A Machine Learning Prediction Model for Non-cardiogenic Out-of-hospital Cardiac Arrest with Initial Non-shockable Rhythm.

Objectives: The purpose of this study was to develop and validate a machine learning prediction model for the prognosis of non-cardiogenic out-of-hospital cardiac arrest (OHCA) with an initial non-shockable rhythm. Design: Data were obtained from a nationwide OHCA registry in Japan. Overall, 222,056 patients with OHCA and an initial non-shockable rhythm were identified from the registry in 2016 and 2017. Patients aged <18 years and OHCA caused by cardiogenic origin, cancer, and external factors were excluded. Finally, 58,854 participants were included. Methods: Patients were classified into the training dataset (n=29,304, data from 2016) and the test dataset (n=29,550, data from 2017). The training dataset was used to train and develop the machine learning model, and the test dataset was used for internal validation. We selected XGBoost as the machine learning classifier. The primary outcome was the poor prognosis defined as cerebral performance category of 3-5 at 1 month. Eleven prehospital variables were selected as outcome predictors. Results: In validation, the machine learning model predicted the primary outcome with an accuracy of 90.8% [95% confidence interval (CI): 90.5-91.2], a sensitivity of 91.4% [CI: 90.7-91.4], a specificity of 74.1% [CI: 69.2-78.6], and an area under the receiver operating characteristic value of 0.89 [0.87-0.92]. The important features for model development were the prehospital return of spontaneous circulation, prehospital adrenaline administration, and initial electrical rhythm. Conclusions: We developed a favorable machine learning model to predict the prognosis of non-cardiogenic OHCA with an initial non-shockable rhythm in the early stage of resuscitation.