Dynamic predictions with time-dependent covariates in survival analysis using joint modeling and landmarking

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Medical Informatics

Background:

• Accurate patient prognosis is crucial for effective clinical practice.
• Physicians utilize various tests and biomarkers to monitor disease progression.
• Optimizing the use of longitudinal data for dynamic survival prediction is essential.

Purpose of the Study:

• To present and compare two statistical techniques for dynamically updated survival probability estimates.
• To evaluate the functional form linking longitudinal and event time processes.
• To assess measures of discrimination and calibration in dynamic prediction.

Main Methods:

• Landmark analysis for dynamic survival prediction.
• Joint models for longitudinal and time-to-event data.
• Comparison of statistical techniques focusing on functional forms and prediction accuracy.

Main Results:

• Both landmark analysis and joint models offer dynamically updated survival probabilities.
• The choice of functional form significantly impacts prediction accuracy.
• Discrimination and calibration measures are key for evaluating dynamic prediction models.

Conclusions:

• Landmark analysis and joint models provide valuable tools for dynamic prognostic prediction.
• Careful consideration of the longitudinal-event time process linkage is vital.
• Robust evaluation using discrimination and calibration metrics ensures reliable clinical decision support.