Session: Drugs & Research
Room: Upson B17
Time: Wed 08:30-10:00

Presenter: Steven Hill (Agency for Healthcare Research and Quality (AHRQ). Center for Financing, Access and Cost Trends)

Discussant: Benjamin Cook (Cambridge Health Alliance/Harvard Medical School)

Abstract

The Medical Expenditure Panel Survey (MEPS) is a unique, nationally representative, source of micro data on health care use and expenditures for all payers. Household respondents report service use while follow-back surveys of pharmacies, hospitals, physicians, and other providers are the primary source of price and expenditure data. The validity of the data is critical, because the MEPS is widely used for national estimates, behavioral modeling, and policy simulations, including many analyses of prescription drug markets.
Previous studies found that respondents reported inpatient hospitals stays well but tended to underreport ambulatory services (office, hospital outpatient department, and emergency department visits) in the MEPS (Zuvekas and Olin, 2009a, 2009b). However, these same studies also suggest that biases introduced into behavioral and distributional analyses are likely small and that simple adjustment strategies can correct effectively for underreporting.
We extend these previous studies to validate prescription drug use reported by household survey respondents and expenditures reported by pharmacies using a sample of Medicare beneficiaries in the 2006 and 2007 MEPS who were linked to their Medicare administrative records. The analytic sample contains 1,815 person-years of MEPS sample members who reported Medicare coverage and who have Medicare Part D drug coverage in the linked administrative data set.
The initial analyses of this sample assess (1) the concordance between the number of drugs and prescription drug fills reported by the household in the MEPS and the numbers in the administrative data, and (2) the concordance between the out-of-pocket and Part D expenditures reported by pharmacies in the MEPS and expenditures found in the Part D claims. We further investigate the sample member and interview characteristics (such as self versus proxy response and recall period) associated with level of concordance.
Next, we investigate whether reporting errors in MEPS lead to systematic biases in behavioral analyses by estimating pairs of utilization regressions using the claims and MEPS household reported utilization measures, respectively, as the dependent variable and comparing the results. The independent variables in each pair include an identical set of sociodemographic covariates. Logistic regression models are estimated for whether the person had any medication use, and negative binomial count data regression models are used for number of prescription drug fills. For total drug expenditures, we use a two part model, where the second part has a square root link. We formally test whether the marginal effects of each covariate is the same in the pairs of regressions. For example, does poor health increase drug spending by the same magnitude whether using the household-reported or claims-based measure? All analyses use MEPS sampling weights (adjusted for the non-matches to Medicare administrative records) and the method of balanced repeated replications (BRR) to adjust for the complex design of the MEPS survey.

Key Terms

measurement error, prescription medications, Medicare

Authors:

Samuel H Zuvekas (Agency for Healthcare Research and Quality. Center for Financing, Access and Cost Trends) , Steven Hill (Agency for Healthcare Research and Quality. Center for Financing, Access and Cost Trends) , Edward Miller (Agency for Healthcare Research and Quality. Center for Financing, Access and Cost Trends) and Marc W Zodet (Agency for Healthcare Research and Quality. Center for Financing, Access and Cost Trends)