Personalized Predictions of Therapeutic Hypothermia Outcomes in Cardiac Arrest Patients with Shockable Rhythms Using Explainable Machine Learning

Figures

• 

  Figure 1. The analytical flowchart employed in the current study. Favorable neurological outcomes were indicated by a CPC score of 1 to 2 points, while unfavorable outcomes were defined as a CPC score of 3 to 5 points. CPC, cerebral performance category.

• 

  Figure 2. Predictive performance of the machine learning models. Receiver operating characteristic (ROC) curves and area under the curve (AUC) constructed through 5-fold cross-validation for the (A) LR, (B) random forest, (C) SVM, (D) XGBoost, and (E) ANN. Each point on the mean ROC curve is plotted based on the average true positive rate and false positive rate calculated from each of the five validation folds. Abbreviations: ANN, artificial neural network; AUC, area under the curve; LR, logistic regression; ROC, receiver operating characteristic; SVM, support vector machine; XGBoost, eXtreme gradient boosting.

• 

  Figure 3. Bar plot and the beeswarm summary plot of the SHAP values of the ANN model. (A) The importance of each feature is ranked by the mean absolute SHAP value. (B) Effect of each variable on the predictive performance of the model. Different colors indicate the magnitude of variable values, with red indicating higher values and blue indicating lower values. The X-axis presents the effect of different variables on the SHAP values. Abbreviations: ANN, artificial neural network; BT, body temperature; CAD, coronary artery disease; COPD: chronic obstructive pulmonary disease; CPC, cerebral performance category; CPR: cardiopulmonary resuscitation; DBP: diastolic blood pressure; DM: diabetes mellitus; ESRD, end-stage renal disease; HR, heart rate; ROSC, return of spontaneous circulation; SBP: systolic blood pressure; SHAP, SHapley Additive exPlanations.

• 

  Figure 4. Associations between the top five crucial features and discharge CPC scores at discharge. Continuous variables include the (A) dosage of epinephrine, (B) BT, and (C) DBP at ROSC. Categorical variables include (D) diabetes mellitus and (E) witnessed collapse. BT, body temperature; CPC, cerebral performance category; DBP, diastolic blood pressure; DM, diabetes mellitus; ROSC, return of spontaneous circulation.

• 

  Figure 5. Exploring patient cases: real-life application of explainable machine learning model. We retrospectively employed the SHAP waterfall plot to illustrate the application of the machine learning model in clinical settings for two patients who underwent therapeutic hypothermia. (A) A patient with OHCA who had a favorable outcome; (B) A patient with IHCA who had a fatal outcome after therapeutic hypothermia. BT, body temperature; CAD, coronary artery disease; COPD: chronic obstructive pulmonary disease; CPR: cardiopulmonary resuscitation; DBP, diastolic blood pressure; DM, diabetes mellitus; HR, heart rate; HTN, hypertension; IHCA, in-hospital cardiac arrest; OHCA, out-of-hospital cardiac arrest; ROSC, return of spontaneous circulation; SHAP, Shapley Additive exPlanations.