Predicting recurrent cardiac arrest within one year after surviving in-hospital cardiac arrest using a machine learning model.

Abstract

Patients discharged alive after in-hospital cardiac arrest (IHCA) have an increased mortality up to a year after hospital discharge. Improving our ability to identify patients at high risk of death one year after IHCA may improve survival through secondary prevention. We aimed to evaluate the possibility to predict recurrent cardiac arrest or death among in-hospital cardiac arrest survivors with support from a machine learning model. All patients (>18 year) discharged alive after IHCA in the Swedish registry for cardiopulmonary resuscitation (SRCR) from 2010 to 2021 were included. Potential predictors included in the model were data related to the index IHCA, comorbidities, socioeconomic data and prescription medication obtained from national Swedish registries. Extreme gradient boosting (XGBoost) model was trained to predict the outcome. Receiver operating characteristics and area under the curve (ROC-AUC) was used to evaluate model performance. Of the 7302 included patients, 22% had developed the outcome. The best performing model with 1241 variables had ROC-AUC 0.72 (95% CI 0.70-0.75). Features of greatest importance were highest serum creatinine measured before IHCA, age and unknown civil status. Comorbidity and cardiac arrest circumstances were important classes of features for this model. In this dataset, an XGBoost model could predict recurrent cardiac arrest or death within one year after IHCA with a modest performance and good accuracy. If these results can be validated, this model could potentially be used clinically to assess the risk of another cardiac arrest in survivors of IHCA.