Joint Analysis of Lung Cancer Longitudinal and Survival Outcomes with a Heteroscedastic Covariance

Music City Center 

Analyzing longitudinal data and survival together is crucial in clinical and epidemiological research for understanding disease dynamics and improving individualized risk prediction. Joint modeling combines mixed-effects models for longitudinal outcomes with survival models, capturing the correlation between these two data types. However, existing joint models often struggle with complex covariance structures, including issues such as ensuring positive definiteness and handling high-dimensional covariance matrices. This paper introduces a novel joint modeling approach using the hypersphere decomposition within a modified Cholesky decomposition (HDMCD) framework. HDMCD effectively manages the covariance matrix complexities by decomposing it into generalized autoregressive parameters, capturing serial correlation, and innovation variances for enhanced prediction accuracy, while ensuring positive definiteness. This new framework improves flexibility and robustness in jointly modeling longitudinal and survival data.

Hypersphere decomposition

Joint model

Longitudinal

Survival