Re-calibrating pure risk integrating individual data from two-phase studies with external summary statistics.

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Decision-Making

Background:

• Accurate risk assessment is crucial for clinical decisions, but existing models often misestimate patient risk in new populations.
• Developing new models for each target cohort is often infeasible due to high costs.

Purpose of the Study:

• To develop a robust method for recalibrating existing risk prediction models to new target cohorts.
• To address limitations of current recalibration techniques, particularly bias arising from differing covariate distributions and two-phase sampling designs.

Main Methods:

• A weighted estimating-equation approach was developed, specifically accounting for two-phase sampling designs.
• This was combined with a weighted empirical likelihood method utilizing summary data on disease incidence and covariates from the target cohort.
• A resampling-based inference procedure was employed for statistical analysis.

Main Results:

• The proposed recalibration method demonstrated nearly unbiased risk estimates across diverse scenarios in simulation studies.
• The method effectively leverages summary information from the target population, even when covariate distributions differ.
• Application to a colorectal cancer study confirmed the generation of a well-calibrated model in the target cohort.

Conclusions:

• The developed method offers a powerful and flexible approach to recalibrate risk prediction models for improved accuracy in target populations.
• It overcomes key limitations of existing methods, providing reliable risk estimates without the need for costly new model development.
• This approach enhances the clinical utility of existing risk prediction tools across different cohorts.