Machine learning-based ACE risk score predicted adverse cardiovascular events in ICI-treated cancer patients with AUC 0.70 on validation and is available online.
Does a machine learning-based risk score accurately predict adverse cardiovascular events in cancer patients treated with immune checkpoint inhibitors?
A novel machine learning-based risk score using 15 clinical and imaging features achieved an AUC of 0.70 for predicting adverse cardiovascular events in cancer patients receiving immune checkpoint inhibitors.
Absolute Event Rate: 0% vs 0%
Abstract Background Cancer patients treated with immune checkpoint inhibitors (ICIs) have an increased risk of adverse cardiovascular events (ACE) 1. Traditional cardiovascular risk scores may not adequately capture ICI-associated cardiovascular toxicities or the unique features that contribute to ACE risk in this population 1,2. Recent studies have developed cardiovascular risk scores for cancer patients, which achieved area under the receiver operating characteristic curve (AUC) values in the 0.65 to 0.85 range 2. Currently, there is no validated ACE risk algorithm designed specifically for ICI patients. Purpose Our study aims 1) to develop an interpretable, machine learning-based ACE risk score algorithm for cancer patients treated with ICI therapy, and 2) to integrate this algorithm into a clinically accessible online calculator interface. Methods We analyzed 5145 cancer patients treated with ICI therapy between 2013 and 2024 at a large academic centre. Patient variables included demographics, comorbidities, laboratory values, cancer type, ICI regimen (single vs dual), and key imaging findings from echocardiography (echo) and cardiac magnetic resonance (CMR) data. The composite ACE outcome comprised myocardial infarction, coronary artery disease (CAD), arrhythmias, heart failure (HF), valvular disease, atrioventricular block, and myocarditis. Data were partitioned into training (80%), test (10%), and holdout validation (10%) sets. An extreme gradient boosting (XGB) classifier was trained using 4-fold cross-validation on the training set, and performance was evaluated on the test set. Shapley Additive Explanation (SHAP) values were used to identify top predictive features. A multivariate logistic regression model was then fit using 15 selected features (based on SHAP ranking and clinical expertise) to form the final ACE risk score algorithm, which was subsequently validated on the holdout validation set. Results ACE occurred in 36.5% of patients in our cohort. The XGB model achieved an AUC of 0.73 on the test set (Figure 1B). The most influential SHAP features included age, body mass index, cancer type, creatinine, CAD, peripheral vascular disease, stroke, HF, hypertension, left ventricular ejection fraction, and global longitudinal strain (Figure 1A, 1D). These features, along with dual ICI status and left ventricular late gadolinium enhancement, were used to train the final ACE risk algorithm, which attained an AUC of 0.70 on the holdout validation set (Figure 1C). We integrate our ACE risk algorithm into a publicly accessible, user-friendly online calculator (Figure 2). Conclusion We present a novel, interpretable, and clinically usable ACE risk score algorithm tailored to cancer patients treated with ICI therapy, which may aid in improving risk stratification and cardiovascular monitoring in this high-risk population.Fig1.Top SHAP features & AUC curves Fig2.ACE risk calculator interface
Cross et al. (Sat,) reported a other. Machine learning-based ACE risk score predicted adverse cardiovascular events in ICI-treated cancer patients with AUC 0.70 on validation and is available online.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: