Construction and Validation of a Hospital Mortality Risk Model for Advanced Elderly Patients with Heart Failure Based on Machine Learning

Figures

• 

  Figure 1. Flowchart.

• 

  Figure 2. Variable screening diagram. (A) LASSO regression solution path roadmap; (B) The LASSO regression error plot; (C) Boruta Feature screening variable selection map; (D) Boruta Details diagram of feature screening.

• 

  Figure 3. Model construction and validation. Figures (A and D) display Receiver Operating Characteristic (ROC) curves on the training and validation sets respectively, with sensitivity on the vertical axis and 1 - specificity on the horizontal axis, and the area under the curve (AUC) quantifies model performance. Figures (B and E) illustrate the relationship between predicted and observed probabilities for calibration assessment, where an ideal curve aligns with the diagonal. Figures (C and F) are Decision Curve Analysis (DCA) plots showing net benefit against threshold, comparing model - guided decision - making with “treat all” and “treat none” strategies.

• 

  Figure 4. Model evaluation. Figure 4 evaluates machine - learning models’ performance. (A and C) Are radar charts for training and validation sets respectively, showing metrics like Sensitivity, Specificity, and F1 for models such as DT, KNN, etc. (B and D) Are corresponding heatmaps, with color - coding to indicate metric values, helping to visually compare model performance across different sets.

• 

  Figure 5. XGBoost Model SHAP values. Figure 5 shows feature analysis via SHAP values. Figure (A) is a feature importance plot for features like CRP, BMI, etc. Bar length indicates importance, with CRP being most important. Figure (B) is a feature dependence plot. Points, colored by feature value, show the relationship between feature values and SHAP values, helping understand feature impact on model prediction.

• 

  Figure 6. The XGBoost Model SHAP dependency.

• 

  Figure 7. XGBoost Model single-sample force map. Figure 7 shows SHAP - based explanations for model predictions. Figure (A) displays positive SHAP values for features like BMI, CRP, etc., indicating their positive contributions to the prediction result f(x)=3.93. Figure (B) presents negative SHAP values for features such as NT - ProBNP, CRP, etc., showing their negative impacts on the prediction result f(x)=−3.81.