Validation of the UB-ROSC Score for Predicting OHCA Survival in Chiayi City, Taiwan

Out-of-hospital cardiac arrest (OHCA) is a critical public health issue, with survival rates varying widely due to multiple well-established factors, including emergency medical service (EMS) system structure, bystander response, initial cardiac rhythm, and patient-level characteristics. This study evaluates the Utstein-Based Return of Spontaneous Circulation (UB-ROSC) score's performance in predicting sustained Return of Spontaneous Circulation (ROSC > 2 h) in a midsized Asian city where EMS delivery is standardized and geographic variation in access is minimal. A total of 209 OHCA cases from Chiayi City in 2024 were analyzed, using Utstein-aligned data to compute UB-ROSC scores. Predictive performance was assessed via receiver operating characteristic (ROC) curves, calibration intercept and slope, the Hosmer-Lemeshow goodness-of-fit test, and bootstrap internal validation, focusing on sustained return of spontaneous circulation (ROSC) (> 2 h). After excluding pediatric and traumatic cases, 209 cases were analyzed. The cohort was 60.3% male, with 47.8% witnessed arrests and 67.0% receiving bystander CPR. The UB-ROSC score demonstrated fair discrimination, with an area under the ROC curve (AUC) of 0.78 (95% confidence interval [CI]: 0.70-0.85), stable on bootstrap and cross-validation. The calibration slope was 0.90 (95% CI: 0.62-1.18) but the calibration intercept was +0.71 (95% CI: 0.36-1.05), and the Hosmer-Lemeshow test was significant (χ ² = 28.6, p < 0.001), indicating systematic underprediction of observed survival. Risk stratification yielded a low-risk group (n = 109) with a 15.6% ROSC rate, a medium-risk group (n = 94) with a 48.9% rate, and a high-risk group (n = 6) with an 83.3% rate. The Chiayi cohort showed a slightly higher observed ROSC rate in the low-risk group, whereas the medium- and high-risk groups fell within the predicted ranges although estimates in the high-risk group were imprecise because of the small sample size. Local recalibration may be needed before applying absolute probability estimates in clinical settings.