UC San Diego

Patients with diabetic kidney disease (DKD) are at high risk for hospitalization, morbidity, and mortality. Early detection of patients with kidney function decline can lead to effective intervention and management of high risk of developing DKD. The human metabolome is a powerful tool for informing the physiological and pathological effects of chronic diseases and could offer direct insights into biochemical pathways potentially linked to kidney dysfunction. Furthermore, functional principal components analysis (FPCA) is a novel approach for modeling and studying the variation of kidney function trajectories for subgroups of diabetic patients, while accounting for complexity in curve estimation. Here, we applied, validated, and extended rigorous statistical approaches that utilize metabolomic markers and functional data methods for uncovering the characteristics of and predicting DKD progression.In Chapter 1, we give an overview of the background and rationale for our distinct research aims. In Chapter 2, we elucidate the choice between fitting a linear mixed model, with serial estimated glomerular filtration rate (eGFR) outcomes, and two-stage methods, with patient-specific eGFR slopes as outcomes, for modeling DKD progression, with metabolites as predictors. Notably, two-stage models offer a suitable modeling alterative to DKD researchers who can readily implement individual eGFR slopes in standard regression models. In Chapter 3, we apply the top-scoring pair (TSP) algorithm to derive simple, parameter-free decision rules (i.e., pair of metabolites) for binary DKD stage classification. As a methodological contribution, we extended the TSP approach to allow adjustment for clinical variables. In Chapter 4, we implement the FPCA approach, which accounts for nonlinear trajectories via nonparametric smoothing while overcoming sparsity and irregularly spaced data. We examined the longitudinal patterns of kidney function trajectories within clinically defined albuminuria-specific groups and expand the FPCA inferential framework for considering whether separate group-level models to prospectively predict group-specific outcome trajectories are needed. Our findings provided insights into modeling choices for DKD progression, markers for renal dysfunction adjusted for clinical variables, dominant modes of eGFR variation, and varying eGFR patterns between albuminuria groups, which can potentially inform therapeutic targets for personalized DKD treatments.