Survival Regression Modeling Strategies in CVD Prediction

Area of Science:

• Biostatistics
• Epidemiology
• Health Informatics

Background:

• Prediction models are crucial for disease prevention, but assessing the added value of novel predictors remains challenging.
• Existing statistical methods for evaluating predictive performance are limited, hindering the adoption of new biomarkers.
• The difference in predictive power between models with and without novel predictors indicates added value.

Purpose of the Study:

• To develop user-friendly Stata commands for implementing novel statistical procedures in model assessment.
• To facilitate the calculation of key statistical indices for evaluating added predictive value.
• To address the lack of accessible software for assessing novel prediction models.

Main Methods:

• Developed Stata commands for Nam-D'Agostino X² goodness of fit test.
• Implemented cut point-free and cut point-based net reclassification improvement (NRI) and integrated discriminatory improvement (IDI) indices.
• Applied commands to real-world data (Tehran lipid and glucose study) to assess predictors for cardiovascular disease (CVD) risk.

Main Results:

• The Stata command 'adpredsurv' was created for survival models.
• The package enables calculation of goodness of fit, NRI, and IDI for survival data.
• Demonstrated utility in assessing the impact of family history, waist circumference, and glucose on CVD risk prediction.

Conclusions:

• The 'adpredsurv' Stata package provides tools for calculating advanced statistical indices.
• Encourages the use of novel methods for evaluating the predictive capacity of new biomarkers.
• Aims to improve the assessment of prediction models in clinical and epidemiological research.