Explainable Machine Learning for Predicting Major Adverse Renal and Cardiovascular Events in Acute Myocardial Infarction With Chronic Kidney Disease: Algorithm Development and Validation.
Feng W, Jiang R, et al. • JMIR medical informatics • 2026
Our study identified 6 important predictors and provided a series of prediction models (XGBoost, logistic regression, random forest, and a nomogram) to enhance risk stratification of in-hospital MARCE for patients with AMI and CKD.
Key Findings
Results
LASSO regression selected six important predictors of in-hospital MARCE in patients with AMI and CKD.
The six variables were Killip class, lactate, estimated glomerular filtration rate, use of diuretics, use of beta-blockers, and use of renin-angiotensin system inhibitors.
Variable selection was performed in the training set using least absolute shrinkage and selection operator (LASSO) regression.
Candidate features included demographics, comorbidities, laboratory results, echocardiography measurements, procedural details, and medication parameters.
Results
XGBoost, logistic regression, and random forest showed comparable discrimination in the hold-out test set, while the decision tree performed worse.
XGBoost AUC 0.882 (95% CI 0.816-0.947).
Logistic regression AUC 0.875 (95% CI 0.799-0.950).
Random forest AUC 0.870 (95% CI 0.800-0.940).
Decision tree AUC 0.814 (95% CI 0.720-0.886).
Results
A logistic regression-based nomogram maintained good discrimination in an external validation cohort.
The nomogram achieved an AUC of 0.873 (95% CI 0.714-0.926) in external validation.
The external cohort came from Wenzhou TCM Hospital of Zhejiang Chinese Medical University, separate from the discovery cohort at the First Affiliated Hospital of Wenzhou Medical University.
Results
SHAP analysis provided global and individual-level explanations of feature contributions in the XGBoost model.
Shapley additive explanations (SHAP) values illustrated both global feature importance and individual patient-level feature contributions.
This was applied to make the XGBoost model interpretable.
Conclusions
Models were evaluated on discrimination, calibration, and classification metrics in a hold-out test set and an external cohort, and are proposed to aid risk stratification and treatment choices.
Four model types were developed: XGBoost, random forest, decision tree, and logistic regression.
The outcome was in-hospital MARCE, a composite of cardiac and renal outcomes, in patients with AMI and CKD who received emergency treatment.
The authors state existing prediction tools lack accuracy and generalizability in this high-risk population.
What This Means
This study looked at patients who had a heart attack and also had chronic kidney disease, a group at high risk of serious complications affecting both the heart and kidneys during their hospital stay. The researchers built computer-based prediction models using data from one hospital and tested them on a held-out portion of that data and on patients from a second hospital.
The models used just six pieces of information: the severity of heart failure signs at presentation (Killip class), blood lactate, kidney function (estimated glomerular filtration rate), and whether patients were given diuretics, beta-blockers, or renin-angiotensin system inhibitors. Several approaches, including a machine learning method (XGBoost), a random forest, and traditional logistic regression, performed similarly well, with AUC values around 0.87-0.88 (1.0 is perfect). A simple decision tree did worse. A scoring chart (nomogram) based on logistic regression also performed well in the second hospital, with an AUC of 0.873. The team also used an explanation technique (SHAP) to show how each factor contributed to predictions.
This research suggests that a small set of routinely available measures could help identify which patients with heart attack and kidney disease are at higher risk of in-hospital heart and kidney complications. This may support risk assessment and treatment planning, though the models were developed in hospitals in one region and would need further testing before broader use.
Feng W, Jiang R, Huang W, Wang C, Wang Z, Chang J, et al.. (2026). Explainable Machine Learning for Predicting Major Adverse Renal and Cardiovascular Events in Acute Myocardial Infarction With Chronic Kidney Disease: Algorithm Development and Validation.. JMIR medical informatics. https://doi.org/10.2196/96654