Predicting Mortality in Intensive Care Unit Patients With Heart Failure Using an Interpretable Machine Learning Model: Retrospective Cohort Study

Figures

• 

  Figure 1. Flowchart of patient selection. ICD: International Classification of Diseases; ICU: intensive care unit.

• 

  Figure 2. The receiver operating characteristic curve among the four models for patients with heart failure. SVM: support vector machine.

• 

  Figure 3. Decision curve analysis of four models plotting the net benefit at different threshold probabilities. SVM: support vector machine.

• 

  Figure 4. The weights of variables importance. Avg: average; BUN: blood urea nitrogen; max: maximum; min: minimum; NIBP: noninvasive blood pressure; RDW: red blood cell distribution width; SHAP: SHapley Additive exPlanation; SpO2: O2 saturation; WBC: white blood cell.

• 

  Figure 5. The SHapley Additive exPlanation (SHAP) values. Avg: average; BUN: blood urea nitrogen; max: maximum; min: minimum; NIBP: noninvasive blood pressure; RDW: red blood cell distribution width; SpO2: O2 saturation; WBC: white blood cell.

• 

  Figure 6. SHapley Additive exPlanation (SHAP) force plot for two selected patients.