Integrating Information from Existing Risk Prediction Models with No Model Details

Area of Science:

• Biostatistics
• Statistical Modeling
• Epidemiology

Background:

• Regression models are crucial for understanding disease risk factors.
• Existing risk calculators often use limited covariates and their internal workings are sometimes unknown.
• Integrating auxiliary information can improve model efficiency.

Purpose of the Study:

• To develop a novel empirical-likelihood-based framework for integrating information from risk calculators into regression modeling.
• To enhance the efficiency of regression parameter estimation by leveraging auxiliary data from risk calculators.
• To apply the framework to prostate cancer risk prediction.

Main Methods:

• Developed a general empirical-likelihood framework.
• Proposed two methods: one using working models to extract calculator information, and another directly using calculator predictions.
• Applied the framework to analyze high-grade prostate cancer risk using the Prostate Biopsy Collaborative Group (PBCG) risk calculator.

Main Results:

• The proposed framework effectively integrates information from risk calculators.
• Theoretical and numerical results demonstrate substantial reduction in the variance of regression parameter estimation.
• The application showed improved analysis of prostate cancer risk by combining conventional factors and molecular biomarkers.

Conclusions:

• Integrating information from risk calculators into regression modeling is a viable and efficient approach.
• The empirical-likelihood framework offers a robust method for enhancing statistical models.
• This approach has significant potential for improving risk prediction in various medical applications, including prostate cancer.