Prediction of preoperative in-hospital mortality rate in patients with acute aortic dissection by machine learning: a two-centre, retrospective cohort study

Figures

• 

  Figure 1. Flow diagram of patient selection. AD, aortic dissection.

• 

  Figure 2. Comparison of areas under the receiver operating characteristic curves (AUCs) among different machine learning models. XGBoost yielded the best prediction accuracy.

• 

  Figure 3. Decision curve analysis (DCA) of the five prediction models. The net benefit curves for these prognostic models are shown. X-axis indicates the threshold probability for mortality and y-axis indicates the net benefit. SVM, support vector machine; XGBoost, eXtreme Gradient Boosting.

• 

  Figure 4. Calibration curves of the machine learning models illustrate how well the predicted probabilities match the actual in-hospital mortality in patients with acute aortic dissection. SVM, support vector machine; XGBoost, eXtreme Gradient Boosting.

• 

  Figure 5. Importance matrix plot of the eXtreme Gradient Boosting (XGBoost) model. All variables were divided into two clusters according to their importance in the development of the final predictive model. AD, aortic dissection; DBPV, diastolic blood pressure variability; HRV, heart rate variability; ODT, onset-to-door time; PAD, peripheral arterial disease; SBPV, systolic blood pressure variability.

• 

  Figure 6. SHapley Additive exPlanations (SHAP) summary plot of the eXtreme Gradient Boosting (XGBoost) model. The higher the SHAP value of a feature, the higher the mortality rate of patients with AD. A dot was created according to the SHAP value of each patient. Dots were coloured according to the values of features. Purple represented higher feature values and yellow represented lower feature values. AD, aortic dissection; COPD, chronic obstructive pulmonary disease; DBPV, diastolic blood pressure variability; HRV, heart rate variability; ODT, onset-to-door time; PAD, peripheral arterial disease; SBPV, systolic blood pressure variability.

• 

  Figure 7. SHapley Additive exPlanations (SHAP) decision plot of the eXtreme Gradient Boosting (XGBoost) model for two selected extreme cases. (A) Prediction of in-hospital mortality for case 4 who survived during hospitalisation. (B) Prediction of in-hospital mortality for case 58 who died during hospitalisation. The contribution of each variable to the predicted mortality rate was shown in the corresponding pentagon-shaped arrow. AD, aortic dissection; DBPV, diastolic blood pressure variability; HRV, heart rate variability; ODT, onset-to-door time; SBPV, systolic blood pressure variability.