Artificial intelligence in contrast-induced nephropathy after coronary interventions: A meta-analysis

Area of Science:

• Cardiology
• Nephrology
• Artificial Intelligence

Background:

• Contrast-induced nephropathy (CIN) is a significant complication post-coronary interventions, increasing patient morbidity and healthcare expenses.
• Machine learning (ML) offers novel methods for predicting CIN by analyzing complex clinical data, potentially enhancing risk assessment and patient outcomes.

Purpose of the Study:

• To evaluate the predictive performance of various machine learning models for contrast-induced nephropathy (CIN).
• To identify the best-performing ML models for CIN risk stratification in patients undergoing coronary interventions.

Main Methods:

• A meta-analysis of seventeen studies involving 21,69,263 patients.
• Synthesized predictive accuracy of ML models using pooled area under the curve (AUC) estimates and heterogeneity metrics.

Main Results:

• The pooled incidence of CIN was 11%. Overall ML models achieved a pooled AUC of 0.74.
• Random Forest (RF) model showed the highest performance (AUC=0.86), followed by Gradient Boosting Machines (GBM) and Extreme Gradient Boosting (XGBoost) (AUC=0.79).
• Ensemble models performed best in test datasets (AUC=0.80), while RF excelled in training datasets (AUC=0.98). The European Society of Urogenital Radiology (ESUR) criteria yielded the highest predictive performance (AUC=0.77).

Conclusions:

• Random Forest (RF), Ensemble models, and XGBoost are the most effective ML models for predicting CIN.
• Ensemble models demonstrated superior performance in test datasets, while RF was consistently superior in training datasets.
• The European Society of Urogenital Radiology (ESUR) definition aids in CIN risk stratification due to its high predictive accuracy.