Validating the 2025 ERC/ESICM neuroprognostication algorithm using a restricted predictor set in a minimal WLST cohort

Abstract

Abstract Background The 2021 European Resuscitation Council (ERC)/European Society of Intensive Care Medicine (ESICM) neuroprognostication guidelines were highly specific but limited by strict entry criteria (Glasgow Coma Scale-Motor [GCS-M] ≤ 3), leaving nearly half of patients with an “indeterminate” prognosis. The updated 2025 guidelines introduced an expanded entry (GCS-M < 6) and a two-tiered sequential algorithm integrating favorable and unfavorable criteria to resolve this clinical uncertainty. This study validated the real-world performance of the 2025 algorithm using a pragmatic, restricted core predictor set in a cohort with a low frequency of withdrawal of life-sustaining therapy (WLST). Methods This retrospective observational study included adult comatose out-of-hospital cardiac arrest (OHCA) survivors treated with targeted temperature management who remained unconscious (GCS-M < 6) at 72 h between 2020 and 2025. Six-month functional outcomes were categorized using the modified Rankin Scale (poor: 4–6). The 2025 algorithm was applied using four unfavorable predictors (absent pupillary/corneal reflexes, status myoclonus, neuron-specific enolase [NSE] > 60 µg/L, gray-to-white matter ratio on brain computed tomography ≤ 1.10 via automated software) and two favorable predictors (GCS-M 4–5 on day 4, NSE ≤ 17 µg/L). Results Among 400 OHCA survivors, 129 were excluded due to early awakening ( n = 90) or death within 72 h ( n = 39). Of the 271 included patients, 233 (86.0%) had poor outcomes, and the withdrawal of life-sustaining therapy rate was low (5.5%). The integrated 2025 algorithm predicted poor outcomes with 76.2% sensitivity (95% CI, 68.4–82.9), 97.1% specificity (95% CI, 85.1–99.9), a 99.1% positive predictive value (95% CI, 95.0–100.0), and a 50.0% negative predictive value (95% CI, 37.6–62.4). Step 2 favorable criteria reclassified 48.9% (68/139) of patients who remained unclassified after Step 1, reducing the unclassified proportion from 55.8% (139/249) to a final indeterminate rate of 28.5% (71/249). Among patients with indeterminate outcomes, 97.2% (69/71) ultimately had poor outcomes. Among 22 patients with discordant favorable and unfavorable predictors, 6 (27.3%) had good outcomes. Conclusions In a cohort with a low frequency of WLST, the 2025 ERC/ESICM algorithm maintained high specificity, while Step 2 substantially reduced the proportion of unclassified patients. However, the restricted predictor set may have limited sensitivity and contributed to persistent indeterminate classifications.