Chronic diseases pose major health and economic burdens for Americans. Not only are chronic diseases the leading causes of disability and death, but they also account for approximately 75% of total health care costs in the United States. Building upon existing life table-based methodologies, this project will use models of chronic disease progression that incorporate current biological understanding of chronic diseases and Markov Chain theory, a statistical theory used to describe discrete stochastic processes, to model individuals' chance of transitioning across different disease progression-related health states. This project has two objectives. First, the project will design biologically plausible models of chronic disease progression for diabetes and chronic obstructive pulmonary disease (COPD) complemented by surveillance data to inform disease-specific multistate life tables. Second, the project will use Markov Chain Theory to estimate the likelihood of transitioning across the health states of these biologically plausible models conditional on age and sociodemographic factors and calculate health-state specific life expectancies. In collaboration with South Carolina's Office of Research and Statistics, the project also will apply this methodology to statewide data on State Health Plan recipients from 1999 to 2008, complemented by information on mortality from the state death registry and the National Death Index - Plus.
The intellectual merit of this research is its interdisciplinary construct, which combines biological models of chronic disease progression, statistical theories of stochastic processes, social theories of the fundamental causes of health disparities, and epidemiologic knowledge of population health. Broader impacts of this research include the potential benefits to public health professionals of a methodology for identifying health disparities in the lifetime burden of specific chronic diseases. In addition, it is likely that this methodology will be generalizable to other chronic diseases and to other populations. As a Doctoral Dissertation Research Improvement award, support is provided to enable a promising student to establish a strong, independent research career.
Population-level disparities in health are of great concern to public health researchers and health policy makers. We explored whether gender and racial residential segregation, a structural-level determinant of health disparities, contributed to the chronic disease burden at the macro-level. We developed a novel methodological approach based on Markov Chain theory and multistate life table techniques to estimate health state-specific life expectancy of a population. Our method employed an innovative weighting mechanism to aggregate annually assessed individual disease trajectories over time, while accounting for initial age-specific cohort sizes. We applied this approach to assess the chronic disease burden, specifically with respect to diabetes, among South Carolina State Health Plan recipients between 1999 and 2008. There were a total of 642,449 South Carolina State Health Plan recipients, excluding Medicare primary enrollees, who were covered at any time from 1999 to 2008. We excluded 30,626 recipients because they did not reside in South Carolina. Of the remaining 611,823 recipients, 81% (n=495,830) were excluded because they did not meet the inclusion criteria for a restricted cohort given they were not covered continuously from 1999 to 2008 or until their death within South Carolina before the end of the study period. Thus, the results of this analysis are based on available claims data for a restricted cohort of 115,993 South Carolina State Health Plan recipients. About 3% (n=3,988) of recipients had been diagnosed with diabetes either in 1999 or before the observation period started, which represented our initial diabetes prevalence rate. The proportion of recipients with diabetes steadily increased over each year of the observation period reaching a maximum of 14.5% in 2008. About 3.0% (n=3,425) of the recipients died over the course of the study period, with 8.2% (n=282) of those deaths related to diabetes. In summary, males had lower total life expectancy, well life expectancy (life expectancy pre-diagnosis of diabetes), and sick life expectancy (life expectancy post-diagnosis of diabetes) than females. We did not observe an impact of racial residential segregation on total life expectancy. Although we did not observe an effect of racial residential segregation on the well life expectancy for males, we did observe an advantage for younger females from low segregated regions and older females from moderately segregated regions. The impact of racial residential segregation on sick life expectancy for younger males was not as substantial as it was for younger females. However, the effect began to level out for both gender groups after about 30 years of age. This dissertation research had several strengths. First, we were able to follow individuals longitudinally over time rather than rely on cross-sectional sampling strategies. Second, our novel weighting mechanism allowed for the aggregation of transition probabilities over time, more heavily weighting larger initial age-specific cohorts given their sample sizes would yield more reliable estimates. Thirdly, our research consisted of a comprehensive multilevel analysis in that we were able to investigate differences in gender, an individual level factor, and in racial residential segregation, a structural level factor. Although our investigation into the impact of racial residential segregation on health state-specific life expectancy was a sociologic strength of the application of our proposed methodology, we do suggest taking caution when interpreting the results due to the relatively low level of racial segregation in the regions we categorized as having moderate residential segregation. However, when we did observe effects of racial residential residential segregation on health state-specific life expectancy, the impact was not constant across the life course. Lastly, our methodology gives public health professionals and health policy makers easily interpretable metrics to compare the chronic disease burden among various subgroups of a population, as we did, across populations, and across time periods. There were three major limitations of our research. First, we did not have information on the State Health Plan recipients’ race/ethnicity. Second, because of data limitations, we did not take diabetes duration into account when estimating the age-specific probabilities of transitioning from the sick state to either of the death states. Third, given our restricted cohort represented a relatively small subset of South Carolina State Health Plan recipients, our results may not be generalizable to the larger, more mobile population of State Health Plan recipients or, more importantly, to the larger population of South Carolina residents, which would include uninsured individuals. In conclusion, our dissertation research applied multistate-life table methodology and Markov Chain theory to address disparities in health state-specific life expectancy at the individual level and at the structural level. Our multidisciplinary assessment provided a comprehensive overview of health disparities that can challenge researchers to explore the underlying mechanisms for these disparities and in addition provide easily interpretable evidence for public health professionals and policy makers to identify the most vulnerable populations and/or subgroups of a population so that effective interventions and/or policies can be implemented.
