Abstract 173: Update And External Validation Of A Prognostic Model To Predict Neurological Outcomes At 90 Days In Out-of-hospital Patients With Cardiac Arrest

Abstract

Introduction: An accurate assessment of postcardiac syndrome (PCAS) is crucial for achieving good neurological outcomes. Herein, we modified and refitted the existing prognostic model in the updated derivation set and externally validated it. Methods: Data used to update the model included those from adult non-traumatic out-of-hospital cardiac arrest (OHCA) patients with return of spontaneous circulation (ROSC) between Jan. 2013 and Dec. 2019 from the CRITICAL study (a multicenter registry in Osaka, Japan). The outcome was the dichotomized 90-day Cerebral Performance Category. The model was updated by logistic regression with the least absolute shrinkage and selection operator regularization. External validation included OHCA patients with ROSC, between Jan. 2013 and Dec. 2019, from the JAAM-OHCA registry (a nationwide multicenter registry in Japan), which geographically differs from the derivation set. The model performance was assessed in the validation set. Results: We updated the model in the derivation set (n = 3391). In the validation set (n = 6185), compared to Model A’s (variables with prehospital or pre-ROSC) C-statistics of 0.941 (95% CI: 0.933-0.950), Model B’s (Model A + first in-hospital documented cardiac rhythm) and Model C’s (Model B + laboratory data available immediately post-ROSC) C-statistics increased to 0.948 (0.940-0.956) and 0.959 (0.953-0.965), respectively (Fig. 1). All models were well-calibrated to the observed outcome. Model C produced higher net benefits at all risk thresholds following decision curve analysis (Fig. 2). A web-based calculator can show the probability of a poor outcome (https://pcas-prediction.shinyapps.io/90d_lasso/). Conclusion: In the validation set, the updated model showed excellent performance in predicting neurological outcomes at 90 days in patients with postcardiac syndrome. Additional laboratory data can provide better prediction by removing ambiguous predictors from the model.