Abstract / Summary
Contrast-induced nephropathy (CIN) is a major complication following coronary interventions, contributing to increased morbidity and healthcare costs. Machine learning (ML) models provide innovative approaches for predicting CIN by integrating complex clinical variables, potentially improving risk stratification and patient outcomes. This meta-analysis evaluates the predictive performance of ML models for CIN, focusing on the best-performing models. Seventeen studies encompassing 21,69,263 patients were analyzed. The predictive accuracy of ML models was synthesized using pooled area under the curve (AUC) estimates and heterogeneity metrics. The pooled incidence of CIN was 11% (95% CI: 9-13%). Overall, ML models achieved a pooled AUC of 0.74 (95% CI: 0.72-0.75). Random forest (RF) model demonstrated the highest performance with an AUC of 0.86 (95% CI: 0.85-0.87), followed by gradient boosting machines (GBM) and Extreme Gradient Boosting (XGBoost), both achieving an AUC of 0.79. In training datasets, RF and XGBoost achieved the highest AUCs of 0.98 (95% CI: 0.97-0.99), with GBM following at 0.88 (95% CI: 0.85-0.90). In test datasets, Ensemble models achieved the best performance with an AUC of 0.80 (95% CI: 0.66-0.94), followed by RF and XGBoost with AUCs of 0.75. External validation results showed an overall pooled AUC of 0.77 (95% CI: 0.71-0.84), indicating strong generalizability of the models. Among CIN definitions, the European Society of Urogenital Radiology (ESUR) criteria yielded the best predictive performance, with an AUC of 0.77 (95% CI: 0.72-0.82). RF, Ensemble models, and XGBoost emerged as the most effective ML models for predicting CIN, with RF showing consistent superiority in training datasets and Ensemble models excelling in test datasets. The pooled CIN incidence emphasizes the clinical burden, and the ESUR definition provided the highest predictive accuracy, supporting its utility in CIN risk stratification.
Primary Source
Medicine
Ask Prognia AI
Have questions about this meta-analysis?
Prognia AI can search this source alongside 35M+ PubMed papers and current ESC, AHA, NICE, and ADA guidelines to give you a fully cited clinical answer.