Methodological Issues in Using Multiple Years of the Medicare Current Beneficiary Survey

Abstract

Background:The analysis presented in this paper examines the multi-year capacity of the Medicare Current Beneficiary Survey (MCBS). Methods:We systematically reviewed the literature for methodological approaches in research using multiple years of the MCBS and categorized the studies by study design, use of survey sampling weights, and variance adjustments. We then replicated the approaches in an empirical demonstration using functional status (activities of daily living (ADL) and 2005 Results: In the systematic review, we identified 22 pooled, 17 repeated cross-sectional, and 17 longitudinal studies. Less than half of these studies explicitly described the weighting approach or variance estimation. In the empirical demonstration, we showed that different study designs and weighting approaches will yield statistically different estimates. Conclusion:There is a variety of methodological approaches when using multiple years of the MCBS, and some of them provide biased results. Research needs to improve in describing the methods and preferred approaches for using these complex data.Keywords: Medicare current beneficiary survey, Health Care Surveys, Survey design, Survey sampling weight, Systematic review, SAS, SUDAAN doi: http://dx.doi.org/10.5600/mmrr.002.01.a04 Briesacher, B.A., Tjia, J., Doubeni, C.A., Chen, Y., Rao, S.R. E2 INTRODUCTIONThe Medicare Current Beneficiary Survey (MCBS) is the principal national survey for informing and evaluating health policies for the Medicare program. Since 1991, the Centers for Medicare & Medicaid Services have sponsored this survey, at a cost of $14.8 million a year to administer. To date, nearly 300 published studies have used the MCBS for critical research questions, most recently for assessing the nationwide impact of the Medicare prescription drug program (Part D) (Madden, Graves, Ross-Degnan, Briesacher, & Soumerai, 2009;Madden et al., 2008). However, one aspect of the MCBS that is not well known is its capability for short longitudinal analyses of up to 3 years. Unlike many other national surveys, the MCBS collects follow up data on each individual for up to 3 years of observation. Important questions about health-related changes over time can be answered by combining multiple years of the MCBS.However, the longitudinal data structure of the MCBS can also complicate analyses and cause confusion about appropriate research approaches. Research that uses multiple years of MCBS data faces the following issues. First, there is a nontrivial overlap in the MCBS sample each year (approximately 60% from one year to the next). This means that the assumption of statistical independence is violated for many standard analytic approaches. Second, the repeated observations introduce an additional level of clustered data beyond the survey design. The MCBS's multi-staged complex sampling design creates data that are clustered in primary sampling units and strata, and that require special statistical approaches for correct estimation and tests. However, the use of multiple years of the MC...

Prescription Drug Spending and Medication Adherence Among Medicare Beneficiaries with Heart Failure

“…26,27 For participants with multiple years of data, a weighted average of their cross-sectional sampling weights was applied because exclusion of multiyear participants in pooled analyses of MCBS data is not recommended. 28 A sensitivity analysis was conducted without recipients of the low-income subsidy (LIS) for Part D cost-sharing assistance because the LIS results in negligible out-of-pocket spending, which could have altered refill behavior. 2 Data were analyzed with SAS version 9.4 (SAS Institute, Cary, NC).…”

“…Nonetheless, when variance estimates account for cluster sampling, it may be unnecessary to correct for the additional autocorrelation from repeated observations. 28 Finally, we excluded survey-reported prescriptions that were not matched to Medicare records, but only about 6% of unmatched prescriptions for common chronic disease medications in the MCBS reflect true out-of-plan use for Part D participants; the rest are likely duplicates. 22 ■■ Conclusions Despite having Part D drug coverage, some Medicare beneficiaries may have faced a modest financial barrier to optimal HF therapy: beta-blocker adherence was inversely related to out-of-pocket beta-blocker spending as a percentage of income in this study.…”