Incentive Formularies: Spending and Utilization for 3 Common Drug Classes

Employers and other healthcare purchasers are increasingly using incentive formularies, which promote the use of generic or preferred brand name medications through the use of lower copayments. As of 2007, 75% of commercially insured individuals had prescription drug coverage that utilized some form of an incentive formulary with 3 or more tiers, whereas a decade ago such coverage was rare.¹ The level of copayments also increased between 2000 and 2007, for nonpreferred drugs by just under 50% and for preferred drugs by more than 65%.²

Some studies have investigated how incentive formularies change the behavior of patients who are using medications for the treatment of chronic medical conditions.^3-9 For instance, Huskamp et al found that among enrollees in 2 different employer groups who had been previously taking a nonpreferred statin drug to control high cholesterol and were switched to a 3-tier incentive formulary, more than one-fifth discontinued their medication in the ensuing 6 months.⁴ Similar discontinuation rates were demonstrated for angiotensin-converting enzyme (ACE) inhibitors and proton pump inhibitors (PPIs). Differences in discontinuation rates, however, were signifi cant for only 1 of the 2 employer groups studied. Goldman and colleagues found similarly high rates of medication nonadherence over a 1-year period for patients with chronic medical conditions after a doubling of copayments.⁹ However, both Huskamp et al and Goldman et al identifi ed medication users with chronic medical conditions on the basis of a single filled drug prescription. Although these studies suggest very high rates of discontinuation following increases in cost sharing, their results have not been replicated in a broader patient population or for persistent users.

We used data selected from more than 1 million continuously enrolled benefi ciaries from a single insurer that managed both medical and prescription drug benefi ts to examine the impact of changes in drug benefi t design on pharmaceutical spending and utilization in 3 specifi c drug classes: statins, ACE inhibitors, and PPIs. The first 2 classes are used to treat asymptomatic conditions that are risk factors for cardiovascular diseases, whereas the third is used to treat symptomatic conditions associated with acid overproduction in the stomach. Our study capitalizes on the diverse pharmacy benefits offered and the consistency of the medical coverage within the single health plan in 2 contiguous regions of the country. The large numbers of enrollees permitted identification of an adequate number of enrollees using prescription drugs in each class that changed to a new benefit design. It similarly permitted identification of a sufficient number of matched enrollees (controls) in the same geographic area that had the same initial benefit design but did not change.

We focused on 3 outcomes. First, we examined how changes in pharmaceutical benefits affected health plan and member prescription drug spending for each drug class in the year after the introduction of a new pharmacy benefit design. In so doing we incorporated the effect of rebates. Second, we examined how the benefit changes affected several utilization measures including discontinuation rates. Finally, we examined shifts in formulary adherence and the use of generic medications and mail order fulfillment.

METHODS

Overview

We assembled a data set of all pharmaceutical claims for the period January 1, 2000, through December 31, 2001, as well as demographic and benefit design information, for 1.25 million members from the Northeast and mid-Atlantic regions of a single health plan.¹⁰ We identified groups of enrollees that had the identical pharmacy benefit design in 2000. Within these groups, we identified cohorts with a pharmacy benefit design change on January 1, 2001. We then identified members who were taking a prescription drug in 1 of the 3 classes and who switched benefit designs, and matched them to control enrollees who maintained the identical benefit design throughout the entire study period. We compared pre- and post-changes using a “difference in-differences” approach.

Study Population

Eligible enrollees were between 0 and 65 years of age; were continuously enrolled in the health plan’s HMO or point-of-service (POS) plans between January 1, 2000, and December 31, 2001; and had prescription drug coverage provided by the health plan. At that time, the HMO/POS products had the largest share of health plan enrollment and were the only products that were routinely offered with a pharmacy benefit. Approximately 70% of HMO/POS enrollees had such coverage. Eligible enrollees were drawn from 11 states in the Mid-Atlantic and Northeast areas. We included only states with a minimum of 10,000 enrollees to ensure adequate numbers of enrollees in each state for comparisons. Because of concerns about selection bias in smaller firms, we eliminated the small number of enrollees associated with firms with fewer than 5 enrollees.

Drug Classes

We focused on 3 commonly used classes of medications that would have sufficient numbers of chronic users to draw reasonable inferences. We selected 2 drug classes used to treat chronic asymptomatic conditions (ACE inhibitors for hypertension and statins for high cholesterol) and 1 class (PPIs) used to treat disorders such as gastroesophageal reflux disease. Only ACE inhibitors had a generic option during this time period.

Pharmaceutical Benefits and Claims

The health plan provided pharmaceutical coverage to its enrollees through its affiliated pharmacy benefits manager. A single national formulary was used for all enrollees, but individual employers could select from a menu of incentive-based formularies ranging from a single-tier formulary to 3-tier formularies that required differential copayments for medications in higher tiers. Individuals within employment groups were not offered a choice of pharmaceutical benefit designs. Incentive-based formularies required enrollees to pay a higher copayment according to whether the drug was generic, formulary preferred, or formulary nonpreferred.

Each pharmaceutical claim included the name of the medication, dosage, days of supply, National Drug Classification code, place of purchase (retail vs mail order), and the amounts paid to the pharmacy by the health plan and the member through copayments or coinsurance. Through an employer file, these data were linked to information on the type and structure of the pharmacy benefit (eg, number of tiers, open vs closed formulary) and copayment amounts. We inferred the tier for each drug prescribed over the 2-year period as generic, formulary preferred, and formulary nonpreferred from the known benefit design by using observed copayments.

Identifying Prescription Drug Users in 3 Medication Classes

To examine persistent users of medications in each class at the time of the pharmacy benefit change, we examined pharmaceutical claims in the 4 months prior to the change. We defined persistence as having at least 2 filled prescriptions during the 4-month period with no break in medication supply of more than 30 days. The vast majority of these patients had filled prescriptions in both November and December. We purposefully excluded patients who did not have evidence of persistent drug use at the end of the year.

Creating Matched Cohorts

We identified groups of enrollees who had the same pharmaceutical benefit design in 2000, a portion of whom were switched to a different benefit design on January 1, 2001. That enabled us to examine a full 12 months of utilization before and after the benefit switch. We identified groups enrolled in 7 different benefit designs in 2000 and the 9 benefit designs that some of these enrollees switched to in 2001. Within each of the 7 groups, we identified cohorts of enrollees taking medications in one of the study classes according to the algorithm defined above.

For each class of medication use within each cohort, we estimated propensity score models predicting the probability that the enrollee belonged to a firm that changed benefit design. Predictors included state, employer size, and baseline demographic and enrollee clinical characteristics, as well as information on formulary adherence in the first time period. Clinical characteristics were summarized using Diagnostic Cost Group (DxCG) scores; these scores ranged from 0.30 to 2.15, with higher scores corresponding to higher predicted total medical costs.¹¹ The average score is set to 1.00; therefore, a score of 1.1 represents a predicted relative increase in cost of 10% compared with the benchmark sample. Each enrollee who was taking one of the medications and whose benefits changed was matched to an enrollee whose benefits did not change based on the estimated propensity score. Only “close” matches were retained—this was operationalized by requiring the estimated log odds of a benefit change between an enrollee who switched and an enrollee who did not switch to be within 0.60 standard deviation.¹² To minimize bias, we matched enrollees exactly on state and, in some models, age (in 5-year increments).

For each benefit change we studied, we created 2 cohorts of matched enrollees within each group. The first cohort included all persistent users identified in the last 4 months of the first year and was used to examine discontinuation rates in the second year. The second cohort required that users of a drug within one of the classes show evidence of use after the switch in their benefit.

Accounting for Rebates

Because of its confidential and proprietary nature, rebate information is generally not available for research studies. Without this information, however, findings with regard to health plan cost savings are biased downward. Moreover, rebates apply to preferred drugs and so differentially affect estimates of spending on preferred drugs relative to nonpreferred drugs. To address this problem, we obtained from the health plan a class-specific (but not drug-specific) estimate of the magnitude of the rebates in place during the time period of study and incorporated these into our estimates of health plan spending for preferred drugs.

Outcomes

Spending. Medication class—specific health plan and out-of-pocket pharmaceutical spending were determined for each member for each month by summing claims for each individual during the time period. For mail order prescriptions, we determined the number of days of supply (to the nearest month) and the amount of the copayment. Mail order fulfillment allows the enrollee to receive a 3-month supply of a drug for 2 copayments instead of 3 copayments. Mail order prescription costs were allocated to the year in which the prescription was filled for prescriptions that spanned both years of the study. The total spending and the copayment were then apportioned equally to each applicable month. Average monthly pharmaceutical spending per member was calculated for up to 12 months prior to the switch starting with the month that the first prescription was filled and for the 12 months after the switch in benefits.

Discontinuation, Medical Possession Ratio, and Persistence. We defined discontinuation as the absence of any new prescription after a 2-month period of no filled prescription. Persistence was defined as continuous use of the medication with a gap of no more than 30 days between the time a prescription would have been exhausted and the filling of the next prescription. Finally, the medical possession ratio, a measure of drug adherence, was defined as the total number of days of medication supplied divided by the total number of days from the first day of the first prescription filled in the year to the last day covered by the final prescription filled in the year. Because all identified patients were persistent users at the end of the first year, persistence was 100% at the end of the baseline year.

Formulary Adherence. We calculated, based on all filled prescriptions, the portion of filled prescriptions (standardized by days of supply) that were for generic (only applicable for ACE inhibitors), formulary preferred, and nonpreferred drugs, and the proportion of total prescription months that was fulfilled through mail order.

Analyses

We used a difference-in-differences analytical framework to examine the outcomes of interest. Prior to matching, differences in observed covariates between cohorts were tested using a 2-sample t test for continuous variables, a X² test for dichotomous variables, and the size of the differences examined using standardized mean differences. After matching, we again examined standardized differences between those that had a benefit change and those that did not to judge the balance of the cohorts. We considered acceptable standardized differences smaller than 10% in either direction for any covariate.

Once we selected matched pairs, estimated differences in outcomes in each group accounted for the matched pairs. Differences in spending and utilization outcomes were assessed using paired t tests for continuous variables and generalized estimating equations for grouped (paired) binomial data. With the exception of the medication possession ratio, for which we calculated an odds ratio, the estimates represent differences in rates calculated by subtracting the mean utilization rate in the cohort experiencing no change in copayments from the mean utilization in the cohort experiencing an increase in copayments. To determine the odds ratio for the medication possession ratio, we calculated the odds of an increase in days of supply for days of exposure for patients whose copayments increased versus the odds of an increase for patients whose copayments did not increase. For the 1 cohort where the copayments were decreased (cohort 6d), we reversed the sign for calculating the pooled estimates.

Finally, in order to provide an overall estimate of the impact of benefit change on utilization outcomes, we pooled the estimated differences across groups using Cochran’s semiweighted estimator.¹³ This estimate weights each group-specific estimate by its precision. Tests of statistical significance are uncorrected for multiple comparisons.

RESULTS

Population characteristics for all members, including those who were unmatched, are presented for each group in

Table 1

. Just more than half of the members were female, and the average age was generally in the low 30s. Across the groups the proportion of members from large employers with more than 3000 covered employees ranged from 34% to 100%. Average comorbidity scores were low. Characteristics of the matched cohorts of patients identified as taking a drug from 1 of the applicable classes are presented in

eAppendices A, B,

and

C

available at www.ajpblive.com. All standardized differences in the matched cohorts were less than 10%, indicating good balance with regard to measured confounders.

Changes in Spending

Prescription drug spending for each drug class for the matched cohorts is presented in

Table 2

. We first present changes in health plan, out-of-pocket, and total spending for statins for those in the cohort who switched to a different formulary compared with the matched cohort who retained the same benefit design. The subsequent columns present the same information for ACE inhibitors and PPIs. Each cohort that changed benefits is contrasted with its matched cohort of members that maintained the same benefit.

For statins, changing from a single-tier or 2-tier to a 3-tier incentive formulary was associated with a small decrease in total spending on the order of approximately 5% depending on the benefit design change. With 1 exception, the changes were all in the expected direction, although few were statistically significant. We then decomposed differences in total spending to examine differences in plan and out-of-pocket spending. Plan spending decreased on the order of 12% to 18%, whereas out-of-pocket spending that resulted from higher copayments for the second and third tier increased by between 28% and almost 300%. To illustrate, we describe findings for 2 of the groups. In 2000, members of group 2 had a single tier benefit design with a $10 copayment for all medications. Cohort 2b switched to a 3-tier incentive formulary with copayments of $10, $15, and $30 for the 3 tiers, whereas cohort 2a-1 continued with the single-tier $10 copayment design. When compared with cohort 2a-1, average per member per month spending on statins for cohort 2b fell by $4.00 (6%, P <.001): spending by the plan fell $11.20 (18%, P <.001) whereas average per member per month out-of-pocket spending increased $7.20 (112%, P <.001).

Group 6 began with a 3-tier incentive formulary. For cohort 6b, copayments increased for each tier, while for cohort 6c copayments decreased for each tier. Compared with the control cohort (group 6a-1), average per member per month spending fell by $3.90 (6%) for cohort 6b but increased by $2.40 (4%) for cohort 6c. Neither of these results was statistically significant. After decomposing spending into plan and out-of-pocket spending, however, spending by the health plan decreased $9.30 (17%, P <.01) for cohort 6b but increased $5.00 (11%, P <.05) for cohort 6c. Similarly, average per member per month out-of-pocket spending increased $5.40 (42%, P <.001) for cohort 6b and decreased $2.50 (16%, P <.001) for cohort 6c. Findings were similar for the other 2 drug classes.

Discontinuation Rates

Across the 3 drug classes, discontinuation rates ranged from an average of 17% for statins to 25% for PPIs (

Table 3

). There were no significant changes in discontinuation rates associated with individual changes in the formulary benefit design. For statins, however, the pooled discontinuation rate for those who switched benefit designs was 2% higher (P = .026) than for those who remained in the same benefit design for the entire study period.

Changes in Persistence, Medical Possession Ratio, Formulary Adherence, and Mail Order Fulfillment

Tables 4, 5,

and

6

show the changes for statins, ACE inhibitors, and PPIs, respectively. Across the 3 drug classes, there were few significant changes in persistence or the medical possession ratio. For instance, those taking statins who initially started with a single-tier $5 copayment and then switched to a 3-tier formulary (cohort 1b) were 66% persistent compared with 75% for the control cohort (cohort 1a, P <.05). For the other statin cohorts, however, there were no significant changes. A similar pattern was observed for the medical possession ratio, where differences between the matched cohorts ranged from 0 to 3 percentage points. As with discontinuation rates, however, for statins the pooled odds of an increase in days of supply after a switch were .81 relative to those with no switch (P <.001), suggesting that overall those taking statins were less adherent to their prescriptions after switching to a new pharmacy benefit.

Combining the estimates across the different benefit changes, there were statistically significant decreases in the use of nonpreferred drugs for both statins and ACE inhibitors (—3%, P <.001, and —4.6%, P <.001, respectively). There were statistically significant increases in mail order fulfillment for all 3 drug classes.

DISCUSSION

Our study has several notable findings. First, when cost sharing increased, overall drug spending by the plan sponsor within each of the classes fell compared with spending in a concurrent control cohort across a variety of benefit types and benefit changes. This fall was accompanied by a concomitant increase in out-of-pocket costs for those taking the medication. Consistent with the changes in relative prices, we also observed small decreases in brand nonformulary utilization and increases in mail order fulfillment.

Although most of the changes we observed represented increases in copayment, we observed 1 instance of a decrease (cohort 6c). Based on this 1 instance, there is no reason to reject the assumption of approximately symmetrical responses to increases and decreases in copayments.

Second, we found that increasing copayment levels and the use of multitier incentive formularies were associated with higher overall discontinuation rates for just 1 of the 3 drug classes examined (statins) and even in that class the magnitude of the overall effect was modest. In addition, for those who continued on the medication, measures of adherence, including both persistence and the medical possession ratio, were largely unchanged, although we did observe an overall decrease in the medication possession ratio for statin users. Thus, we found some support for the notion that drug copayments affect the behavior of patients on chronic drug regimens, but the effect is not consistent across different classes of drugs, different measures of behavior, and perhaps most importantly, different populations. These findings add to the small number of studies that have shown either no effect on discontinuation rates or a small effect.^5,6,14-17

The increasing prevalence of incentive formularies makes it important to understand how they affect the use of drugs and, in particular, the use of drugs used to treat chronic medical conditions such as hypertension and hypercholesterolemia. If formulary policies lead to decreased adherence, this may lead to higher medical costs in the future as well as suboptimal clinical outcomes.^18,19 Our study is notable for examining a diverse mix of pharmacy benefit designs, including 1 that lowered copayments, and for using carefully matched concurrent comparison cohorts selected from a population of more than 1.2 million enrollees. In addition, all members were drawn from 2 contiguous regions of the country with identical medical coverage from a single health plan over the entire period of the study. Finally, to our knowledge, this is the first study of specific drug classes that accounts for rebates in the analyses.

Our findings of modestly increased discontinuation rates for those who were switched to incentive formularies contrast with at least 2 other previous studies in the literature.^4,9 Our study, however, differed in important ways from these studies. First, we required evidence of persistent drug use at the end of the first year for a patient to be included in the analysis. Second, we created matched cohorts of users of each of the drugs using a propensity score approach. Studies that required only a single prescription or 2 prescriptions over the prior 6 months might have included enrollees who were taken off of the medication or had displayed evidence of nonadherence prior to switching benefits. Our method identified the subset of patients who might have been the most committed to taking their medication. Even within this subset, however, we observed high discontinuation rates among both those who switched their pharmacy benefit and those who did not, as well as a suggestion of slightly increased discontinuation rates overall for those who switched to a benefit design with higher copayments. Some have argued that rather than raise copayments for life-saving medications such as statins and ACE inhibitors, these copayments should be lowered to increase adherence.²⁰

Our findings related to costs demonstrate that cost savings by the plan sponsor are largely accounted for by increased out-of-pocket spending by enrollees. The magnitude of the changes in out-of-pocket spending can be large in percentage terms, with out-of-pocket payments often doubling or tripling. Competition among plans may mean plan savings that in turn translate into lower premiums, although any such savings may be distributed differently across persons and households than the increased out-of-pocket payments.

Our study is subject to several limitations. First, we studied commercially insured enrollees of a single large health plan in 2 contiguous regions of the country. Our results, therefore, may not generalize to the elderly, to the poor, or to other regions of the country. Second, because we lacked data on enrollee characteristics such as socioeconomic status, we could not discern whether incentive formularies had differential effects on certain types of enrollees, such as the poor or underserved minority populations. Third, we studied multiple benefit changes for 3 classes of drugs, and we did not formally adjust for multiple comparisons. We also were not able to ascertain whether there was a threshold amount above which copayment increases became increasingly noticeable to enrollees. Our findings should be interpreted with this in mind. Finally, our study was limited to a single year of follow-up after the introduction of the new pharmacy benefit, and the effects we observed might change over time. Future studies should examine the extent to which these observed differences persist with time.

In conclusion, we found that switches to incentive formulary arrangements with higher levels of copayments for 3 specific classes of drugs generally led to overall lower drug costs for health plans, but that these health plan savings were accompanied by substantial increased out-of-pocket spending for health plan members such that overall spending was only modestly impacted. The inclusion of rebates allowed us to present a more accurate picture of the magnitude of these changes from the health plan perspective. Our study did not, however, fully replicate previous findings of increased rates of discontinuation for those on chronic medications.