RESEARCH DESIGN AND METHODS
4.2 Data Sources
In this retrospective study, two data sources were used to test the hypotheses mentioned in Chapter 1: Medicare Current Beneficiaries Survey (MCBS; 2006-2010), and Area Health Resources Files (AHRF; 2006-2010). Based on the conceptual
framework, both individual-level and contextual-level data is necessary for this project. MCBS provides the individual-level information on independent variable (type of Part D plans), primary outcomes of interest (health services use and costs and medication adherence) and covariates at the individual level (demographics, socioeconomics, health conditions and health status, and health-seeking attitudes). AHRF data provides the county-level covariates that are relevant to health services use (e.g., local availability of health care system) and instrumental variable (e.g., PDP penetration rate). The following sections describe the data files and linkage of these data files for further analyses.
4.2.1 Medicare Current Beneficiary Survey (MCBS)
Medicare Current Beneficiary Survey (MCBS) is a longitudinal and multi- purpose survey of nationally representative sample of Medicare beneficiaries, including both aged and disabled beneficiaries residing in the community or long-term care
facilities in the U.S. and Puerto Rico.138 MCBS is conducted continuously by Centers for Medicare & Medicaid Services (CMS), Department of Health and Human Services. MCBS has been previously used in evaluating the impact of HMOs on health care utilizations and expenditures compared to Fee-For-Service (FFS) plans.139,140
MCBS employs a multistage stratified random sampling design and a rotating panel design. The purpose of this multistage sampling design is to reduce the costs of traveling for interviews while maintaining the national representativeness of Medicare beneficiaries. In the first stage of sampling, 107 geographic primary sampling units (PSUs), which consist of counties or groups of counties, were selected to represent the nation. Within PSUs, samples were restricted to address (zip codes) within a total of 1,163 sub-PSU areas selected using systematic sampling. To better represent the areas of
the nation, MCBS added or replaced several PSUs, primarily Western and Southwestern, which had experienced major growth in their elderly population, since the 1980 census. The survey sample was drawn by systematic random sampling within age strata from an enrollment list of Medicare beneficiaries residing in these areas. Approximately 16,000 Medicare beneficiaries were interviewed each round, and only 12,000 beneficiaries completed all four interviews each calendar year due to rotating panel design.138,141 The response rate is around 80%. All personal identifying information was removed for confidentiality purposes.142-144
Initial interviews are conducted each fall, and collect information on demographics, socioeconomic characteristics and medical conditions. The follow interview is divided into rounds three times yearly, to collect information on the use of health care services and prescription drugs, health insurance coverage, and sources of payment. Data related to the health status is collected in the third round. The annual interview lasts around 1 hours, and covers a variety of demographic and behavioral questions such as income, assets, living arrangements, satisfactory to health care systems, and access to medical care. MCBS interviewed the sampled person directly, but if he/she was unable to answer the questions, he/she would be asked to designate a proxy
respondent, usually a family member or close acquaintance who was familiar with his/her care. All interviewers are trained and retrained, particularly in analyzing insurance statements, apportioning payments, and dealing with the stresses of interviewing the patients who are chronically ill. Spanish translation is provided for Hispanic persons who cannot speak English.
Medicare claims, and filled in and corrected survey-reported payment amounts with more accurate information from bills submitted to and paid by Medicare. Hence, the final database consists of data from survey and administrative claims.142-144
MCBS contains two modules – the Cost and Use and the Access to Care modules. This study used data from these two modules of MCBS 2006-2010. The data files used for this study is presented in Table 5.1. In the Cost and Use module, RIC K file provides survey information for each beneficiary, RIC 1 and RIC 2 files provide self or proxy- reported demographics, socioeconomics, and clinical conditions, RIC 4 files provide self or proxy-reported insurance coverage for each beneficiary, RIC IPE, OPE, MPE, DUE, and FAE files provide self-reported medical records on an event level. RIC PME files provide pharmacy claims from both self-reports and claims. In the Access to Care module, RIC 3 files provide information on beneficiaries’ CRN and care-seeking attitudes. The key variables used in this study are described in Appendix A.
Table 4.1 List of Data Files in MCBS
Module Record Type Contents
Access to Care RIC 3 Access to Health care
Cost and Use
RIC K Key information
RIC A Administrative identification
RIC 1 Survey identification
RIC 2 Survey health status and functioning
RIC 4 Survey health insurance
RIC 5 Living conditions
RIC X Survey cross-sectional weights
RIC DUE Dental Events
RIC IPE Inpatient hospital events
RIC MPE Medical provider events
RIC OPE Outpatient hospital events
Module Record Type Contents
RIC IUE Institutional Events
RIC FAE Facility Events
4.2.2 Area Health Resource File (AHRF)
Area Health Resource File (AHRF) is a comprehensive health resource database that is administered by Health Resources and Services Administration (HRSA). AHRF collects data for every county in the U.S., and covers more than 50 data sources, such as American Medical Association, U.S. Census Bureau. AHRF provides a broad range of information on health care providers and facilities, and population characteristics on the county level. AHRF is available for public use and released annually. Data from AHRF 2014-2015 was used for this dissertation to provide information on the contextual factors related to health services use, and the instrumental variables (e.g., PDP penetration, % white collar jobs). The following table presents the key variables extracted from AHRF files and their data sources.
Table 4.2 Variables and Data Sources in the AHRF
Variable Data source Year
Primary physicians American Medical Association
Physician Masterfile 2010
Hospital beds American Hospital Association
Annual Survey of Hospitals 2006-2010
Per capita income U.S. Department of Commerce 2010
Number of population U.S. Census Bureau 2010
Education American Community Survey (ACS)
Summary File, U.S. Census Bureau 2010
PDP penetration Centers for Medicare & Medicaid
Services (CMS) 2008-2010
% white collar worker American Community Survey (ACS)
4.2.3 Linkage of data
MCBS data files were linked with AHRFs using county codes for each beneficiary. The following figure demonstrates the steps involved in pooling the data files for sample selection.
Figure 4.1 Flowchart of Database Preparation