Development of a machine learning algorithm model to predict intraoperative hypotension in elderly patients undergoing thoracic and abdominal surgeries

Figures

• 

  Figure 1:. The statistical flowchart of clinical data. IOH, intraoperative hypotension; LR, logistic regression; Bayes, Bayes model; KNN, K-nearest neighbors; SVM, support vector machine; NN, neural network; DT, decision tree; Xgboost, extreme gradient boosting; RF, random forest.

• 

  Figure 2:. Screening of Boruta characteristic variables. This graph illustrates the importance of feature variables as determined by the Boruta algorithm. The independent variables on the right side of ShadowMax are the feature variables filtered by Boruta. The x-axis displays the individual feature variables, while the y-axis represents their importance scores.

• 

  Figure 3:. The ROC, and PRC of the clinical data (A) represented ROC of eight models; (B) represented PRC of eight models. The figure shows the process of model selection. Figure 3 shows the performance of eight models in terms of ROC (A) and PRC (B). Among them, RF has the highest AUC-ROC (A) and AUC-PRC (B) values.

• 

  Figure 4:. ROC, PRC, calibration curve, and DCA. (A) represented ROC of the test set; (B) represented PRC of the test set; (C) represented calibration curve of the test set; (D) represented DCA of the test set.

• 

  Figure 5:. Importance ranking of characteristic variables. The figure shows the importance ranking of feature variables filtered by Boruta in the optimal model (RF). The y-axis displays each feature variable, while the x-axis represents the contribution level of its corresponding outcome variable.

• 

  Figure 6:. Partial dependency graph of characteristic variables. This graph shows the quantitative trend of each feature variable filtered by Boruta for the outcome variable in the optimal model (RF). The y-axis displays the probability of occurrence of outcome variable, while the x-axis represents the value of its corresponding feature variable.

• 

  Figure 7:. Force plots of the single-sample predictions. Force plots show the visualization process of placing a random sample into the optimal model (RF) for prediction. The graph shows the name, value, and contribution level to the outcome variable of each feature variable in this sample, as well as the model’s the predicted value of the outcome variable.