INDIVIDUAL DYNAMIC PREDICTION FOR CURE AND SURVIVAL BASED ON LONGITUDINAL BIOMARKERS

Area of Science:

• Biostatistics
• Medical Informatics
• Clinical Research

Background:

• Accurate patient prognosis prediction is crucial for optimizing personalized treatment and extending survival.
• Longitudinal biomarker data are essential for prognosis, but predicting cure remains challenging.
• Existing models often lack flexibility in capturing complex survival patterns.

Purpose of the Study:

• To develop and validate a comprehensive joint model and a landmark cure model incorporating potentially cured patients.
• To provide formulas for predicting individual cure and survival probabilities using biomarker history.
• To enhance predictive performance compared to standard cure models.

Main Methods:

• Proposed a joint model for longitudinal and survival data and a landmark cure model.
• Utilized flexible hazard functions beyond proportional hazards.
• Incorporated proportions of potentially cured patients.
• Derived formulas for individual cure and survival probability predictions.

Main Results:

• Simulations demonstrated superior predictive performance of the proposed models.
• Improvements were measured using time-dependent AUC, Brier score, and integrated Brier score.
• The models were successfully applied to chronic myeloid leukemia patient data.

Conclusions:

• The proposed comprehensive and flexible cure models significantly outperform standard models.
• These models offer enhanced capabilities for predicting patient cure and survival.
• The approach facilitates personalized treatment strategies and improved patient outcomes.