A tutorial on dynamic risk prediction of a binary outcome based on a longitudinal biomarker

Area of Science:

• Biostatistics
• Clinical Prediction Modeling
• Oncology

Background:

• Dynamic risk predictions are crucial for timely identification of high-risk patients.
• Predicting future binary outcomes with repeatedly measured biomarkers remains challenging.
• Existing methods for updating predictions with longitudinal biomarker data require further investigation.

Purpose of the Study:

• To provide an overview of four distinct approaches for dynamic risk prediction of a future binary outcome using longitudinal biomarker data.
• To compare the performance of likelihood-based and Bayesian methods, including two-stage and joint models.
• To illustrate the application of these models for predicting post-molar gestational trophoblastic neoplasia (GTN) using human chorionic gonadotropin (hCG) measurements.

Main Methods:

• Four prediction approaches were evaluated: likelihood-based two-stage method (2SMLE), likelihood-based joint model (JMMLE), Bayesian two-stage method (2SB), and Bayesian joint model (JMB).
• Models were applied to weekly updated predictions of GTN using age and repeated hCG measurements.
• Model performance was assessed using discrimination and calibration measures, with internal validation via bootstrapping.

Main Results:

• All four approaches demonstrated comparable predictive and discriminative performance for GTN prediction in the applied dataset.
• A simulation study indicated that joint models outperformed two-stage methods when biomarker variability (within- and between-patient) increased.
• The study provides practical insights into obtaining and updating dynamic predictions for binary outcomes.

Conclusions:

• The evaluated methods offer viable strategies for dynamic risk prediction using longitudinal biomarker data.
• Joint models are recommended over two-stage methods in scenarios with substantial biomarker variability.
• The findings support the use of dynamic predictions for improved patient management in conditions like GTN.