Abstract:Objective To develop and validate a decision tree model to identify in-hospital cardiac arrest among chest pain patients at the emergency department. Methods Data of 3 146 patients who complained of nontraumatic acute chest pain and rescued in the emergency department were collected. Seventy-one patients experienced cardiac arrest were treated as the case group, and 142 patients without cardiac arrest were treated as the control group. Decision tree analysis was performed to determine the early warning system of cardiac arrest, and 10-fold cross validation was used to verify. Performance of the decision tree model was compared with the GRACE, TIMI and HEART scores. Results The decision tree included three layers and five nodes of diastolic blood pressure, Killip class, troponin I, duration of chest pain, and creatine kinase. The area under the ROC curve of the decision tree model was 0.893, which was significantly higher than TIMI score (0.817) and HEART score (0.801) (P<0.05 for both), and was also higher than GRACE score (0.857) but without significant difference (P>0.05). Conclusion The decision tree model with good accuracy, intuitive results, and clear logic in predicting cardiac arrest can be used as a decision-making reference for medical staff on risk management of chest pain.