External validation of cardiac arrest-specific prognostication scores developed for early prognosis estimation after out-of-hospital cardiac arrest in a Korean multicenter cohort

Abstract

We evaluated the performance of cardiac arrest-specific prognostication scores developed for outcome prediction in the early hours after out-of-hospital cardiac arrest (OHCA) in predicting long-term outcomes using independent data. The following scores were calculated for 1,163 OHCA patients who were treated with targeted temperature management (TTM) at 21 hospitals in South Korea: OHCA, cardiac arrest hospital prognosis (CAHP), C-GRApH (named on the basis of its variables), TTM risk, 5-R, NULL-PLEASE (named on the basis of its variables), Serbian quality of life long-term (SR-QOLl), cardiac arrest survival, revised post-cardiac arrest syndrome for therapeutic hypothermia (rCAST), Polish hypothermia registry (PHR) risk, and PROgnostication using LOGistic regression model for Unselected adult cardiac arrest patients in the Early stages (PROLOGUE) scores and prediction score by Aschauer et al. Their accuracies in predicting poor outcome at 6 months after OHCA were determined using the area under the receiver operating characteristic curve (AUC) and calibration belt. In the complete-case analyses, the PROLOGUE score showed the highest AUC (0.923; 95% confidence interval [CI], 0.904–0.941), whereas the SR-QOLl score had the lowest AUC (0.749; 95% CI, 0.711–0.786). The discrimination performances were similar in the analyses after multiple imputation. The PROLOGUE, TTM risk, CAHP, NULL-PLEASE, 5-R, and cardiac arrest survival scores were well calibrated. The rCAST and PHR risk scores showed acceptable overall calibration, although they showed miscalibration under the 80% CI level at extreme prediction values. The OHCA score, C-GRApH score, prediction score by Aschauer et al., and SR-QOLl score showed significant miscalibration in both complete-case (P = 0.026, 0.013, 0.005, and < 0.001, respectively) and multiple-imputation analyses (P = 0.007, 0.018, < 0.001, and < 0.001, respectively). In conclusion, the discrimination performances of the prognostication scores were all acceptable, but some showed significant miscalibration.