A Bayesian Joint Model of Multiple Nonlinear Longitudinal and Competing Risks Outcomes for Dynamic Prediction in...

Area of Science:

• Oncology
• Biostatistics
• Health Informatics

Background:

• Predicting clinical events in Multiple Myeloma (MM) is complex due to disease progression influenced by biomarkers like M-protein.
• Understanding the impact of longitudinal biomarker patterns on therapy transitions and survival is crucial for patient management.

Purpose of the Study:

• To develop a Bayesian joint model for analyzing multiple longitudinal biomarkers and competing risks of death and therapy transitions in MM.
• To evaluate simultaneous and sequential estimation approaches for the proposed joint model.
• To validate the model using real-world data for dynamic prediction of clinical events.

Main Methods:

• Proposed a Bayesian joint model integrating longitudinal biomarkers (e.g., M-protein) with competing risks (death, next line of therapy).
• Explored two estimation strategies: simultaneous and a corrected two-stage sequential approach.
• Applied the model to a retrospective US MM patient cohort (2015-2022), using training and testing datasets for validation.

Main Results:

• The study successfully developed and validated a joint modeling framework for MM.
• Both simultaneous and sequential estimation methods were assessed for their efficacy and computational efficiency.
• The model demonstrated capability for dynamic prediction of clinical events based on longitudinal data.

Conclusions:

• The proposed Bayesian joint model effectively integrates longitudinal biomarkers and competing risks for improved clinical event prediction in MM.
• The evaluated estimation methods offer flexible approaches for model implementation.
• This framework supports personalized treatment strategies and outcome prediction in MM patients.