Identifying Prognostic Factors and Developing Accurate Outcome Predictions for In-hospital Cardiac Arrest by Using Artificial Neural Networks

Background: Accurate estimation of neurological outcomes after in-hospital cardiac arrest (IHCA) provides crucial information for clinical management. This study used artificial neural networks (ANNs) to determine the prognostic factors and develop prediction models for IHCA based on immediate preresuscitation parameters.

Methods: The derived cohort comprised 796 patients with IHCA between 2006 and 2014. We applied ANNs to develop prediction models and evaluated the significance of each parameter associated with favorable neurological outcomes. An independent dataset of 108 IHCA patients receiving targeted temperature management was used to validate the identified parameters. The generalizability of the models was assessed through fivefold cross-validation. The performance of the models was assessed using the area under the curve (AUC).

Results: ANN model 1, based on 19 baseline parameters, and model 2, based on 11 prearrest parameters, achieved validation AUCs of 0.978 and 0.947, respectively. ANN model 3 based on 30 baseline and prearrest parameters achieved an AUC of 0.997. The key factors associated with favorable outcomes were the duration of cardiopulmonary resuscitation; initial cardiac arrest rhythm; arrest location; and whether the patient had a malignant disease, pneumonia, and respiratory insufficiency. On the basis of these parameters, the validation performance of the ANN models achieved an AUC of 0.906 for IHCA patients who received targeted temperature management.

Conclusion: The ANN models achieved highly accurate and reliable performance for predicting the neurological outcomes of successfully resuscitated patients with IHCA. These models can be of significant clinical value in assisting with decision-making, especially regarding optimal postresuscitation strategies.