Dynamic prediction of cumulative incidence functions by direct binomial regression

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Dynamic prediction models are crucial for updating patient risk assessments over time.
• Competing risks present challenges in accurately predicting event probabilities.
• Existing methods for cumulative incidence prediction need enhancement for dynamic updates.

Purpose of the Study:

• To extend dynamic prediction methods for the cumulative incidence function in competing risk settings.
• To integrate direct binomial regression with inverse probability of censoring weights and landmarking.
• To develop flexible models accommodating complex time-varying covariate effects.

Main Methods:

• Direct binomial regression with inverse probability of censoring weights.
• Incorporation of landmarking for dynamic prediction.
• Generalized estimating equations for model fitting.
• Wald tests for assessing time-varying covariate effects.

Main Results:

• The proposed method enables dynamic prediction of cumulative incidence in competing risks.
• Models demonstrate flexibility in handling complex time-varying covariate effects.
• A simulation study validated the performance of the developed method.

Conclusions:

• The combined approach of direct binomial regression and landmarking offers a powerful tool for dynamic risk prediction.
• This method provides updated and more accurate predictions as patient information evolves.
• The approach is applicable to complex clinical scenarios, such as bone marrow transplant outcomes.