Random Survival Forest With Multiple Imputation Analysis for Case-Cohort and Generalized Case-Cohort Studies

Area of Science:

• Epidemiology
• Biostatistics
• Machine Learning

Background:

• Case-cohort and generalized case-cohort designs offer efficient alternatives to full cohort studies in epidemiology.
• Existing research primarily addresses estimation and inference in semiparametric survival models.
• Nonparametric survival prediction methods are under-explored for these efficient study designs.

Purpose of the Study:

• To introduce novel methods for nonparametric survival prediction in case-cohort and generalized case-cohort studies.
• To evaluate the performance of proposed random survival forest approaches with different imputation techniques.
• To enhance the utility of case-cohort designs for predictive modeling in epidemiology.

Main Methods:

• Proposed random survival forest with multiple imputation by chained equation (RSF-MICE).
• Proposed random survival forest with substantive model compatible imputation (RSF-SMCI).
• Conducted simulation studies to assess finite-sample performance.

Main Results:

• Both RSF-MICE and RSF-SMCI demonstrated superior performance in simulations across various scenarios.
• The proposed methods provide effective survival predictions within case-cohort frameworks.
• Successfully applied the approach to predict incident diabetes in a real-world study.

Conclusions:

• RSF-MICE and RSF-SMCI are effective and efficient for survival prediction in case-cohort studies.
• These methods advance nonparametric survival analysis for epidemiological research.
• The developed approach has practical applications in predicting disease incidence.