Effectiveness of Artificial Intelligence Models for Cardiovascular Disease Prediction: Network Meta-Analysis

Comprehensive Summary

Cardiovascular disease is the leading cause of death worldwide. This study performed a network meta-analysis comparing machine learning and deep learning models for predicting four major outcomes: heart failure, stroke, hypertension, and diabetes. Seventeen studies involving 285,213 patients from 2016 to 2021 were included, following PRISMA and QUADAS-2 guidelines. Data was analyzed using random-effects network meta-analysis in R. Studies found that deep learning models achieved strong performance for heart failure prediction with an AUC of 0.843 (95% CI 0.840–0.845), outperforming logistic regression and random forest models. Among machine learning methods, gradient boosting machines reached 91.1% accuracy for heart failure prediction, artificial neural networks best predicted diabetes with OR 0.0905 (95% CI 0.0489–0.1673), random forest models predicted hypertension with OR 10.85 (95% CI 4.74–24.83), and support vector machines performed best for stroke with OR 25.08 (95% CI 11.48–54.78).

Outcomes and Implications

AI algorithms can assist clinicians in early identification of cardiovascular risk by analyzing patient data from routine clinical sources. Deep learning models specifically have well-posed applications to improve triage and prevent for heart failure and other major diseases. Before being used at the bedside, large-scale validation and model transparency are needed to ensure safe and equitable clinical application, while also reducing potential false positives.