Machine Learning Model for Predicting Risk Factors of Prolonged Length of Hospital Stay in Patients with Aortic Dissection: a Retrospective Clinical Study

Figures

• 

  Artificial intelligence (AI)-based machine learning algorithms were used to predict risk factors for prolonged length of hospital stay (LOS) in patients with aortic dissection

• 

  Fig. 1. The flowchart of this study. The flowchart of this study consists of data collection and model development. EMR, electronic medical record; KNN, k-nearest neighbors; GNB, gaussian naive bayes; SVM, support vector machine; CNB, complement naive bayes; MLP, multi-layer perceptron neural network

• 

  Fig. 2. The Boruta algorithm for feature selection. The features were selected for developing models when the importance of features exceeds the ShadowMax value

• 

  Fig. 3. The training and validation ROC curves of machine learning-based models. (A) The comparison of machine learning-based models in the area under the receiver operating characteristic curve in the training set. (B) The comparison of machine learning-based models in the area under the receiver operating characteristic curve in the validation set. KNN, k-nearest neighbors; GNB, gaussian naive bayes; SVM, support vector machine; CNB, complement naive bayes; MLP, multi-layer perceptron neural network

• 

  Fig. 4. Calibration plots of machine learning-based models. The comparison of machine learning-based models in calibration curve fitting is shown by Brier score and 95% confidence interval (CI). KNN, k-nearest neighbors; GNB, gaussian naive bayes; SVM, support vector machine; CNB, complement naive bayes; MLP, multi-layer perceptron neural network

• 

  Fig. 5. Decision curve analysis. The decision curve analysis (DCA) for the XGBoost model predicting prolonged length of stay in patients with aortic dissection. The DCA curve shows that the miscalibration offsets improved discrimination: the XGBoost model has a higher net benefit compared to the clinical default strategies of “treat all” or “treat none”, over the entire range of reasonable threshold probabilities

• 

  Fig. 6. The XGBoost model’s interpretation based on the SHAP. The positive or negative effects of top twenty variables on prolonged length of stay in patients with aortic dissection (AD)

• 

  Fig. 7. Individual analysis of SHAP. The SHAP force plot for explaining results of individual’s prediction in the validation cohort. SHAP, SHapley Additive explanation