Predictability of Mortality in Patients With Myocardial Injury After Noncardiac Surgery Based on Perioperative Factors via Machine Learning: Retrospective Study

Figures

• 

  Figure 1. A flowchart of our retrospective study design. peri-op hs cTn I: perioperative high-sensitivity cTn-I; peri-op hs cTn T: perioperative high-sensitivity cTn-T; Pts: patients; SMN-TINCO: Samsung Medical Center Troponin in Noncardiac Operation.

• 

  Figure 2. Performance comparison of each 30-day mortality prediction model with the range of (A) AUROC and (B) F1 score. AUROC: area under the receiver operating characteristic; CART: classification and regression trees; GBM: generalized boosted regression model; GLMNET: lasso/ridge/elastic net; kNN: k-nearest neighbors; LDA: linear discriminant analysis; RF: random forests; SVM: support vector machines; XGB: extreme gradient boosting.

• 

  Figure 3. SHAP summary plot of 30-day mortality prediction extreme gradient boosting model. According to the SHAP values of each feature, antiplatelet prescription at discharge (ie, purple dots) was associated with a lower probability of death (ie, the left side of the vertical dotted line). Higher C-reactive protein levels at discharge (ie, purple dots) were associated with a higher probability of death (ie, the right side of the vertical dotted line). Insulin prescription at discharge (ie, purple dots) was associated with a higher probability of death (ie, the right side of the vertical dotted line). CABG: coronary artery bypass graft; COPD: chronic obstructive pulmonary disease; ESA: European Society of Anaesthesiology; ESC: European Society of Cardiology; SHAP: Shapley Additive Explanations.

• 

  Figure 4. Internet app for predicting 30-day mortality of patients with MINS. CCB: calcium channel blockers; MINS: myocardial injury after noncardiac surgery.