Our analysis of the effects of Medicare Part D, a large-scale natural experiment among uninsured seniors, showed that the implementation of the benefit was associated with a sizable initial reduction in out-of-pocket drug spending and a meaningful increase in the use of selected essential medications.
The extent to which Medicare Part D stimulated increased dispensing of overused versus underused medications requires further study. Warfarin use was the least affected by the introduction of Medicare Part D, likely because of its relatively low cost and narrow therapeutic range requiring closer physician monitoring. In contrast, PPIs, a medication class that is frequently overused, experienced the steepest increase in use.
The benefit of Medicare Part D was not evenly distributed throughout the year. A sizable proportion of patients reached the coverage gap before the end of the year, which resulted in meaningfully reduced dispensing of previously used essential medications. Consequently, overall use of statins and clopidogrel, drugs with proven effectiveness, reached a plateau and started to decline in late 2006. Although the coverage gap may have been a necessary component of the Part D legislation to allow for passage of the bill, our results suggest that it mitigated gains in access to prescription drugs in our study setting.
Consistency with previous studies
Our results are consistent with those of two previous studies, which estimated 5.9 percent and 12.8 percent increases in prescription drug use and 13.1 percent and 15.6 percent decreases in out-of-pocket spending, respectively, among all seniors, regardless of pre- and post-Part D insurance status.20
We found that among previously uninsured seniors, drug use increased 3–37 percent and out-of-pocket spending decreased 37–58 percent, with the results varying by drug class. Assuming that the effects of Medicare Part D were limited largely to the one-third of seniors who were previously uninsured, our estimates should be approximately three times those previously reported. The demand elasticities implied by our results are 0.35 for warfarin, 0.44 for statins and clopidogrel, and 0.76 for PPIs. Previous estimates used by Mark Pauly in forecasting the effect of Medicare Part D (0.4) and calculated by Frank Lichtenberg and Sean Sun from an analysis of Walgreens data (0.72) are within the range observed in our study.21
One advantage to our class-based analysis is that it allowed for a detailed assessment of the effect of the policy change on medication use for patients with specific diseases; Part D evaluations that lump all drug classes together risk generalizing findings that are in fact quite heterogeneous.
Our findings were not entirely unexpected. A preponderance of evidence indicates that increased patient cost sharing has a strong effect on overall medication use and adherence to specific therapies.22
Reduced drug usage as a consequence of reaching a coverage gap has been associated with worse health outcomes.23
Our study found that increasing and limiting coverage similarly affected the use of under-used and overused medications, consistent with results from the RAND Health Insurance Experiment, which showed patients’ difficulty discriminating between more and less essential treatments under increased cost sharing.24
Impact of Part D on generic drug use
There was a rapid uptake of newly marketed generic statins and esomeprazole under Medicare Part D. Advocates of the use of private drug plans to deliver prescription drugs to seniors have argued that such plans could reduce costs for prescribed medications by steering patients toward more cost-effective medications such as generic versions of their current medications.25
We did not directly test the effect of Part D on generic medication use, but our findings are consistent with this contention.
Strengths of our analysis
Implementation of Medicare Part D on 1 January 2006 constituted a large-scale natural experiment. Interrupted time-trend analyses of drug coverage changes are considered among the most valid study designs, short of randomization.26
By establishing baseline time trends of drug usage in 2005 that were then followed into 2006, we can be confident that a comparison of post– to pre–Part D utilization trends provides a valid estimate of the Part D effect, assuming no other major interventions in early 2006 that would have affected drug use. Because the patients we studied were continuous users of the same pharmacy chain before and after the policy change, we believe that changes in the composition of our population over time do not account for the substantial utilization changes that occurred immediately after the policy change in this aggregate-level analysis.27
Although individual subjects’ characteristics could have changed over time as a result of aging, and we did not have diagnostic information to fully characterize patients’ health states, only an implausibly massive and sudden change in health or a substantial, simultaneous health system intervention could explain our findings.28
Our study is based on drug-use data that include patients with and without drug benefits, since they were collected independent of coverage status.
Limitations of our analysis
Despite these strengths, several limitations must be considered. Generalizability to all Part D–eligible patients without prior drug coverage may be limited because the patients in our study were continuous users of the same pharmacy chain before and after the study period. Although this does not rule out that patients received drugs in other pharmacies concurrently, our findings are generalizable to patients who tend to prefer a particular pharmacy or live near only one pharmacy. The U.S. Northeast was underrepresented in our study, but it is unlikely that the uninsured in the Northeast would respond differently to Part D than uninsured patients elsewhere in the United States. We could not account for over-the-counter (OTC) medication use, which may have complicated the PPI drug class analysis, because enrollment in Part D may have led some patients to switch from OTC to prescription medications.
We also did not have information about enrollment in specific prescription drug plans, which are likely to vary greatly in their particular formularies and cost-sharing requirements. We generated specific but not necessarily sensitive algorithms to assign insurance status, including the coverage gap. Testing variations in the assignment algorithm of insurance status in 2006 confirmed the robustness of our definitions. The high specificity ensured high internal validity of the policy-effect estimates because we avoided contamination of our study populations. However, it resulted in an underestimation of the proportion of seniors without drug coverage and possibly underestimated the proportion of patients who registered for drug benefits in 2006 or those reaching the coverage gap.29
Only detailed insurance data would allow a precise assessment of drug benefits, but these data would not include drug usage data for previously uninsured patients, rendering the data useless for our study question. Additional studies linking pharmacy-level data to health plan data would allow researchers to better assess the population of patients who use multiple pharmacies and to better identify the presence or absence of drug coverage.
We also did not include a concurrent control group in this analysis to more completely account for any underlying temporal trends in medication use. However, an analysis of patients dually eligible for Medicare and Medicaid, who had coverage through Medicaid during 2005 and were automatically enrolled in Medicare Part D in 2006, was conducted with this data set and indicated no significant change in the use of medication classes in our study after the implementation of Part D.30
This finding diminishes but does not completely eliminate concerns regarding the absence of controls. Also, we did not consider the cost of Medicare Part D to the government.
As seniors participated in Part D plans, it is possible that some received their medication via mail, which would not be recorded in our data. This would lead to an underestimation of the increased use of medications under Part D and could possibly overstate the reduced use during the coverage gap. Although regression to the mean of the coverage gap effect cannot be fully ruled out in an uncontrolled time-series analysis, a cohort study approach showed similar results.31
We also cannot determine precisely which patients entered the coverage gap. Only 16 percent of plans provided coverage in the coverage gap in 2006, but numerous cost-sharing strategies were used.32
Our estimates of total drug spending per patient were based on cost estimates using 80 percent of the average wholesale price, leading to imprecision of cost estimates for individual drugs.33
Additionally, we may have randomly misclassified the exact date on which patients entered the coverage gap, which would lead to an underestimation of the observed reduction in usage. Despite the measurement error, there does seem to be a strong relationship between patients’ out-of-pocket spending and medication use that closely corresponds to the time that we would expect patients to be entering the coverage gap.
The introduction of Medicare Part D was a mixed blessing for seniors without prior drug benefits. To the credit of the benefit, patients who enrolled were more likely to use essential medications, including clopidogrel, statins, and to a lesser extent warfarin, that are likely to result in better health outcomes. However, a sizable proportion of sicker patients reached the coverage gap in the first year and experienced a drop in the use of drugs they had used before reaching the coverage gap, which may result in worse health outcomes.
Additionally, while private drug plans may promote the use of generic drugs, there is also evidence that coverage within these plans encourages greater use of drugs that tend to be overused as well as those that are underused, and it may not adequately distinguish between the two. Increased implementation of benefit designs that require evidence of clinical appropriateness before authorizing medication use, or implementation of value-based benefit designs that reduce cost-sharing requirements for the most effective medications, may help encourage more appropriate and cost-effective medication use in Part D. Moreover, efforts to provide additional coverage in the coverage gap for certain essential medications may assist in optimizing coverage and the health of seniors.