Hierarchical Bayesian mixture model for subgroup-specific cardiometabolic phenotypes

Anthropometric, cardiometabolic, and renal traits—including height, adiposity, blood pressure, blood glucose, cholesterol, and kidney function—affect numerous noncommunicable and infectious diseases. These traits share some genetic, environmental and nutritional determinants while also having distinct causes. Their complex distributions and correlations further vary among population subgroups. Traditional clustering methods for multiple continuous variables often overlook subgroup-specific distributions or hinder comparisons across separately clustered subgroups. We developed a Bayesian hierarchical mixture model to coherently identify anthropometric, cardiometabolic and renal phenotypes across population subgroups. Our model balances flexibility and comparability by adapting a Bayesian hierarchical mixture model to include subgroup-specific cluster weights and empirically driven shifts in cluster means. This structure allows subgroup-specific variations on phenotype characteristics while maintaining a shared structure and interpretation. We applied this model to anthropometric, cardiometabolic and renal data of men from three different age subgroups in the US National Health and Nutrition Examination Surveys. The results demonstrate the model’s ability to identify relevant phenotypes within each age subgroup and maintain comparability across them. The model can be applied to a range of other unsupervised clustering tasks related to health, sociodemographic or environmental data that have complex patterns with similarities and differences across subgroups.