadn’t filled a statin prescription as reported in a 6-month follow-up survey. A few studies assessed fill rates for asthma and respiratory medications. In a rural Australian patient population, Watts et al

 reported that only 251 of 359 (70%) respiratory medication prescriptions were dispensed and retrieved. In a medical records review of 32 pediatric Medicaid patients, Bronstein et al

 reported that 30.3% and 30.9% of preventive and acute respiratory prescriptions, respectively, went unfilled after a full year. In a recent study, Williams et al15 studied inhaled corticosteroid fill rates among 1064 asthma patients, comparing electronic prescription records with pharmacy claims. In this population, 92% of patients filled their inhaled corticosteroid prescription within 90 days (83% within the first 7 days).

Previous studies reported that primary adherence rates vary by numerous patient, prescriber, and treatment characteristics, including patient age and sex,

 therapeutic class of prescribed medication,

 patient—physician relationship and patient satisfaction with care,

 and insurance coverage. In our study, noncompliance was more common for dyslipidemia therapy than asthma-controller therapy. That result was consistent with previous findings in which illness burden,

 patient perception of the consequences of treatment avoidance,

 and lower patient motivation to treat asymptomatic conditions all were implicated as causes of nonadherence.

Our study participants had pharmacy insurance; however, primary adherence rates were significantly higher if the member’s out-of-pocket cost for the prescription was less than $10, a result that is consistent with self-reported reasons for primary nonadherence,

 and with studies of secondary adherence that consistently report lower adherence with higher out-of-pocket costs.

 With the exception of the study by Ekedahl and Månsson,

 who reported marginally higher nonadherence rates among the elderly, each study referenced above indicates a higher primary adherence rate among older adults. Yet our results indicate an inverse relationship between age and primary adherence, with progressively lower rates among patients age 45 to 54 years and 55 to 64 years compared with patients age 18 to 44 years. It is particularly interesting to note that this inverse relationship persists after adjustment for numerous potential confounders, especially recent pharmacy claims.

As there is a positive correlation between prescription medication use and age, it is possible that the generally poorer health status of older adults and greater use of the healthcare system are the actual drivers of adherence behavior in older aged populations. We speculate that this relationship exists because a history of filling prescriptions indicates a general willingness to fill prescriptions as well as a higher level of illness burden, which can translate to an increased motivation to initiate and remain on treatment.

 of respiratory medication fill rates, patients with a recent history of short-acting beta-agonist use were more likely to fill their prescription, a result consistent with our findings, though our results were not limited to medications indicated for the treatment of similar health conditions.

There are limitations to the current study. First, administrative pharmacy data are surrogate measures for medication adherence and do not guarantee that the patient consumed the medication. Second, a 6-month washout period is insufficient to state with certainty that all eligible prescriptions were intended to initiate therapy. However, exploratory analysis on this topic suggests a low misclassification rate (<5%), and extended washout periods reduce sample sizes by excluding members with less continuous benefit eligibility. These results also were limited to the experiences of a select prescriber population who were identified and targeted for iScribe adoption based on prescribing activity. The study also was limited to only 2 chronic medication classes, which had substantial variation in primary adherence. We would not recommend generalizing these results to other therapeutic classes. Further, the study population excluded participants age ≥65 years; a conscious limitation to prevent the introduction of Medicare Part D from affecting our adherence results. Finally, there are several potential predictors of nonadherence that we could not account for with e-prescribing and paid pharmacy data, such as burden of illness, patient—physician communications, and patient knowledge, attitudes, and beliefs. Future research should include both older adults and other chronic medication classes. Results of such studies would have significant policy implications as the Centers for Medicare & Medicaid Services evaluates e-prescribing requirements and standards.

As healthcare costs continue to increase, greater focus is being placed on the costs associated with chronic conditions. One known way to mitigate costs and associated clinical effects is through medication management. The first step in appropriate medication management is initiation of the correct treatment. It is essential to ensure that patients initiate treatment, especially for the management of chronic health conditions, and our results confirm that a substantial proportion of chronic medication prescriptions go unfilled.

Our results provide at least 3 insights that may translate to improved interventions and educational opportunities. First, a rescription was likely to be filled quickly or not at all, a fact that ought to improve the timing of interventions. Second, patient age and pharmacy history are easily acquired and could support physician education and point-of-care information to assist prescribers with the removal of treatment barriers faced by patients. Finally, the evidence that only prescriptions with an out-of-pocket cost of less than $10 per prescription were most likely to be filled should inform prescription benefit designs intended to maximize adherence to therapy.

Articles

This study linked e-prescribing records to paid pharmacy claims to identify primary nonadherence rates and factors associated with unfilled prescriptions.

Determinants of Primary Nonadherence in Asthma-Controller and Dyslipidemia Pharmacotherapy

	Among the commercially insured, pharmacy benefits define the availability, accessibility, and cost of prescription drugs. Routinely, total drug costs are paid by both the insurance provider and the plan participant, with the plan participant’s share of the drug price often determined by tiered (or escalating) copayments for generic, preferred brand, and non—preferred brand drugs.1 To address growth in pharmacy expenditures, insurance providers have steadily increased copayments and adopted other utilization management strategies, including prior authorization, step-therapy requirements, and limits on the quantity of medication dispensed.2,3 

	Although effective at controlling pharmacy trend (yearover- year increases in pharmacy expenditures), prescription drug benefits that shift greater costs to consumers or increase barriers to access can result in lower rates of drug treatment, poorer adherence, and increased rates of treatment discontinuation. 4-7 Unfortunately, when pharmacy benefit designs reduce the use of essential therapies (eg, antidiabetics), reduced adherence may have deleterious effects on health status.8-10 

	To counteract this effect, some insurers have implemented value-based insurance designs (VBIDs).11 Value-based insurance designs selectively lower or even eliminate copayments for highly effective preventive medications prescribed for the management of chronic health conditions in an effort to improve medication adherence, improve health outcomes, and lower healthcare costs.11,12 Hence, insurers are increasingly drawn to VBID for medications indicated for prevalent health conditions for which empirical evidence suggests that treatment adherence will improve outcomes. 

	To date, few studies have investigated the effect of changing from a tiered formulary to a VBID with a corresponding drop in copayments for all drugs in key therapeutic classes. Moreover, none of the existing studies have been large enough to evaluate initiation of therapy and persistency of therapy. We examined the impact of implementing a VBID on adherence to and utilization of insulin and oral antidiabetic pharmacotherapy in a large cohort of patients.

	We conducted a retrospective pre—post controlled study, comparing commercially insured plan participants whose pharmacy benefit design changed from a traditional, 3-tiered formulary to a VBID with a random sample of participants drawn from plans that experienced no change in the traditional pharmacy benefit design. 

	Identification and Selection of the Study Population

	The VBID cohort was comprised of age-eligible plan participants from 3 CVS Caremark clients that implemented a VBID on January 1, 2007. Plans with concurrent educational intervention or disease management programs provided by CVS Caremark were excluded to avoid potential confounding intervention effects. A contemporaneous control group was selected at random from plan participants with extant tiered prescription drug plans. We developed an external control group selection procedure that matched the control group to the VBID group on different characteristics such as adjudication platform, client type (ie, employer, health plan, Medicare, Medicaid), maximum allowable benefit, deductible, mail or retail coverage, and mean out-of-pocket costs at the group level. After the groups were selected and matched, plan participants in the VBID group were individually matched to participants in the control groups by sex (male, female), age group (<40, 40-64, 65+ years), and geographic region (Northeast, South, Midwest, West, other). 

	Eligible study participants were continuously eligible for pharmacy benefits for the 12 months before and after January 1, 2007 (the VBID implementation date). The control group comprised plan participants who were continuously enrolled in traditional, 3-tiered formulary plans during the study baseline and who remained in such plans in the follow-up period.

	As of January 1, 2007, participants in the VBID group had the cost of tier 1, generic antidiabetic medications lowered from $15 to $0 per 30-day prescription. For tier 2, preferred brand drugs, patient cost-sharing levels were lowered from $30 in 2006 to amounts from $10 to $15 (depending on the drug/client) in 2007. Users of tier 1 and tier 2 medications in the control group, however, had no changes in their copayment of $10 during the study period. The copayment for non—preferred branded agents (tier 3) was between $30 and $35 in both groups throughout the baseline and follow-up periods. The average copayment for insulin decreased from $28 to $0 for VBID participants in the postchange period, whereas traditional participants faced an increase from $23 to $28 in the same time period. The average out-of-pocket cost for oral antidiabetic drugs did not change for the control group at $18, but for the VBID group the cost for oral agents decreased from $21 to $10. 

	Antidiabetic medication use was categorized by therapeutic classes identified by either 4- or 6-digit Medispan Generic Product Identifier (GPI) codes for oral antidiabetic agents (antidiabetic combinations [GPI 2799], biguanides [GPI 2725], the thiazolidinedione [TZD] class of insulin-sensitizing agents [2766xx], and sulfonylureas [GPI 2720]) or insulin (GPI 2710). Both groups were further categorized into (1) initiators—participants with no pharmacy claims in the therapeutic class between July 1 and December 31, 2006, and with an index prescription for an antidiabetic medication between January 1 and June 30, 2007; and (2) continuers/ discontinuers&mdash;plan participants who had at least 2 paid claims in the therapeutic class of interest during the baseline period and available days supply within the final 45 days to December 31, 2006. Note that initiators included those using 1 antidiabetic agent who added another.

	To assess changes in medication use, we examined administrative pharmacy claims from January 1 through December 31, 2006, the baseline period, and utilization in the first year in which the VBID was implemented (January 1 through December 31, 2007), the follow-up period

	We assessed medication adherence by determining treatment initiation and discontinuation rates and compliance. We defined the initiation rate as the number of plan participants who started therapy in the follow-up period divided by the total number of plan participants with no history of antidiabetic medication use during the baseline period. We defined the discontinuation rate as the proportion of all antidiabetic users with a lapse of at least 60 days following the exhaustion of days supply, including those users with 2 antidiabetic agents who discontinued 1 agent. Plan participants who changed from one antidiabetic therapeutic class to another during the follow-up period were excluded from this calculation. 

	We defined compliance using the medication possession ratio (MPR), calculated as the sum of the days supply of medication divided by the number of days between the first fill and the last refill, excluding the days supply of the last refill. The MPR scores were calculated among continuers only, were truncated at 100% for plan participants who consistently refilled early, and were calculated annually for calendar years 2006 and 2007 to determine changes in compliance (MPR difference). For plan participants on multiple antidiabetic classes (eg, metformin and insulin), MPR scores were a weighted average of class-level MPRs, weighted by days of therapy (in the drug class) in the follow-up period.

	The impact of implementing a VBID on medication adherence was analyzed by comparing the MPR, changes in MPR, and initiation and discontinuation rates between the 2 study populations. Analyses of variance and t tests were used for analyzing continuous data; the Pearson c2 test was used for categorical data. To adjust for population differences in age and sex, logistic regression models were constructed to assess the probability of discontinuing therapy, and linear regression models were constructed to predict follow-up MPR among initiators and difference in MPR (follow-up minus baseline) among continuers. When multivariable methods did not influence the relationship between benefit design and adherence (eg, linear regression models), only unadjusted results were reported. Data were analyzed using SAS version 9.1 (SAS Institute Inc, Cary, NC). All statistical tests were conducted at the P <.05 significance level.

	This research was conducted in compliance with applicable laws governing the use and disclosure of protected health information, including the privacy regulation issued pursuant to the Health Insurance Portability and Accountability Act of 1996.

	There were a total of 20,173 and 190,889 plan participants in the VBID and control groups, respectively (Table 1). The groups were similar with regard to age and sex. Nearly 10.3% of the VBID group and 8.1% of the control group used antidiabetic therapy during the baseline year, with higher rates of both insulin use (3.2% vs 1.8%; P <.01) and oral antidiabetic use (9.5% vs 7.2%; P <.001) among VBID participants. These treatment levels were appropriate and within the estimated 10.7% prevalence of diagnosed and undiagnosed diabetes among people age 20 years or older in the United States.13 However, the VBID population had substantially lower baseline MPR scores for insulin (MPR >80% for 52.8% vs 63.5%; P <.001) and for oral antidiabetic agents (MPR >80% for 53.8% vs 67.6%; P <.001).

	During the follow-up period, therapy initiation rates among VBID plan participants were significantly higher than those among controls (insulin, 0.8% vs 0.4%; any oral antidiabetic drug, 2.3% vs 1.4%) (Table 2), but were within the national average, based on the annual age-adjusted incidence (or new cases) of diabetes of 5.0 to 12.8 per 1000 persons.14 The higher rates of initiation in the VBID group were noted in each therapeutic class, overall, and among each age and sex group. Furthermore, among VBID initiators, the subsequent rate of discontinuation was significantly lower in each therapeutic category compared with controls (Table 2). After adjusting for differences in age and sex, there were no differences in the likelihood of discontinuation in either TZDs or sulfonylureas. However, initiators in the control group had significantly higher odds of discontinuing metformin (odds ratio [OR] = 1.7; 95% confidence interval [CI] = 1.2, 2.2), oral antidiabetic combinations (OR = 2.5; 95% CI = 1.5, 4.3), and insulin (OR = 1.9; 95% CI = 1.3, 2.8), which translated to significantly higher odds of discontinuing any oral antidiabetic therapy and all antidiabetic therapy overall (Figure). In other words, an OR greater than 1, if statistically significant, indicated that therapy discontinuation was more likely in the control group, and an OR less than 1 indicated that discontinuation was less likely in the control group. Furthermore, there were no significant differences in average MPR scores for either insulin or the oral antidiabetics during the year subsequent to initiating therapy (Table 2). Among continuers, VBID plan participants were significantly less likely than controls to discontinue insulin therapy in 2007 (33.1% vs 45.8%; P <.001); however, there was no difference in the discontinuation rates of oral antidiabetic therapy (Table 2). These results persisted after adjustment for age and sex (Figure). 

	Adherence to antidiabetic therapy overall as measured by changes in MPR increased significantly (4.9% points; P <.001) in the VBID group (MPR difference 4.9%; SE = 0.007), whereas plan participants in the control group who remained in tiered plans decreased in the same period (MPR difference −2.3%; SE = 0.004). Improvements in MPR also were observed across the 2 therapeutic classes in the VBID group compared with the control group: insulin users (MPR difference 8.2% vs −1.7%) and oral antidiabetic users (MPR difference 2.2% vs −2.8%) (Table 2). These results persisted after adjusting for age, sex, and mail-order utilization.

	Among patients with diabetes, glycemic control is achieved and maintained with proper diet and regular exercise, and often is supported by long-term use of prescription medications.14,15 The American Diabetes Association standards of medical care in diabetes recommend monotherapy as an adjunct to diet to lower blood glucose in patients with type 2 diabetes (non—insulin dependent) or in combination with a sulfonylurea, metformin, or insulin when diet plus monotherapy do not result in adequate glycemic control. However, patients routinely delay or underuse antidiabetic pharmacotherapy.15,16 Although there are numerous reasons for undertreatment, prescription drug costs remain one of the most important barriers.9,16 

	This study examined changes in prescription drug utilization behavior after implementation of a pharmacy benefit design that substantially lowered out-of-pocket costs for antidiabetic medications. In this study, converting from a traditional tiered formulary benefit to a VBID with substantially lower out-of-pocket costs for generic and brand medications resulted in higher levels of medication adherence (with MPRs increasing an average of 4.9%), an increase in treatment initiation rates, and a decrease in discontinuation rates. 

	The significantly higher treatment initiation rates in the VBID cohort, and the correspondingly low discontinuation rate in this population, suggest that some commercially insured individuals remain price sensitive and defer treatment because of cost. The reported level of therapy use was appropriate if the initiator group consisted only of newly diagnosed participants but fairly modest even in the VBID group if existing diabetes patients with unmet treatment needs were included. National statistics reported that among adults with diagnosed diabetes (type 1 or type 2), 14% take insulin only, 13% take both insulin and oral medication, 57% take oral medication only, and 16% do not take either insulin or oral medication.15 Although patients with less advanced disease may manage their diabetes through diet and exercise alone and delay initiating oral antidiabetic drug therapy until they need more intensive glycemic control, the observed lower rate of treatment initiation among those in need of insulin in the control group suggests that plans with high levels of cost sharing also will reduce the use of essential medications.

	Fewer plan participants ceased taking oral antidiabetic drugs than insulin, with no difference in discontinuation of oral agents between the VBID group and the controls (both initiators and continuers). One possible explanation for this similarity is that these plan participants, irrespective of plan design, already experienced a decrease in out-of-pocket spending due to the wide availability and use of generic oral antidiabetic drugs, so they would be less likely to discontinue their medications. Yet the VBID reduced the discontinuation rates among insulin users. We speculate that discontinuation of insulin was influenced directly by the cost share of medications that they would be or were taking—copayments for insulin decreased from $28 to $0 for VBID participants in the postchange period, whereas traditional participants faced a 22% increase to $28 in the same time period. The out-of-pocket costs for oral antidiabetic drugs for the VBID group decreased more than 50% but did not change for the control group, which may have stabilized utilization in the control group. These findings are consistent with those of 2 other recent studies of benefit design changes in commercially insured populations that showed similar associations with utilization for diabetes and other chronic diseases, although neither of these studies reported the same degree of therapy discontinuation that we observed.17,18 

	Although previous studies found that continuing users of chronic medications were less sensitive to price in responding to copayment increase,19,20 in our study the VBID dramatically improved adherence among continuers. It is possible that other plan benefit features (eg, use of generics) could influence adherence independently of copayment level. It also is noteworthy that adherence rates subsequent to initiation were similar for insulin and the oral antidiabetic classes, a result that runs contrary to previous studies in which adherence to insulin has been shown to be significantly lower than adherence to oral medications.21 This benefit change lowered costs for insulin therapy more than it did for oral therapy (out-ofpocket spending for insulin and oral antidiabetic agents decreased 79% and 52%, respectively, in the postchange period); therefore, difference in adherence rates between therapeutic classes might have been equalized. 

	Several limitations to our study should be noted. First, because administrative pharmacy claims do not include diagnosis, we could not identify the diagnosed cases that were untreated, nor differentiate between type 1 and type 2 diabetes patients. However, in adult populations, type 2 diabetes represents about 90% to 95% of all cases.13 Second, despite the matching criteria, baseline utilization patterns were quite different between the VBID and control groups, although participant demographics were homogeneous. The results of this study need to be interpreted in light of these group differences. Finally, our results are limited to the first year after a benefit design change. The effects of sustained benefit changes on overall medication use likely continue beyond the first year and need to be confirmed with follow-up studies. Future research should quantify the impact of adherence to therapy on improved health outcomes and determine the causal factors leading to higher initiation rates for the VBID group. Additionally, studies that include a wider range of therapeutic classes, including acute and chronic medications, are needed.

	Value-based insurance designs may help overcome some of the barriers to oral antidiabetic and insulin use. By reducing cost sharing and out-of-pocket costs, plan participants on diabetes medications receiving value-based plan coverage were more likely to initiate therapy, less likely to discontinue it, and had better medication adherence. These findings suggest that shifting plan participants from a traditional tiered plan design to a VBID is a viable alternative to existing health plan designs.

Among the commercially insured, pharmacy benefits define the availability, accessibility, and cost of prescription drugs.