Comprehensive Summary
Mohammadi et al. conducted a PRISMA-guided systematic review of 35 studies using AI models to predict postoperative outcomes following congenital heart surgery. Most studies (published 2019-2023) focused on mortality or survival (n = 16), length of stay (n = 7), complications (n = 6), and ventilator duration (n = 4), encompassing 56 to 221,335 patients. Logistic regression (n = 16), boosting algorithms (n = 4), and random forests (n = 9) were most frequently utilized for algorithms. Reported AUCs ranged from 0.52 to 0.997, with most >0.7. AI models consistently outperformed conventional risk stratification scores such as RACHS-1 and STS-EACTS, particularly for mortality prediction. However, 60% of included studies had unclear or high risk of bias, and only 6 achieved external validation, highlighting methodological heterogeneity and limited generalizability.
Outcomes and Implications
AI-based prediction models demonstrate strong potential in improving postoperative risk assessment in congenital cardiac surgery, outperforming traditional scoring systems in both short-term and long-term outcomes. Yet widespread clinical adoption remains premature due to small, heterogeneous cohorts, limited calibration reporting, and lack of multicenter validation. Future research should prioritize transparent model development, external validation, and feature optimization to enable integration into surgical decision-making and preoperative care regimens.