Key result
An ensemble machine learning model incorporating 15 clinical features outperformed the Mehran score in predicting contrast-associated acute kidney injury, achieving an AUC of 0.828 in internal validation.
Why the study?
Patients with ACS frequently require coronary angiography and may develop contrast-associated acute kidney injury, but effective preventive measures are lacking, making efficient interpretable predictive models crucial.
Do machine learning models improve the prediction of contrast-associated acute kidney injury compared to the Mehran score in patients with acute coronary syndrome undergoing angiography?
Observational (n=1,013)
Yes
Do machine learning models improve the prediction of contrast-associated acute kidney injury compared to the Mehran score in patients with acute coronary syndrome undergoing angiography?
Effect estimate: AUC 0.828 (95% CI 0.779-0.876)
A machine learning ensemble model using 15 clinical features outperformed the traditional Mehran score for predicting contrast-associated acute kidney injury in ACS patients.
May support CA-AKI risk stratification in ACS; hypothesis-generating and requires prospective validation.
Abstract Background Patients with acute coronary syndrome (ACS) often need coronary angiography (CAG). After CAG, patients with ACS may develop contrast-associated acute kidney injury (CA-AKI). However, effective preventive measures are lacking in treatment for CA-AKI. Therefore, an efficient interpretable predictive model of CA-AKI is crucial. Methods We enrolled 1013 ACS patients who received percutaneous coronary intervention or coronary angiography in Sir Run Run Hospital, Nanjing Medical University, and Nanjing First Hospital from September 2020 to December 2021. To screen features, the sliding windows sequential forward feature selection technique (SWSFS) was used. The model was built using five machine learning (ML) algorithms: logical regression (LR), random forest (RF), support vector machines (SVM), extreme gradient boosting (XGBT), and ensemble model (ENS). We evaluated predictive performance by comparing the model with Mehran score. The model features were explained through shapley additive explanations (SHAP) and a web-based calculator was built. Results CA-AKI occurred in 215 patients (21.27%). In the training set, SWSFS identified 15 variables. The top 5 variables included diuretics, creatine phosphokinase MB isoenzyme, unstable angina, lactate dehydrogenase, and Triglycerides × Total Cholesterol × Body Weight Index (TCBI). Overall, ML models outperformed Mehran score. In the internal and external validation sets, the ENS model obtained the highest AUC of 0.828 (95%CI: 0.779 ~ 0.876) and 0.811 (95%CI: 0.750 ~ 0.867). SHAP explained the 15 selected features' importance and contribution. We also built a web-based calculator for clinical use. Conclusions Based on machine learning, 15 clinically accessible features were screened. The established model and the web-based calculator had the potential for real-time risk assessment of CA-AKI in clinical practice.
No takes yet. Share an insight, caveat, or question.
Ma et al. (2023) conducted an observational in Acute coronary syndrome (ACS) (n=1,013). Ensemble machine learning model (ENS) vs. Mehran score was evaluated on Prediction of contrast-associated acute kidney injury (CA-AKI) (AUC 0.828, 95% CI 0.779-0.876). An ensemble machine learning model incorporating 15 clinical features outperformed the Mehran score in predicting contrast-associated acute kidney injury, achieving an AUC of 0.828 in internal validation.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: