PrevCardioOncAI: Machine Learning Algorithms for Predicting Cardiovascular Disease in Cancer Survivors.

Area of Science:

• Cardio-oncology
• Machine Learning in Medicine
• Biostatistics

Background:

• Cardiovascular disease (CVD) is a primary cause of mortality in cancer survivors.
• Accurate prediction of CVD risk in cancer survivors is a significant clinical challenge.
• Machine learning (ML) offers a potential solution for objective and precise CVD risk assessment.

Purpose of the Study:

• To evaluate the performance of various ML algorithms and regularized logistic regression for CVD risk prediction in cancer survivors.
• To compare the predictive accuracy of established ML techniques with newer, more complex algorithms.
• To assess the utility of these models for identifying survivors at high risk of developing CVD.

Main Methods:

• A multicenter study involving 3835 cancer survivors with 89 features collected over 20 years.
• Models were trained on random and time-split samples and validated on a separate cohort of 329 patients.
• Performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Main Results:

• Regularized logistic regression achieved AUCs of 0.845 (heart failure), 0.792 (coronary artery disease), and 0.806 (composite CVD).
• Advanced ML models like Bayesian additive regression trees and random forests showed comparable performance for heart failure prediction.
• Regularized logistic regression also effectively predicted de novo composite CVD (AUC 0.826).

Conclusions:

• Both regularized logistic regression and advanced ML models exhibit similar, institutionally transferable predictive capabilities for CVD in cancer survivors.
• These computational tools can aid in risk stratification and the development of targeted prevention strategies.
• Longitudinal data analysis is key for advancing cardio-oncology through predictive modeling.