A Dynamic Prediction Model Supporting Individual Life Expectancy Prediction Based on Longitudinal Time-Dependent...

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Chronic Disease Management

Background:

• Traditional survival predictions (e.g., hazard ratio) are abstract for clinicians and patients.
• Patients desire intuitive survival time estimates, especially at follow-up visits.
• Longitudinal time-dependent covariates complicate accurate survival prediction in chronic diseases.

Purpose of the Study:

• To develop a dynamic restricted mean survival time (RMST) prediction model.
• To incorporate longitudinal time-dependent covariates into survival time predictions.
• To provide more intuitive and accurate survival time estimates for clinical decision-making.

Main Methods:

• Proposed a dynamic RMST prediction model using joint modeling techniques.
• Accounted for longitudinal time-dependent covariates.
• Validated the model through Monte Carlo cross-validation and application to a primary biliary cirrhosis (PBC) cohort.

Main Results:

• The dynamic RMST model demonstrated superior performance compared to the static RMST model in simulations.
• The model accurately described the trajectory of longitudinal time-dependent covariates.
• In the PBC cohort, the dynamic RMST model achieved an average C-index of 0.81, outperforming static RMST regression.

Conclusions:

• The dynamic RMST prediction model offers enhanced predictive accuracy for survival times.
• The model provides a more scientific basis for clinical decisions in chronic disease management.
• It enables dynamic prediction of average survival times at various patient follow-up points.