See the full schedule here.
Title: Statistical Modeling of BMI Trajectories and Mortality Risk Using a Multi-Decade Japanese Health Cohort: Applications of Proposed Functional Clustering and Survival Methods
Keywords: Longitudinal Health Cohort, Functional Data Analysis, Survival Modeling, Wasserstein-Based Time Series Modeling
Abstract: This presentation explores the statistical association between longitudinal variation in body mass index (BMI) and mortality risk using data from a large-scale Japanese health cohort initiated in the late 1950s, comprising several thousand participants followed over multiple decades. This cohort provides a rare opportunity to analyze long-term individual-level trajectories in a real-world population setting. The first part reviews the approach of Cologne et al. (2019, JAMA Network Open), where residual variation in BMI trajectories was used as a covariate in Cox proportional hazards models to assess mortality risk. Their findings highlight the importance of modeling intra-individual variability beyond average trends. The second part introduces two statistical methods recently developed by the speaker: Functional Convex Clustering and Joint Modeling with the Cox Model. These methods provide a flexible framework for clustering individual longitudinal trajectories and integrating them into time-to-event analysis. Finally, I briefly present ongoing work on distributional time series modeling in Wasserstein space via Koopman operator decomposition, outlining future directions in geometric and operator-based functional data analysis.
