Study Design, Data, and Sample
This was a retrospective, cross-sectional archival record study. Data was obtained from the 2012 Maryland Clinical Public Use Data. Because this is a public data source containing no personal identifiers, no Institutional Review Board application was necessary to access the data for research purposes.
We choose Maryland as the site for our study because this state applies the same capitation model for all enrollees in government insurance plans. Past studies that include variables on social determinants of health were confounded by financial considerations, which influence the access to and level of care due to costs. In Maryland, a federal waiver exempts hospitals in the state from the national Medicare fee schedule. As a result, Maryland hospitals are placed under the same payer system to contain cost growth, improve access to care, and implement payment innovations. Consequently, confounds in utilization rates and differences in the restriction of care between Medicare FFS and Medicare MC patients are minimized because the payment rate to healthcare providers in Maryland is the same for both groups of patients.
The data included 694,488 patients who were hospitalized in Maryland hospitals in 2012. Among these patients, 85,712 (12.3%) were diagnosed with DM. We limit the DM sample to those who were at least 65 years old who had the opportunity to choose between Medicare and private insurance. We excluded Medicaid enrollees because it is not a federal-only insurance program but involves state partnership. The final sample for analysis included 43,519 DM patients comprising 37,825 Medicare FFS, 1,926 private FFS, 2,736 Medicare MC, and 1,032 private MC patients.
Three health outcomes were examined in this study. The first was hospital length of stay (LOS) measured in days, which indicates the efficiency and quality of hospital care. Related to quality of care is the 30-day readmission from the indexed DM-related discharge, which was coded as 1 for yes and 0 for no. The third outcome was whether the patient was hospitalized as having end-stage renal disease (ESRD), which was coded as 1 for yes and 0 for no. ESRD is a costly and disabling condition associated with DM.
Control and Predictor Variables
Three groups of control variables were included in the analysis as they have influence on health outcomes even though they were not the focus of this study. The first group of control variables were related to hospital characteristics where the patient received care. Specifically, hospital size and teaching status have been associated with access to specialists and resources, which influence health outcomes. Hospital size was coded as 1 if the hospital had more than 400 beds and as 0 otherwise. Operating costs increase for hospitals with over 400 beds due to diseconomies of scale. The teaching status of the hospital was coded as 1 if the hospital is an academic teaching hospital and as 0 otherwise. 30-day mortality rate was found to be lower in teaching hospitals compared to non-teaching hospitals.
The second group of control variables that were related to health outcomes but were not the focus of this study were the patients’ demographics, behavioral and biological factors. Age, measured in years, was included because almost 80% of US adults 65 years and older have some form of dysglycemia, which has significant implications for DM. Gender, coded as 1 for male and 0 for female, was included because DM is more common among men. Marital status, measured as 1 for married and 0 otherwise, was included because unmarried individuals tend to have higher health risks than their married counterparts. Smoking, measured as 1 for smoking and 0 otherwise, has been associated with increased risk of Type 2 DM. Asthma, measured as 1 for asthma and 0 otherwise, and tuberculosis, measured as 1 for tuberculosis and 0 otherwise, were included because DM was associated with impaired pulmonary function. Depression, measured as 1 for depression and 0 otherwise, was included because mood has been associated with adherence to medication regimens for glycemic control. Finally, obesity, measured as 1 for obesity and 0 otherwise, is associated with DM because obese individuals secrete resistin, a hormone that causes insulin resistance.
The third group of control variables were the patients’ social determinants of health. The first variable was SES. Since the 2012 Maryland Clinical Public Use Data did not provide the patients’ income data, we used the patients’ zip codes to approximate average household income in accordance with common practices in the literature. A second variable was the size of the patients’ residential neighborhood. Patients who live in metropolitan areas, with populations greater than 1 million people, were coded as living in Urban Neighborhoods and were coded as 1 to reflect the degree of congestion and access to community resources and 0 otherwise. Rural areas have fewer physicians per capita than urban areas, which indicate inequitable healthcare access. A third variable was race, measured as two dichotomous variables, White and Black, because race and ethnicity significantly affect health outcomes due to food insecurity, which can affect glycemic control. By controlling for social determinants of health disparities that has traditionally been the result of a lack of insurance access, we limit selection bias associated with financial considerations that include only healthier patients who have access to insurance and healthcare.
Our predictor variables were the four different types of insurance plans: Medicare FFS, Medicare MC, private FFS, and private MC insurance plans.
To partial out the effects of the control variables, we used multiple hierarchical regression analysis to estimate the effects of insurance plans for LOS and logistic hierarchical regression analyses to estimate for 30-day readmission and ESRD. Hierarchical regression analysis determines whether insurance plans explained additional variance on the outcome variables, after hospital characteristics, patient demographics, behavioral and biological factors, and social determinants of health were considered. We entered the data in incremental blocks of information first by hospital characteristics, second by patient demographics, behavioral and biological factors, third by social determinants of health, and fourth by insurance to separate the various influences on the outcome variables. We report the standardized regression coefficients or beta, which considers the standard errors of the regression coefficients, for multiple regression analysis and Odds Ratio for logistic regression analyses, which indicates the likelihood that an outcome results from the influence of the predictor variable. An Odds Ratio below 1.0 indicates a likelihood of less than 100% that an outcome will occur, whereas an Odds Ratio above 1.0 indicates a likelihood greater than 100% that an outcome will occur.