Noncommunicable diseases (NCDs) such as heart diseases, kidney diseases, and cancer are the leading causes of mortality and morbidity worldwide. NCDs are often the cause of a combination of risk factors, including obesity, hypertension, dyslipidemia, hyperglycemia, and impaired renal function, which traditionally have been studied in isolation. This thesis adopts a data-driven approach by utilizing clustering techniques to analyze these risk factors jointly and identify cardiometabolic and renal phenotypes within populations.
The first aim of this thesis is to identify cardiometabolic and renal phenotypes in the United States. I used data from the National Health and Nutrition Examination Survey (NHANES) spanning from 1988 to 2018, and identified ten cardiometabolic and renal phenotypes in the US population, highlighting significant health trends. Notably, there was a decrease in phenotypes associated with high blood pressure and cholesterol, con- trasted by an increase in those characterized by severe obesity and poor kidney function, indicating a shift from phenotypes with high blood pressure and cholesterol towards poor kidney function and severe obesity.
In a second part, I extended this approach by developing a methodological framework that enables the comparison of cardiometabolic and renal phenotypes across multiple population subgroups and introduced a novel Bayesian hierarchical Gaussian mixture clustering model. I applied this model to data from three di↵erent age groups from NHANES data and illustrated the advantages of using this model rather than clustering these age groups together or separately.
The identification of clinically relevant phenotypes and their trends in the US, as well as the development of a new model for a multivariate comparison of cardiometabolic and renal health across populations, provide a foundation for future research on NCD risk factors and their distribution in various populations.