R-4: Multiple Chronic Conditions and Healthcare Access Disparities


The Medical Expenditures Panel Survey (MEPS) sample is based on household units selected as a sub-sample of households participating in the previous year's National Health Interview Survey (NHIS).  The MEPS provides a representative sample of the U.S. civilian non-institutionalized population (excluding those in prisons or nursing homes).  This sampling strategy will provide unbiased national and Census region estimates of health care expenditures and other health parameters that are precise enough to inform policy, particularly on priority subgroups.  The MEPS over-samples subgroups of persons who are Hispanic, African American, Asian, and those with a low-income status to increase the precision of estimates.  This will allow us to examine minority disability populations that are often too small to examine in other datasets.

Because the sampling frame for the MEPS is the NHIS sample, we can estimate the sample size of people with disabilities in each MEPS panel using the NHIS prevalence estimate of 15.8%. Based on an overall sample size on the Household component of approximately 19,000 adults per panel, this yields an estimate of approximately 3,000 adults with varying levels of disability in each panel. Data are weighted to yield national estimates. Each panel is independent of panels from previous years, allowing multiple panels to be combined. 

To provide reliable estimates of health disparities in care access among small subgroups, we will use a dataset that combines data across seven panels of respondents .  For their NIDRR-funded Disability and Rehabilitation Research Program (DRRP), our partners at the University of New Hampshire have already combined the seven most currently available MEPS datasets, yielding an estimated, unweighted sample size of 27,474. They have committed to providing us with this combined, state-of-the-art database for use in this project so that we may build on what they are doing. We hope to answer additional important research questions that resources and time prevent them from pursuing in their DRRP.

Data Collection and Measurement 

We will use the MEPS dataset as the data source for this investigation. Initiated in 1996, the MEPS collects data on how specific health services are used by Americans and the costs and methods of payments for these services. It has become the premier dataset for examining health care access disparities in vulnerable populations, with reviews conducted among elders, the uninsured, rural Americans, children with special health needs and women. Although the MEPS collects information on a number of disability identifiers and the Agency for Healthcare Research and Quality (AHRQ) regards individuals with disabilities as a “priority population,”  to our knowledge we have conducted the only studies with individuals with disabilities as the subject of a health disparity study using MEPS data. We will draw upon the methodology we developed for these studies in the currently funded RRTC/MICL for identifying subgroups of disability (physical disability, intellectual development, mental health, and sensory disability). In addition, a series of MEPS Methodology Reports provide guidance for variable construction.

Measurement Techniques

Key variables of interest relate to the factors identified by the Andersen Behavior Model and the requests from HHS (2010) and NIDRR (2011) to be related to multiple chronic conditions and health care access among persons with disabilities.  Like other existing national survey datasets that include measures of disability, the MEPS does not collect data on system-level factors such as neighborhood characteristics, urban/rural factors, or health provider and health facility availability. In order to address this need, the MEPS will be linked to the Area Resource File (ARF) to examine additional system-level factors along with MEPS individual- and system-level variables. Primary data for the ARF are supplied by both public (e.g., Bureau of the Census, Centers for Medicaid and Medicare Services, National Center for Health Statistics) and private (e.g., American Hospital Association, American Medical Association, American Dental Association) organizations. ARF datasets are released annually.  Each year’s release contains more than 7,000 variables describing health facilities, health professions, economic activity, socioeconomic and environmental characteristics.  The basic file contains geographic codes and descriptors that enable it to be linked to other secondary data files, including the MEPS. Data for 2006, 2005 and 2004 are currently available. 

Table. Items from MEPS and ARF by Conceptual Group
Conceptual GroupVariable: Potential Items from MEPS and ARF
Health Status

General health: Short form (SF)-12, SF-physical, self-rated health (x5), days missed work

Mental health: Kessler Index, SF-12 mental, self-rated mental health (x5)

Oral Health: Lost all teeth

Health Care Access

Availability: usual source of care

Utilization: number of office visits, got clinical screening

Perceived access: problems getting care

Satisfaction with care

Personal Health PracticesBMI, smoking, physical activity
MSA, region of country, urban/rural status (ARF), population per square mile (ARF)
   Health Care System
Physicians by detailed specialty and major professional activity (ARF), Number of hospitals (ARF), Number of federally qualified health centers (ARF)
Race, ethnicity, age, education, income (% poverty level), marital status, family size
Insurance type, employment, transportation,b to dental care
Population: NeedIll or injured and needed care (in 12 months), made routine care appointment, need care/test/treatment, need to see specialist, number of conditions

Data Analysis

Logistic regression and descriptive and exploratory analyses will be conducted using SPSS 19.0 for Windows. Prior to running analyses, data distribution for all variables will be examined using multivariate kurtosis, indicating use of Maximum Likelihood Estimation (MLE) or Maximum Likelihood Robust (MLR) procedures. Missing cases will also be examined and estimated by the full information maximum likelihood method.

Logistic Regression

In order to identify predictors of comorbidity of chronic diseases in the disability sample, we will use a logistic regression analysis with a dichotomous dependent indicator variable of those persons experiencing comorbidity (2 or more chronic conditions). Predictor variables are identified as related to the Andersen model, including predisposing characteristics, as well as environment variables, including location and health care system access. From this model we will be able to identify those variables which most influence the occurrence of comorbidity in the disability population (personal health practices, external environment, health care environment, and predisposing factors).

Descriptive and Exploratory Analyses

We will use descriptive analyses to identify the different common combinations (dyads and triads) of comorbidity within each of the disability subgroups. Those dyads and/or triads of chronic illness with high frequency counts will be further analyzed using exploratory descriptive analyses to determine the following: 

  • Are the most common combinations of chronic disease similar among the four disability subgroups?
  • Does health care access or health status differ across the most common combinations of chronic disease (health status, health care access variables)?
  • Do the population enabling and need variables differ across the most common combinations of chronic disease (enabling factors, needs variables)?

Finally, do those variables identified in the regression analysis as significant predictors of comorbidity differ across the most commonly occurring combinations of chronic disease variables (personal health practices, external environment, health care environment, predisposing factors)?