Associations of mortality rates with rate and income among United States Medicare participants

The racial disparities in mortality risks adjusted by socioeconomic status (SES) is addressed by a large amount of literature but still not well understood. In this thesis, we develop and apply statistical methods and models to investigate the relation among race, SES, and the risk of all-cause mortality, using a large data set that consists of more than 4 million Medicare enrollees who are 65 years and older residing in 2095 zip codes in the Northeast region of the United States, with SES mainly measured by income.; The thesis work contains three major analyses. We first develop and apply hierarchical statistical models to estimate the individual-level association between race and mortality risks, adjusted by both individual- and zip code-level income, for the Medicare population. We fit the models using a Bayesian approach via Markov chain Monte Carlo, and we apply multiple imputation to fill in the missing data. Results show a higher risk of death for Blacks compared with Whites, both adjusted and not adjusted for income. After the adjustment of both individual- and zip code-level income, there is a statistically significant reduction in the absolute difference while limited reduction in the relative difference between the mortality risks of the black population and white population.; We secondly propose a spatial smoothing approach for data masking to preserve confidentiality, and we investigate the bias of parameter estimates resulting from analyses using the masked data for Generalized Linear Models as a function of both the form and the degree of masking. We apply the approach to the study of racial disparities in mortality risks for the Medicare population, and the bias of the estimated odds ratio of death comparing Blacks versus Whites highly depends on both the form and the degree of masking.; We further apply linear regression models as well as CART and random forests to study the difference between individual-level income and zip code-level median income, for a representative sample of the Medicare population. The results show that individual level marital status and education are the most important predictors of the individual- and zip code-level income difference.