The American Journal of Managed Care November 2008 - Special Issue
Medicare Part D's Effects on Elderly Patients' Drug Costs and Utilization
Objectives: To analyze Medicare Part D’s net effect on elderly patients’ use of and out-of-pocket (OOP) costs for prescription drugs and to compare these with standardized results from prior studies.
Study Design: Our dataset contains 1.4 billion prescriptions from Wolters Kluwer Health from December 2004 through December 2007 for patients whose age as of 2007 was more than 57 years.
Methods: Days’ supply per capita, OOP cost per day’s supply, and number of individuals filling prescriptions were compared before and after January 2006 for those over age 66 years versus those age 58-64 years. Adjustment was made for under-reporting of pure cash prescriptions in the data.
Results: Elderly patients’ utilization in the first year of Part D increased compared with that of near-elderly patients by 8.1% for days’ supply and 4.8% for the number of individuals filling prescriptions, and their OOP costs declined by 17.2%. Although elderly patients’ OOP costs in the second year were reduced an additional 5.8%, days’ supply increased by only an additional 1.0%. Correcting for the under-reporting of pure cash prescriptions yielded effects of 8.1% and -3.5% for days’ supply and -15.6% and -7.2% for OOP costs in 2006 and 2007, respectively. A standardized comparison with previous estimates from Walgreens data showed that our utilization estimates were 2.6 times larger.
Conclusion: Part D lowered elderly patients’ OOP costs and increased utilization, primarily during the first year of the program. Magnitudes vary substantially across studies because of differences in data and methods.
(Am J Manag Care. 2008;14(11 Spec No.):SP14-SP22)
Using near-elderly patients as a control group, we concluded that Medicare Part D:
- Decreased elderly patients’ relative out-of-pocket costs for prescription drugs substantially in both 2006 and 2007.
- Increased their relative utilization of prescription drugs substantially from 2005 to 2006 but not from 2006 to 2007.
- These 2006 utilization estimates are 2.6 times larger than comparable results from previous studies using different data.
Part D can affect elderly patients’ net prescription drug utilization in several ways. First, Part D enrolled some elderly persons who previously lacked drug coverage; if demand is not perfectly inelastic, new coverage would increase the quantity of drugs consumed through lower OOP costs. To examine this extensive margin, we analyzed changes in the number of elderly filling any prescription before and after January 2006. Utilization also could have increased at the intensive margin for those previously covered if their coverage under Part D was more generous than their previous coverage (eg, because of lower patient OOP costs or fewer supply-side restrictions such as prior authorization requirements.) Conversely, net utilization could have decreased if prior coverage was more generous than Part D coverage or if employers responded to Part D’s introduction by reducing the generosity of retiree coverage.
Two existing peer-reviewed studies also examined this aggregate effect. Both prior studies used data from only Walgreens pharmacies. Yin et al used data covering September 2004-April 2007 for those age 66-79 years, with the experience of those age 60-63 years as a control group.3 They found that Part D increased use of medications by 1.1% and OOP costs fell by 8.8% during the initial open enrollment period (January-May 2006). During the following 11 months (June 2006-April 2007), they found that utilization increased 5.9% and OO P costs decreased by 13.1%. Lichtenberg and Sun also used Walgreens data to compare drug use among those age >65 years with those age <65 years from September 2004 to December 2006; they found an increase of 12.8% in quantity and an 18.4% reduction in OOP costs.4 In this article we reconcile the seemingly disparate findings from these 2 studies and compare them with our own results, paying attention to the various strengths and weaknesses of each data source and method.
Studies on the effect of Part D on access to medications also have been conducted with survey data. Using the Health and Retirement Study, Levy and Weir found that although there was substantial take-up of Part D by the uninsured, prescription consumption did not appear to increase.5 Neuman et al conducted a survey of Medicare beneficiaries in fall 2006 and reported that, compared with employer coverage, OOP costs were higher under Part D.6 Before implementation of Part D, Pauly forecast that Part D would reduce elderly patients’ OOP costs by 14% overall, by halving costs for the 25% who were uninsured.7 Using a 0.4 price elasticity of demand measure, he estimated a 6% increase in utilization as a result of Part D.
We analyzed a large dataset of prescriptions covering the period from December 2004 through December 2007 from Wolters Kluwer Health’s Source Lx database.8 Our dataset comprises 1,417,366,769 prescriptions filled by 34,198,008 unique patients, whose age as of 2007 was >58 years. McKesson Corporation (formerly Per-Se Technologies, and NDCHealth before that) collects these data primarily from their switches that transmit electronic claims between pharmacies and third parties (eg, insurance companies and pharmacy benefit managers) that help pay for prescriptions. Source Lx covers all 50 states and includes approximately 31% of the nation’s prescriptions, filled at approximately 32,000 pharmacies (not restricted to any particular pharmacy chains), including retail, mail-order, and hospital pharmacies, as well as other institutions that fill prescriptions such as long-term care facilities.8 The dataset includes prescription-fill level information on patient age, OOP costs versus third-party payment amounts, number of days’ supply, the pharmacy’s ZIP code, pharmacy type, and an encrypted patient identifier that tracks usage over time. Descriptive statistics are provided in the eAppendix (available at www.ajmc.com).
We compared 2 consistent cohorts: a group always age eligible for Medicare since January 2006 (age >66 years as of 2007) versus a group always age ineligible for Medicare (age 58-64 years as of 2007). (Our data included the patients’ birth year but not month, so we excluded those who turned 65 during either 2006 or 2007. Because a small fraction of elderly persons are not eligible for Medicare and some persons under age 65 years are eligible, our treatment-control group definitions may cause small underestimates of the effects of Part D. See Lichtenberg and Sun.4) Using these groups, we made nationwide difference-in-difference comparisons of the aggregate change from January 2005 to December 2007. Use of a control group was important to separate Part D’s effects from those of other factors that changed over time. For example, several widely used products such as Zoloft lost patent protection in 2006. To the extent that near-elderly patients also experienced the effects of these other changes, our method identified Part D’s effects better than a simple comparison of 2005 data with 2006 data for elderly patients.
In addition to showing the aggregate effects of Part D for the first and second years of the program, we also report monthly estimates to consider how Medicare’s effects varied during the initial open enrollment and transition period in early 2006, early in the calendar year before patients had met their deductibles, and later in the year when some entered the donut hole. To justify use of our control group, we used data prior to the implementation of Part D to validate the underlying assumption that the pre–part D time trend did not significantly differ between the 2 age groups.
We present results for 3 main outcomes: days’ supply, number of patients filling prescriptions, and patient OOP costs per day’s supply. We also analyzed total prescriptions filled, but we did not report these results separately because they were virtually identical to those for days’ supply. Because our 2 utilization outcomes (days’ supply and number of patients with prescriptions filled) were compared in terms of totals for the 2 cohorts over time, we had to adjust for different rates of mortality for the 2 cohorts, or we could have found relative decreases in total use occurring for the older cohort simply because of more deaths. To do this adjustment, we normalized our totals by the size of the national population estimated or projected by the US Census Bureau for each cohort at each point in time and report per capita changes.9,10 We report the percent changes over time in 3 outcomes to calculate the difference-in-difference impact of Medicare Part D.
Correction for Biases
Certain features of our dataset pose hurdles to estimating the impact of Part D on use of prescription drugs by the elderly and need to be explained up front. First, our data largely but not exclusively tracked the same pharmacies over time. This unbalanced panel could bias our results if the pharmacies that entered or left the data catered to a single age cohort disproportionately. To address this concern, we tested the robustness of our results on a panel of retail pharmacies that consistently reported data during our study period. These included 77.4% of the total days’ supply in the full dataset.
Three sources of drug procurement are under-reported or absent in these data (“pure cash,” mail order, and free samples). These may have systematically changed as a result of Part D, causing us to overestimate or underestimate the true impact of Part D on the actual quantity of medications used by seniors.
Our data under-report pure cash transactions (where there is no third party to receive a claim) because they represent only 3.7% of our prescriptions but are thought to comprise approximately 12% of nationwide prescriptions for our age groups. Cash transactions for purchasers with drug discount cards, claims filled before deductibles were met, and rejected claims by insured people for uninsured (off-formulary) drugs are not subject to this under-reporting issue. This under-representation of pure cash transactions could have caused us to overestimate the effects of Part D to the extent that Part D disproportionately shifted elderly patients’ 2005 cash purchases not observed in our data to insured purchases observed in our data in 2006 and 2007. Second, mail-order pharmacies are under-represented in the Wolters Kluwer Health data; they represent 2.1% of scripts in our data, whereas they are 13.8% for all ages nationwide in 2007.11 This under-representation would have caused us to underestimate Part D’s utilization effects if Part D disproportionately increased elderly patients’ use of mail-order pharmacies. Third, prescription sales data omit free samples by definition. That would have caused an overestimate of Part D’s effects on actual drug utilization if Part D caused use of free samples to decrease.
To correct for the bias due to missing pure cash prescriptions, we used age- and year-specific estimates of the fraction of prescriptions filled that were pure cash when prescriptions with all payment sources were fully represented, as reported by Wolters Kluwer Health from other proprietary datasets. (These other data are from pharmacies that report all transactions. However, these pharmacies likely have higher proportions of pure cash payments than other pharmacies, so these corrected results rely on an upper bound estimate of the extent of pure cash, while our uncorrected results rely on a lower bound estimate.) Comparing the 2 numbers told us by how much we needed to multiply the volume of pure cash observed in our data to make up for the fraction that we were missing. We performed this correction for quantity of days’ supply for each age cohort and year. Likewise, we adjusted our estimated effects on overall OOP cost by adding in the missing pure cash observations. In so doing, we assumed that the average prices of missing pure cash transactions were similar to the average prices of the observed pure cash transactions (ie, the missing transactions were missing at random). Unfortunately, age-specific data on mail-order use were not available to implement a similar correction for their under-representation. In the Discussion section we present some evidence that this limitation might not bias our results because the available data suggest that mail order did not change systematically over time.