The integrated, staff-model HMO offered a single MAPD plan for individual subscribers, with a coverage gap between U.S.$2,250 in total drug costs and U.S.$3,600 in out-of-pocket expenditures in 2006. Individual subscribers had no deductible and U.S.$10 generic and U.S.$40 brand copayments for up to a 30-day supply before the gap. After U.S.$3,600 in cumulative annual out-of-pocket expenditures, beneficiaries had U.S.$3 generic and U.S.$10 brand copayments. In 2005, individual subscribers had generic-only benefits with U.S.$10 generic copayments (for up to a 100-day supply). In 2006, other MAPD beneficiaries with employer-supplemented insurance had no coverage gap and lower copayments than individual subscribers, that is, U.S.$5–30 generic and U.S.$10–75 brand copayments for up to a 100-day supply; their benefits were similar in 2005 as in 2006. The Integrated MAPD plans were available in California.
The network-model HMO offered two MAPD plans in 2006: one with a coverage gap as described above, and one with generic-only coverage during the gap. Neither plan had a deductible and both had a four-tier copayment before the gap: U.S.$8.50 for generics, U.S.$26– U.S.$27 for preferred brands, 50 percent coinsurance for nonpreferred brands, and 33 percent coinsurance for specialty drugs. During the gap, beneficiaries in the generic-only plan had U.S.$8.50 generic copayments, but no brand coverage. During catastrophic coverage, beneficiaries in both plans paid the greater of 5 percent coinsurance or U.S.$2 generic and U.S.$5 brand copayments. Beneficiaries with the standard coverage gap in 2006 were most commonly enrolled in MA plans with no or limited (e.g., capped) brand-name drug coverage in 2005; the majority of beneficiaries with generic-only gap coverage in 2006 had unrestricted generic and brand coverage in 2006. The gap plans were available in multiple states; the generic-only plan was only available in select counties within California, where beneficiaries had a choice of the two plans.
Study Design and Population
Our focus was to compare the standard Part D coverage gap with partial and complete gap supplementation; thus, we conducted comparisons holding other system variables constant. The two MAPD plan sponsors differ in their levels of integration, care management practices, plan offerings, geographic locations, and drug formularies, which could affect drug costs and adherence; therefore, we conducted separate, parallel analyses of beneficiaries with and without supplemental gap coverage within the same Medicare Advantage system. Specifically we compared the following:
- Beneficiaries with a coverage gap versus beneficiaries in employer-supplemented plans without gaps within an Integrated MAPD; and
- Beneficiaries with a gap versus beneficiaries with generic-only gap coverage within a Network MAPD.
The study included beneficiaries continuously enrolled from January 1, 2005 through December 31, 2006, 65+ years old, with ≥1 oral diabetes prescription dispensed in 2005. We selected the cohort based on prior year drug use to focus our analyses on the effects of Part D-related cost-sharing, and specifically the coverage gap, on costs and adherence for diabetes patients receiving ongoing drug therapy. We excluded dual-eligible beneficiaries (Medicaid-Medicare) and those receiving Medicare's low-income subsidy because they had substantially different cost-sharing levels. In analyses examining hypertension and lipid (cholesterol) drugs, we included the subset of beneficiaries who had ≥1 drug in the respective class dispensed in 2005.
Potential differential selection of Medicare drug plans is a concern for any nonrandomized study. By conducting separate within-system analyses, we mitigated concerns related to differential selection of plans offered by different plan sponsors. Within the Integrated system, there was only a single plan available for individual Part D subscribers, that is, the coverage gap plan. The benefits for beneficiaries with employer-supplemented insurance were determined at the employer, not the individual, level, which reduces selection concerns within the Integrated system. Within the Network system, however, over half of the study population lived in areas where they could choose between the basic coverage gap plans and the enhanced generic-only coverage plans, with a higher premium for the latter. To reduce potential selection bias that could result from healthier patients with lower levels of drug need choosing the less generous coverage gap plan, we excluded subjects in these plans that lived in areas where they also had a choice of the generic-only plan. Identifying beneficiaries with the same chronic condition, diabetes, further increases comparability by focusing on beneficiaries with more clinically homogeneous needs. In addition, we used propensity scores to reduce bias due to imbalances in measured characteristics of the comparison groups, for example, beneficiaries with generic-only gap coverage versus a standard coverage gap (with no choice of the generic-only plan).
We examined total drug costs and out-of-pocket expenditures for Part D drugs using health plan pharmacy data. We also examined costs for three diabetes-related classes: oral diabetes, hypertension, and lipid drugs. Total costs are the amount that beneficiaries would have paid if the drug was not covered by their plan (e.g., during the gap), and the amount includes the acquisition cost and dispensing fee. Out-of-pocket costs were calculated as patient costs and included copayments/coinsurance, or full price, during uncovered periods.
To measure annual and monthly adherence to oral diabetes, hypertension, and lipid drugs, we calculated the proportion of days covered (PDC) using dispensing data. Adherence was defined as having PDC≥80 percent in the year or month for the entire regimen (Fung et al. 2007
) and allowed drug supply to carry over from month to month. When examining adherence to oral diabetes drugs, we censored subjects if they were dispensed insulin to isolate changes in adherence from prescribed changes in the diabetes drug regimen (e.g., potential replacement of oral agent with insulin).
To examine differences in annual 2006 total drug costs and out-of-pocket expenditures between beneficiaries with a coverage gap and those with either no gap or generic coverage, we used one-part general linear models and log transformed costs (Buntin and Zaslavsky 2004
). To examine differences in adherence (PDC≥0.80) to each of the three drug classes, we used logistic regression.
To examine changes in drug costs and adherence before and after beneficiaries exceed the coverage gap threshold, we plotted monthly differences among those who reached U.S.$2,250 in total drug costs aligned by the month in which they exceeded this threshold. Analyses were limited to the 6 months before and 3 months after subjects reached the gap threshold in 2006 because the majority exceeded the threshold in later months during the year. We estimated mean monthly costs and adherence levels for the gap versus supplemented gap groups using a generalized estimating equations approach and treating all subjects as if they were in each group; standard errors were estimated using the delta method (Oehlert 1992
). Models included monthly indicators and interactions between month and coverage gap group to examine differences in each month before and after reaching the gap threshold. In sensitivity analyses, we conducted within-person fixed effects analyses, which are robust to time-stable unmeasured differences between groups; these analyses yielded consistent findings.
To calculate propensity scores, we estimated the probability of having a standard coverage gap (versus no gap or generic-only gap coverage) using logistic regression models and included the propensity score as a continuous variable in all analyses. The logistic models adjusted for age (65–74, 75–84, 85+), gender, plan membership tenure, and neighborhood socioeconomic status based on the 2000 U.S. Census and median household income at the block group level. We also adjusted for comorbidities (hypertension, hyperlipidemia, coronary artery disease, depression, osteoarthritis, and chronic kidney disease) based on diagnoses in 2005. As a proxy for diabetes severity, we included an indicator for use of oral diabetes medications alone or with insulin in 2005. When assessing costs or adherence for specific drug classes, we adjusted for the number of drugs in beneficiaries' 2005 regimens, and the mix of generic and brand drug use in 2005. In cost analyses, we controlled for prior year total drug spending for all Part D drugs or the respective drug class, depending on the outcome. Because beneficiary-level race/ethnicity was not available in both health systems, we did not include race/ethnicity in the propensity score. In sensitivity tests, we included census-based race/ethnicity (Network MAPDs) and individual-level race/ethnicity (Integrated MAPDs); point estimates and statistical inference for the main predictor (amount of gap coverage) were similar.