Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Med Care. Author manuscript; available in PMC 2012 September 1.
Published in final edited form as:
PMCID: PMC3151472

Medicare Part D and Changes in Prescription Drug Use and Cost Burden: National Estimates for the Medicare Population, 2000–2007

Becky A. Briesacher, PhD,(2) Yanfang Zhao, MA,(4) Jeanne M. Madden, PhD,(1) Fang Zhang, PhD,(1) Alyce S. Adams, PhD,(3) Jennifer Tjia, MD, MSCE,(2) Dennis Ross-Degnan, ScD,(1) Jerry H. Gurwitz, MD,(2) and Stephen B. Soumerai, ScD(1)



The full effect of Medicare Part D, after the initial policy transition period and across the U.S. Medicare population, remains unclear.


To estimate nationally-representative changes in prescription drug use and out-of-pocket drug costs two years after implementation of Part D.

Design, Setting, and Participants

We examined study outcomes over 8 years (2000–2008) and estimated changes after Part D, accounting for prior trends. Our analyses used the community-dwelling sample of the Medicare Current Beneficiary Survey (unweighted unique n=38,798). Actual post-Part D outcomes were compared to projected values using 2000–2005 data. Subgroup analyses and standardization weights were used to address population-level shifts over time in health status and demographic characteristics.

Main Outcome Measures

Annual prescription drug fills and out-of-pocket drug costs.


We observed significant average per person increases of 1.8 prescription fills (95% confidence interval [CI]: 1.1, 2.5) in 2006 and 3.4 prescription fills (95% CI: 2.7, 4.1) in 2007 above pre-Part D increases of 0.9 prescription fills per year. Average out-of-pocket drug costs decreased significantly by $143 (95% CI:−182.5,−103.1) in 2006 and $148 (95% CI: −181.2, −114.1) in 2007 above average pre-Part D increases of $12 per year. Prescription fills did not change for beneficiaries with fair to poor health until 2007 when large increases occurred (increases of 3.7 to 11.0 fills above pre-Part D trends). Poor beneficiaries without Medicaid had no reductions in out-of-pocket drug costs in 2006 or 2007.


After the transition year of 2006, the impact of Part D appeared larger and more consistent across the Medicare population. Of note, sick and poor beneficiaries experienced significant improvements in prescription drug use in 2007.

Five years after its implementation, Medicare’s outpatient prescription drug insurance program for disabled and older adults (Medicare Part D) has become a staple of the Medicare system. As of 2010, more than half of the Medicare population (over 28 million out of 46 million beneficiaries) had enrolled in the program and over 1,576 stand-alone prescription drug plans and Medicare Advantage organizations competed for these enrollees.(1) Medicare Part D has even received its first major revision---the gradual phase out between 2011 and 2020 of the no coverage zone known as the “doughnut hole”.

To date, there have been over sixty published evaluations of whether and how much Part D expanded prescription drug coverage, increased use of medications, and decreased out-of-pocket costs.(2) However, surprisingly, most of these studies have major methodogical limitations. Nationally-representative Part D utilization data were not available until recently, so for the first four years of program operation, most of what is known about Part D has come from research using data from pharmacy chains or single health systems.(2) While informative, these early studies are insufficient since they include non-representative populations (which cannot be characterized in pharmacy chain data), do not fully capture patient drug utilization, and were generally limited to the transitional year, 2006. Rigorous research methods were also often lacking. About two-thirds of the early Part D studies did not use actual Part D data or apply robust longitudinal study designs that controlled for pre-existing trends or use actual Part D data. Some studies compared prescription drug use of Part D enrollees to that of non-Part D enrollees, although adverse selection has been clearly documented in voluntary enrollment.(3) As a result, current estimates of the average impacts of Medicare Part D in its first year of operation span a fairly large range; relative changes in overall out-of-pocket drug (OOP) costs range from a 32% decrease to a 13% increase, while relative increases in overall prescription drug fills range from 6% to 44%.(49) With few exceptions, almost none of these studies have provided estimates that can be generalized to the Medicare population or extended beyond the transitional period of 2006.(67, 1011)

The lack of nationally-representative evaluations of Part D is an important gap in current research. Our prior research on respondent-reported cost-related medication nonadherence suggests that the impact of Part D was neither uniform across the Medicare population nor limited to the first year of implementation. In a national survey of medication nonadherence due to costs we could find no evidence of relief among vulnerable groups such as the disabled and individuals with multiple chronic conditions in the first year after Part D implementation. Furthermore, between 20% and 26% of at-risk beneficiaries continued reporting major financial burdens or medication nonadherence due to drug costs as late as 2007.(1213)

The objective of the present study is to provide a comprehensive evaluation of drug utilization and OOP costs that is generalizable to the entire community-dwelling Medicare population. We report overall trends in prescription drug use and OOP drug costs in the 6 years before and 2 years after implementation of Medicare Part D (the latest data available for prescription drug use and OOP drug costs). Our estimation of the impact of Part D is based on the difference between actual utilization data, standardized to baseline population characteristics, and projected results based on historical trends from 2000 to 2005. Subgroup analyses were conducted to detect evidence of any adverse impacts in potentially vulnerable populations, many of whom have not been previously studied in Part D evaluations.


The Medicare Current Beneficiary Survey (MCBS) is a continuous face-to-face panel survey of a representative national sample of Medicare beneficiaries conducted by the Center for Medicare and Medicaid Services (CMS).(14) Since 1991, the MCBS has provided detailed longitudinal data on annual samples of Medicare beneficiaries with a current sample size of approximately 12,000 community-dwelling and institutionalized elderly and disabled. The rich variety of measures includes demographic information, income, assets, living arrangements, family supports, health status, changes in health status, functioning, health behaviors, health insurance coverage, drug coverage, health services utilization (including copayments, deductibles, and non-covered services), and access to medical care.

The sample for the MCBS is drawn from Medicare enrollment records according to a multi-stage rotating panel sampling plan, with the sample replenished each year. Respondents are interviewed in person three times a year using Computer Assisted Personal Interviewing (CAPI), resulting in very high response rates (initially ~85%). The typical MCBS interview lasts approximately one hour. Interview cycles begin with the fall interview, which includes demographic and household composition, as well as health insurance, and health status. The subsequent interviews collect detailed information on health care utilization and expenditures. Each respondent is asked to keep a record of insurance statements, receipts, and prescription bottles, in order to enhance the accuracy of data collection. Our analysis used the Cost and Use files from the MCBS that are the main source of medication use data and are now available through 2007.


The sample included community-dwelling Medicare enrollees from 2000 through 2007. We excluded institutionalized respondents because prescription drug expenditures are not captured for this population (n=5,882 out of 91,855 total person-years in the 8-year Cost and Use survey population). We then excluded individuals who did not respond to the prescription drug use survey or individuals with three or fewer months of entitlement (n = 5,755) to ensure a reliable time-frame for estimating annual trends. Our annual samples ranged from 11,167 in 2000 to 11,995 in 2007. Accounting for overlapping panels across years of data, the total number of unique individuals in this study was 38,777. For individuals with 4 to 11 months of entitlement in a calendar year, we weighted their data to reflect the partial year contribution (e.g., annual observations for a person observed only 4 months were subject to a weight of 0.33). Three-year sample panels were constructed to examine unadjusted changes in medication use within the same individuals included in the survey from 2005 to 2007. (Two-year sample panels were also constructed for sensitivity analyses, and results were consistent so are not reported here.)

Study variables

Prescription fills and out-of-pocket drug costs

Prescription drug fills and costs were drawn from the MCBS Cost and Use files in all years (2000 to 2007).(15) Beginning in 2006, the MCBS included prescription drug claims for Part D enrollees in addition to self-reported drug use. Our preliminary comparison of 2006 prescription drug fills by data source showed a 16.8% undercount by self-reported drug fills compared to drug claims, which is consistent with a 17.7% undercount found in a previous validation study using 1999 MCBS data.(15) For this study, we used only self-reported drug fills to ensure consistency across all study years. By excluding the claims-only prescription drug use, our estimates of changes attributable to Part D are not biased by the additional and as yet unvalidated data source. All drug costs were adjusted to 2007 values using the consumer price index to account for inflation.(16)

Prescription drug coverage

We ascertained drug coverage either through a “yes” response to the survey question “Does your (supplemental insurance) provide drug coverage,” a medication bill provided during the interview documenting third-party payment, or from data in Medicare’s administrative files, as in previous reports.(17) We also constructed a measure of changes in drug coverage status between 2005 and 2007 with the following categories: had no drug coverage in all three years, gained drug coverage after Part D (2006 or 2007), or had drug coverage in all three years. We excluded people who experienced a loss of drug coverage after Part D because of insufficient sample size (n=37).

Covariates include: age; sex; race and Hispanic ethnicity; geographic residence; disease burden as measured by a count of specific conditions; Medicaid enrollment; income; and general health status.(13, 18) All variables are self-reported by survey respondents, except for sex, Medicaid enrollment, and age, which come from Medicare’s administrative files. Self-reported income is known to be underreported so we inflated the income by 20%, as recommended previously, and categorized the adjusted income in relation to the federal poverty levels.(19) Our comparison of the adjusted income categories to those reported in the Current Population Survey showed consistent results.

Statistical analyses

We described demographic and health characteristics of the MCBS population in 2000 and 2007, weighted to represent the overall population of community-dwelling Medicare beneficiaries.(14) We calculated unadjusted means of prescription drug fills and OOP prescription drug costs with 95% confidence intervals from 2000 to 2007.

To model changes in prescription drug fills and OOP drug costs, we used time-series regressions with a first-order autoregressive correlation structure(20) and only pre-Part D data (2000–2005). For each outcome, the model contained an intercept and an indicator of the yearly trend. Using parametric bootstrapping techniques,(21) we conducted 10,000 simulations of post-Part D outcome means (2006 and 2007) based on the null hypothesis that the data would follow the previously estimated trend. We then compared the simulated results to the observed results. We created percentile intervals based on the proportion of simulated values that exceeded the observed value (in the case of average prescription drug fills) or fell below the observed value (in the case of average OOP drug costs), assuming a one-sided test. We repeated these analyses separately in 18 subgroups based on sex, race/ethnicity, geographic residence, poverty status, and morbidity burden, stratified by self-reported health status (excellent to good vs. fair to poor).

Lastly, we repeated these procedures using adjusted outcome means that were standardized to account for changes in characteristics of the study sample between 2000 and 2007.(2223) Weights were derived from stratum-level characteristics of the standard population in 2000 and applied to the subsequent yearly estimates to create expected means, holding demographic characteristics constant. We then tested for differences between the standardized outcome means and our times series estimates.

We conducted all analyses using Stata version 10 (StataCorp LP, College Station, Texas) and SAS version 9.2 (SAS Institute Inc, Cary, North Carolina), and the a priori level of statistical significance was p<0.05. This study was approved by the Human Subjects Committee of Harvard Pilgrim Health Care.


Characteristics of Medicare Beneficiaries, 2000 and 2007

The demographic and health characteristics of the community-dwelling Medicare population between 2000 and 2007 were very stable over time with some exceptions (Table 1). There were slight increases in the proportions of non-elderly disabled Medicare beneficiaries and beneficiaries aged 85 or older. The proportion of beneficiaries with Medicaid coverage grew as did the proportion in the wealthiest tier. The percentage of respondents having three or more morbidities increased from 46% of the population in 2000 to 53% in 2007. Finally, about a quarter of beneficiaries reported having no drug coverage in 2000 compared to only 4% in 2007.

Table 1
Demographic and Health Characteristics of Community-Dwelling Medicare Beneficiaries, 2000 and 2007

Changes in Study Outcomes for Medicare Beneficiaries Overall, 2000–2007

Figure 1 displays unadjusted annual averages of prescription drug use and OOP drug costs among community-dwelling Medicare beneficiaries. We observed an average increase of 5.4 prescription fills per person in the first two years following Medicare Part D implementation (from 31.1 fills in 2005 to 34.2 in 2006 and 36.5 in 2007), which is larger than any increase during the pre-policy years from 2000 to 2005 (historical trend: average increase of 0.88 fills per year). Concurrently, there was a $110 decrease in the average annual OOP drug costs 2 years after implementation of Medicare Part D (from $611 in 2005 to $499 in 2006 and $501 in 2007), relative to the historical average increase of $12 per year). The lack of overlap in the 95% CIs between 2005 and 2006 and between 2006 and 2007 suggests significant increases in prescription fills in both 2006 and 2007. Significant decreases in OOP drug costs also occurred between 2005 and 2006, but not between 2006 and 2007.

Figure 1
Unadjusted Averages for Annual Prescription Drug Use and Out-of-Pocket Drug Costs Among Community-Dwelling Medicare Beneficiaries Overall, 2000–2007

Table 2 compares the adjusted average changes in prescription drug use and OOP drug costs after implementation of Part D, estimated using time-series regression and population standardization methods. Accounting for historical trends and holding population characteristics constant, we estimated a 1.8 increase (95% CI: 1.1, 2.5; p<0.001) in annual prescription fills per person and a $142 decrease (95% CI: −182.5,−103.1; p<0.001) in annual OOP drug costs per person in the first year after Part D implementation. In the second year, the estimated increase in fills doubled to 3.4 fills above the predicted average (95% CI: 2.7, 4.1; p<.001) and the decrease in OOP drug costs per persons was sustained at $148 below predicted (95% CI: −181.2,−114.1; p<.001).

Table 2
Estimated Changes in Annual Prescription Drug Fills and Out-of-Pocket Drug Costs Following Part D Implementation among Community-Dwelling Medicare Beneficiaries

Changes in Study Outcomes for Subgroups of Medicare Beneficiaries, by Health Status

Results from the subgroup analyses are summarized in Figures 2A and 2B (details available in the supplemental appendix). Among beneficiaries in excellent to good health, average prescription drug use increased significantly by 0.7 to 4.7 fills in 2006 and by 1.0 to 4.8 fills in 2007 above projected levels based on pre-Part D historical trends. All increases were significant except those for rural beneficiaries. For beneficiaries with fair to poor health, average prescription use did not significantly change after Part D until 2007, when the increases were generally large. For example, based on historical trends before Part D implementation, the average use of prescription drugs among black non-Hispanic beneficiaries in fair to poor health was predicted to be 44.1 fills in 2006 (95% CI: 38.2,49.9; p=0.411) compared to their actual average use of 44.8 fills. One year later, however, their use increased significantly to 6.2 fills over the predicted 2007 level (actual mean 55.6 fills, 95% CI: 46.5, 56.6; p=.014). By 2007, all Medicare subgroups reporting fair to poor health had experienced significant increases in prescription use ranging from 3.7 to 11.0 above estimated levels except for the wealthiest beneficiaries whose observed and predicted levels of prescription use were similar.

Figure 2
Figure 2A. Absolute Differences between Observed and Predicted Average Annual Prescription Drug Fills among Medicare Subgroups based on Demographic Characteristics, according to Self-reported Health Status

In comparison, average annual OOP drug costs decreased significantly by $86 to $302 in 2006 over predicted levels across nearly all subgroups of Medicare beneficiaries, regardless of health status, and these decreases were sustained in 2007 (Figure 2B). For example, based on historical trends before Part D implementation, the average costs of prescription drugs among dual-eligible Medicaid beneficiaries in excellent to good health was predicted to be $256 in 2006. Their actual 2006 average cost was $144 (p<0.001). Significant decreases in OOP drug costs over predicted levels were experienced by all groups except one, individuals in fair to poor health with incomes less 100% FPL who were not enrolled in Medicaid.

Changes in Study Outcomes for Medicare Beneficiaries by Insurance Status

Figure 3 displays the observed changes in average prescription drug use and OOP drug costs among the 3-year cohort of Medicare beneficiaries who were continuously enrolled from 2005 to 2007, classified by changes in drug coverage status over the three years. We observed a larger absolute increase in drug use among beneficiaries who gained or maintained drug coverage after Medicare Part D implementation (gained drug coverage: 26.0 fills in 2005 to 32.7 fills in 2006 and 37.1 fills in 2007; maintained drug coverage: 32.5 fills in 2005 to 37.3 fills in 2006 and 39.0 fills in 2007) as compared to those who had no drug coverage from 2005 to 2007 (10.4 fills to 11.3 fills and 11.9 fills). During the same time period, Medicare beneficiaries who gained drug coverage after Part D experienced an absolute decrease of $333 in average OOP drug costs between 2005 and 2006 and the lower level was sustained in 2007 (a nonsignificant additional decrease of $104). The year-to-year overlaps in 95% CIs for groups that had consistent drug coverage or consistent lack of coverage suggest that the observed changes in OOP costs for these groups were not significant.

Figure 3
Changes in Average Annual Prescription Drug Use and Out-of-pocket Costs among 3-year Cohorts of Medicare Beneficiaries, 2005–2007


Our study found that after the transition year of 2006, the impact of Part D appeared larger and more consistent across the Medicare population. Of note, sick and poor beneficiaries experienced significant improvements in prescription drug use in 2007. Like the earlier studies, we found that the first year of Part D was associated with a modest but significant 6% increase in average prescription drug use over predicted levels (32.8 vs. 31.0) across the overall Medicare community-dwelling population.(2) However, we found a much larger impact in 2007, namely a significant 14% increase in average prescription drug use over predicted levels (35.0 vs. 31.6). Furthermore, the significant 23% decrease in average OOP drug costs over the predicted levels ($618 vs. $475) that we detected in 2006 was sustained in 2007 ($624 vs. $476). The stability of the lower OOP drug costs in the second year of operation is a new finding with important implications. If we take into account the large increases in prescription use in 2007, this means Medicare beneficiaries actually paid, on average, closer to 35% less in OOP drug costs per fill in 2007 than predicted levels based on historical trends. These are the largest estimates of the impact of Part D on OOP drug costs to date.

Furthermore, we found for the first time that nearly all of the increases in average prescription drug use detected in 2006 occurred among Medicare beneficiaries in excellent to good health. Our analysis found no significant increases in average prescription drug use in 2006 among most beneficiaries with fair to poor health including the non-elderly disabled, Black non-Hispanics, metropolitan residents, and poor beneficiaries within 200% FPL but no Medicaid assistance. However, we did find a lagged Part D impact for these same groups in 2007; Medicare beneficiaries in fair to poor health experienced significant and sometimes large increases of up to 11 additional prescription fills over predicted levels in 2007. In comparison, decreases in OOP drug costs occurred almost uniformly across health status groups in both 2006 and 2007.

Our study offers a unique combination of nationally-representative data spanning 6 years before and 2 years after the implementation of Part D and a rigorous longitudinal study design. Our use of time-series regressions and standardization weights allowed us to provide robust estimates of Part D impacts, especially after the transitional initial year. This study also offers one of the first assessments of whether the program had consistent impacts across vulnerable poor and chronically ill populations. Using self-reported drug fills to ensure consistency across all study years is another unique feature of this study. Use of self-reported drug fills overcomes several common biases of claims-based studies, including their omission of prescriptions filled at out-of-network pharmacies, free samples, and prescriptions not paid by insurance

The increases in prescription drug use and decreases in OOP drug costs in the community-dwelling Medicare population are consistent with our earlier findings that self-reported cost-related medication nonadherence significantly decreased in this population after implementation of Part D.(13) We can now support those findings with empirical evidence about similar changes in the actual use of medications and costs. Furthermore, we found no increase in medication use following Part D in the sickest segments of the Medicare population until 2007, echoing our earlier finding that the sickest groups reported only lagged improvements in cost-related nonadherence and foregoing other needs to pay for drugs.(13) We postulate that 2006 was a transitional period for two reasons. First, the deadline for enrollment in 2006 was delayed to May and thus that year captures partial year effects, as demonstrated by others.(6, 11) Second, there is some evidence that previously uninsured people had to learn to use and optimize the new insurance coverage, especially the first time entering a gap in coverage.(24)

Our study has several limitations. The six years of pre-policy data provide an important comparison and context for our analyses. However, additional post-policy years would provide more clarity about the long-term impacts of implementing Part D. (MCBS drug utilization data after 2007 are not yet available.) Certainly, the first year of the new program was a transitional period and results from 2006 may underestimate the true effects of Part D; however, the extent to which 2007 data represent stable effects is still unknown. Other factors unrelated to Part D (such as contemporaneous changes in the financial condition of Medicare beneficiaries) may also have influenced the observed changes in drug use and costs both before and after implementation of Part D. Thus, our results should still be considered early evidence until longer-term data are available.

In conclusion, in comparison to historical trends, we found significant population-level increases in prescription drug use and decreases in out-of-pocket drug costs in the first and second years after implementation of Medicare Part D. However, some subgroups of beneficiaries in fair to poor health experienced only lagged changes in prescription drug use, despite immediate decreases in drug costs. Further research is needed to determine whether essential medications were most affected by implementation of Part D, and whether increased access to those medications impacted health outcomes in nationally representative populations.

Supplementary Material


1. Hoadley J, Cubanski J, Hargrave E, et al. Medicare Part D Spotlight. Menlo Park, CA: The Henry J. Kaiser Family Foundation; 2010. Part D Plan Availability in 2011 and Key Changes Since 2011.
2. Polinski JM, Kilabuk E, Schneeweiss S, et al. Changes in drug use and out-of-pocket costs associated with Medicare Part D implementation: a systematic review. J Am Geriatr Soc. 2010;58:1764–1779. [PMC free article] [PubMed]
3. Levy H, Weir D. NBER Working Papers. Cambridge, MA: National Bureau of Economic Research; 2009. Take-Up of Medicare Part D: Results from the Health and Retirement Study.
4. Millett C, Everett CJ, Matheson EM, et al. Impact of Medicare Part D on seniors’ out-of-pocket expenditures on medications. Arch Intern Med. 2010;170:1325–1330. [PubMed]
5. Zhang Y, Donohue JM, Lave JR, et al. The effect of Medicare Part D on drug and medical spending. N Engl J Med. 2009;361:52–61. [PMC free article] [PubMed]
6. Yin W, Basu A, Zhang JX, et al. The effect of the Medicare Part D prescription benefit on drug utilization and expenditures. Ann Intern Med. 2008;148:169–177. [PubMed]
7. Ketcham JD, Simon KI. Medicare Part D’s effects on elderly patients’ drug costs and utilization. Am J Manag Care. 2008;14:SP14–22. [PubMed]
8. Lichtenberg FR, Sun SX. The impact of Medicare Part D on prescription drug use by the elderly. Health Aff (Millwood) 2007;26:1735–1744. [PubMed]
9. Nair KV, Jan SA, Wolfe P, et al. Impact of Medicare Part D on Utilization and Expenditures. The American Journal of Pharmacy Benefits. 2010;2:209–218.
10. Madden JM, Graves AJ, Ross-Degnan D, et al. Cost-related medication nonadherence after implementation of Medicare Part D, 2006–2007. Jama. 2009;302:1755–1756. [PubMed]
11. Basu A, Yin W, Alexander GC. Impact of Medicare Part D on Medicare-Medicaid dual-eligible beneficiaries’ prescription utilization and expenditures. Health Serv Res. 2010;45:133–151. [PMC free article] [PubMed]
12. Hsu J, Fung V, Price M, et al. Medicare beneficiaries’ knowledge of Part D prescription drug program benefits and responses to drug costs. Jama. 2008;299:1929–1936. [PubMed]
13. Madden JM, Graves AJ, Zhang F, et al. Cost-related medication nonadherence and spending on basic needs following implementation of Medicare Part D. JAMA. 2008;299:1922–1928. [PubMed]
14. Adler GS. A profile of the Medicare Current Beneficiary Survey. Health Care Financ Rev. 1994;15:153–163. [PubMed]
15. Poisal JA. Reporting of drug expenditures in the MCBS. Health Care Financ Rev. 2003;25:23–36. [PubMed]
16. U.S. Department of Labor. Consumer Expenditures in 2005. 2007. Bureau of Labor Statistics.
17. Briesacher BA, Stuart B, Poisal JA, et al. Methodological Issues in Estimating Prescription Drug Coverage Using The Medicare Current Beneficiary Survey. Washington, DC: AARP Public Policy Institute; 2006.
18. DeSalvo KB, Fan VS, McDonell MB, et al. Predicting mortality and healthcare utilization with a single question. Health Services Research. 2005;40:1234–1246. [PMC free article] [PubMed]
19. Goldman DP, Smith JP. Methodological biases in estimating the burden of out-of-pocket expenses. Health Services Research. 2001;35:1357–1364. discussion 1365–1370. [PMC free article] [PubMed]
20. Hamilton JD. Time Series Analysis. Princeton, NJ: Princeton University Press; 1994.
21. Efron B, Tibshirani RJ. An Introduction to the Bootstrap. Boca Raton, FL: Chapman & Hall/CRC; 1993.
22. Brownson RC, Petitti DB, editors. Applied Epidemiology: Theory to Practice. New York: Oxford University Press; 1998.
23. Kleinbaum DG, LKL, Morgenstern H. Epidemiologic Research: Principles and Quantitative Methods. New York: Van Nostrand Reinhold; 1982.
24. Schneeweiss S, Patrick AR, Pedan A, et al. The Effect Of Medicare Part D Coverage On Drug Use And Cost Sharing Among Seniors Without Prior Drug Benefits. Health Aff (Millwood) 2009 [PMC free article] [PubMed]