How Effective Is the Medicare Part D Drug Plan?

The Medicare Part D drug plan, introduced in 2006, fundamentally changed the drug coverage landscape for U.S. seniors. Although Part D does not cover all drug costs, it nonetheless decreased financial uncertainty, improved access to pharmacotherapy, and thus presumably improved the health of seniors who previously had no or limited drug coverage. In this issue, Briesacher and colleagues (1) report a study that attempts to estimate the effect of Part D on the health of the entire community-dwelling Medicare population 5 years after plan implementation. Earlier studies found that Part D reduced the nondrug health care spending of persons with limited drug coverage before 2006 (2, 3). But Briesacher and colleagues note that these studies focused on the effects only 1 to 2 years after program implementation. To conduct their study, Briesacher and colleagues analyzed 11 consecutive years (6 before Part D and 5 after) of data from the Medicare Current Beneficiary Survey (MCBS). The MCBS is intended to be representative of the entire Medicare population. The authors focused on various health outcomes of community-dwelling MCBS respondents: fair or poor self-assessed overall health, limitations to activities of daily living, frequency of hospitalization or emergency department visits, and mortality. To identify the effect of the Part D policy, the authors used the interrupted time-series design; they extrapolated the prepolicy trend in mean outcomes into the postpolicy period (an estimate of counterfactual outcomes) and compared these with actual postpolicy outcomes. Estimates were obtained by using a statistical model that controlled for age, sex, and race of the beneficiaries. Somewhat surprisingly, the program had little apparent effect on mean health outcomes. These results were robust to both the statistical model used and the analyses that focused on beneficiaries who reported having been diagnosed with cardiovascular disease. What are we to make of these results? Is the annual public expenditure of $70 billion on the program all for naught? It may be premature to dismiss the program as ineffective for 3 reasons. First, the interrupted time-series design requires that extrapolation of the prepolicy trend in outcomes to the postpolicy period represents what the outcomes would have been in the absence of the policy. In Table 1 from Briesacher and colleagues' article, there is evidence that the composition and number of comorbid conditions of community-dwelling MCBS respondents changed over the years. In particular, the fraction of enrollees with 3 or more comorbid conditions was higher in 2010 (62.0%) than 2000 (55.6%). The fraction of enrollees who also received Medicaid, and who were presumably less affluent and possibly less healthy, was higher in 2010 (14.7%) than 2000 (13.1%). If these changes occurred smoothly over time, then the study design is valid. But the policy effect is underestimated if these changes occurred mainly after 2005. This may have been the case; community-dwelling MCBS respondents who reported relatively poor health after 2005 might have been in even poorer health without Part D coverage. Some of these respondents would have been institutionalized and thus excluded from the study sample. Therefore, the focus on community-dwelling beneficiaries induced a form of selection bias. Whether this bias materially affected the estimated policy effect depends on the fraction of Medicare enrollees whose institutionalization status was affected by Part D. The fraction is probably small, but it would be good to check before rendering a verdict on Part D's effect on health. The second reason to exercise caution is that some of the outcome measures that Briesacher and colleagues analyzed may be insensitive to the health domains that would be affected by improved pharmacotherapy. Seniors who are limited in carrying out their activities of daily living because of impairments in ambulation, manual dexterity, vision, hearing, speech, or cognition may still face these impairments even if they receive appropriate pharmacotherapy. Other study outcomes, such as fair or poor self-assessed overall health or mortality, should be sensitive to the receipt of pharmacotherapy. But for many of the chronic health problems in seniors, such as hypertension, hyperlipidemia, and diabetes, there is a considerable lag between the receipt of pharmacotherapy and a change in these outcomes. Antihypertensive drugs affect the risk for heart attack in future years but otherwise would not result in physiologic changes that would be noticeable to most patients. As Briesacher and colleagues acknowledge, the 5-year follow-up might still be too short to see an effect. The final reason that caution is warranted is that Part D will mainly improve the health of the subset of Medicare enrollees for whom Part D actually improved drug coverage. It should have little effect on persons who already had similar or better coverage. But Lichtenberg and Sun (4) report that most (72%) of the prescriptions paid for by the program would have already been covered by another sourceonly 28% of the prescriptions covered by Part D in 2006 were new. Although Briesacher and colleagues acknowledge this issue, they also had concerns with the 2 studies that focused on persons with poor coverage before 2006 (2, 3). They suggest that these analyses were probably biased "because of associations among coverage choices and other characteristics and behaviors" (1). It is unclear how such bias would manifest. Certainly, cross-sectional comparisons would be problematic but these studies were longitudinal and tracked the prepolicy and postpolicy outcomes of groups of Medicare enrollees whose coverage differed before 2006. The pre- to postpolicy change in outcomes of persons with good coverage throughout was used to control for the latent secular trends that might affect the change in outcomes of persons for whom Part D improved drug coverage. It is not obvious why the former group would be a poor comparator with the latter. Therefore, I echo Briesacher and colleagues' call for further long-term longitudinal research on this issue. One possibly fruitful avenue is to retain their approach, which is fundamentally sound, but include all Medicare enrollees (including the institutionalized) and then compare the health outcomes of groups who differ in the extent to which Part D improved coverage. The Congressional Budget Office (5) notes that the "near poor"those with incomes ranging from 100% to 200% of the federal poverty levelwere disproportionately uninsured before 2006. Finally, prescription drug use, similar to other types of health care use, is right-skewed in that 20% of persons use 80% of the care, so it may be better to focus not on the mean of the distribution of health care use but on the right tail of this distribution. This can be accomplished using quantile regression and related techniques.