Using Machine Learning to Predict MACEs Risk in Patients with Premature Myocardial Infarction

Fig. 1.. Flowchart. PMI, premature myocardial infarction; RSF, random survival forest; LASSO, least absolute shrinkage and selection operator; DCA, decision curve analysis; XGBoost, extreme gradient boosting; RCS, restricted cubic spline; COX, COX proportional hazards model; K-M, Kaplan-Meier; C-index, concordance index.

Fig. 2.. The process of variable selection. (A) LASSO coefficient path diagram. The horizontal axis shows the log-transformed regularization parameter (Log(λ)), and the vertical axis represents variable coefficients. As λ increases, stronger penalties shrink more coefficients to zero, highlighting the most relevant variables. (B) LASSO cross-validation plot. The horizontal axis shows Log(λ), and the vertical axis represents the partial likelihood deviance. Red dots represent deviance values from 10-fold cross-validation, with error bars indicating standard error. The left dashed line marks λ_min (minimum deviance), and the right dashed line marks λ_1se (a simpler model within one standard error of λ_min). (C) Ranking the importance of variables in the random survival forest (RSF) model. (D) Venn diagram of the intersection of variables screened by RSF and LASSO. LASSO, least absolute shrinkage and selection operator; ApoB, apolipoprotein B.

Fig. 3.. RCS plot of the final screening variables. (A) BMI; (B) UA; (C) LVEF; (D) HbA1c; (E) FFA; (F) BNP; (G) ApoB. The relationship between the independent variable and MACEs is shown, as well as the HR and its confidence interval (the red-shaded part in the figure, when the red-shaded part crosses 1, it means that the HR value is meaningless).

Fig. 4.. Performance evaluation of four models. (A) Receiver operating characteristic (ROC) curves of different models at 12, 24, and 36 months. From left to right are COX regression, Random Survival Forest, XGBoost, and DeepSurv models. (B) Decision curve analysis of four models at 12 months, 24 months and 36 months.

Fig. 5.. Survival analysis of high-risk and low-risk populations in the RSF model. Kaplan-Meier curves for the training set (left) and testing sets (right), and the number of people who did not develop MACEs over time.

Fig. 6.. Feature importance ranking, SurvSHAP(t) prediction interpretation analysis and risk calculator. (A) Feature importance ranking of RSF model and forest plot. (B) The figure shows the temporal changes in the SurvSHAP(t) values of the key variables in the RSF model. The x-axis represents time (in months), while the y-axis represents the magnitude of the SurvSHAP(t) value, which quantifies the contribution of each variable to the prediction of survival at a given time point. The figure shows how the predicted survival probability of a single sample at different time points is affected by various features. (C) Risk calculator for MACEs in patients with PMI. The left page inputs the value of the variable, and the right page outputs the risk score and level.