Risk Projection for Time-to-event Outcome Leveraging Summary Statistics With Source Individual-level Data

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Prediction Modeling

Background:

• Accurate prediction of chronic disease risk is crucial in clinical practice.
• Existing prediction models often perform poorly when applied to new populations due to differing baseline risks and patient characteristics.
• This necessitates methods for recalibrating models for external cohorts.

Purpose of the Study:

• To develop a novel statistical approach for recalibrating existing chronic disease risk prediction models to new target populations.
• To update the baseline risk within the prediction model using information from the target cohort while assuming consistent relative risks of predictors.
• To provide a robust and efficient method for improving the accuracy of risk predictions in diverse clinical settings.

Main Methods:

• A weighted estimating equation approach is proposed to recalibrate prediction models.
• The method incorporates survival probabilities for the disease of interest and competing events.
• It also utilizes summary information of risk factors from the target population.

Main Results:

• The proposed estimators are shown to be consistent and asymptotically normal.
• Extensive simulations demonstrate robustness to differences in risk factor distributions between source and target populations.
• The method gains efficiency when risk factor distributions are similar, provided precise target population data.

Conclusions:

• The novel weighted estimating equation approach effectively recalibrates chronic disease risk prediction models for new populations.
• The method is robust and can improve prediction accuracy by updating baseline risk with target population data.
• This approach has practical implications for applying prediction models across different cohorts, as illustrated by its application to colorectal cancer risk prediction.