Personalized Predictions of Therapeutic Hypothermia Outcomes in Cardiac Arrest Patients with Shockable Rhythms Using...

Area of Science:

• Cardiology
• Neurology
• Artificial Intelligence in Medicine

Background:

• Therapeutic hypothermia (TH) is crucial for cardiac arrest patients, but outcomes vary.
• Predicting individual patient responses to TH is essential for optimizing treatment.
• Personalized prognostication can enhance clinical decision-making in post-cardiac arrest care.

Purpose of the Study:

• To develop and validate machine learning models for predicting neurological outcomes in cardiac arrest patients undergoing TH.
• To identify key predictors of neurological outcomes in this patient population.
• To demonstrate the utility of a personalized machine learning approach for TH prognostication.

Main Methods:

• A multi-center retrospective cohort study of 209 cardiac arrest patients with shockable rhythms treated with TH in Taiwan.
• Development of artificial neural network (ANN) models using pre-treatment patient characteristics.
• Interpretation of the optimal ANN model using SHapley Additive exPlanations (SHAP) for feature importance.

Main Results:

• The ANN model achieved high predictive performance (AUC = 0.9089, accuracy = 0.8330).
• Key predictors identified include epinephrine dose, diabetes status, temperature at ROSC, witnessed arrest, and diastolic blood pressure at ROSC.
• SHAP analysis enabled interpretation of model predictions, highlighting personalized prognostic factors.

Conclusions:

• Machine learning, specifically an ANN model, effectively predicts neurological outcomes for cardiac arrest patients receiving TH.
• The model provides personalized prognoses by integrating individual patient data and event details.
• This approach facilitates precise risk stratification and optimizes personalized treatment planning for TH.