Prediction of survival after pediatric cardiac arrest using heart rate variability and machine learning.

Abstract

Early prognostication of the outcome in pediatric cardiac arrest (CA) patients is crucial for clinical decision-making. Heart rate variability (HRV) has shown potential in predicting outcomes after CA in adult patients. This study investigates whether HRV can be used to predict survival outcomes after pediatric CA using machine learning techniques. This retrospective study included children with CA, who achieved return of spontaneous circulation (ROSC), and were admitted to the pediatric intensive care unit (PICU) of a tertiary hospital between 2012 and 2021. A 5-min electrocardiogram (ECG) segment acquired at 24 h after CA was used to calculate HRV parameters (time-, frequency-, and non-linear domains). These parameters were used to train a random forest model. The primary outcome was 12-month survival or death. Model performance was evaluated using receiver-operating characteristics (ROC) analysis and predictive values. Feature importance was assessed using Shapley values. A total of 76 patients (male: 63.2%, median age: 2.5 [IQR: 0.4-8.0] years) were divided into survival (34) or death (42) groups based on 12-month outcomes. The machine learning model achieved an accuracy of 77.6% and a positive predictive value of 0.879 for mortality prediction. The most influential features for model predictions were the frequency-domain parameters total power and very-low frequency (VLF) power, with lower values associated with an increased probability of death. Analysis of HRV at 24 h after ROSC may serve as a strong predictor of 12-month survival after pediatric CA.