Artificial intelligence for the prediction of all-cause mortality and readmission in heart failure: a meta-analysis of twenty studies

Background: Accurate risk prediction of heart failure patients is essential for identifying high risk patients, developing targeted treatment strategies and prognostication. Existing traditional risk scores offer modest discriminative value, and rely on rigid predictor variables. A form of artificial intelligence, machine learning (ML) models provide alternate risk stratification that may improve predictive accuracy. This systematic review and meta-analysis compared machine learning models with traditional risk scores for predicting all-cause mortality and readmission in patients with heart failure. Methods: PubMed, EMBASE, Web of Science and Cochrane databases were searched until 1st May, 2024 for studies comparing ML models with traditional statistical methods for prediction of all-cause mortality and hospital re-admission following an index admission with CHF. The primary outcome was comparative discrimination measured by C-statistics with 95% confidence intervals between ML models and traditional methods in estimating the risk of all-cause mortality and hospital readmission at 30 days. Results: Twenty observational studies were included (558,233 patients). The summary C-statistic of the top-performing ML models for all-cause mortality was 0.76 (95% CI, 0.72-0.80), compared to traditional risk scores 0.71 (95% CI, 0.68-0.74). The difference in C-statistic between ML models and traditional methods was 0.05 (95% CI 0.04-0.06, p<0.05). Of all included studies, 6 models were externally validated. Calibration was inconsistently reported. Conclusion: ML models demonstrated superior discrimination of 30-day all-cause mortality and hospital readmission for heart failure patients when compared to traditional risk scores. Before integrating into clinical practice, further research is required to overcome methodological and validation limitations. ; No Full Text