Neuroprognostication after cardiac arrest in patients without withdrawal of life-sustaining therapy: a prospective observational multicenter study

Abstract Background Neuroprognostication after cardiac arrest (CA) is critical for guiding treatment decisions. However, concerns about self-fulfilling prophecy related to withdrawal of life-sustaining therapy (WLST) complicate outcome prediction. This study evaluated the accuracy of guideline-recommended prognostic markers within 14 days after CA for predicting poor 12-month outcomes in patients without WLST during the first four weeks after CA. Methods This prospective multicenter observational study enrolled adults who remained comatose 72 h after CA across eight German hospitals between 2014 and 2017. Patients with WLST during the first four weeks, stroke, pre-existing disorders of consciousness, or terminal malignancy were excluded, leaving 101 patients for analysis. Prognostic markers assessed included pupillary light and corneal reflexes (PLR + CR), EEG, somatosensory evoked potentials (SEP), neuron-specific enolase (NSE) concentration, and the best Coma Recovery Scale–Revised (CRS-R) score. Poor outcome was defined as a modified Rankin Scale score of 4–6 at 12 months. Results Poor outcomes occurred in 67.3% of patients. Individual markers demonstrated high specificity but limited sensitivity, with variable false-positive rates (FPRs). Repeated assessments and the combination of two or more unfavorable markers markedly reduced FPRs. Within 14 days, the presence of two or more unfavorable markers yielded no false positives and was strongly associated with poor outcome (adjusted OR 34.7). Adding the best CRS-R score to established markers significantly improved prognostic discrimination (AUC 0.925 vs. 0.888). Conclusions In patients who remained comatose three days after CA and in whom multimodal prognosis becomes clinically relevant, guideline-recommended assessment identified poor-outcome patients with high specificity and consistency with prior multimodal prognostication cohorts. Although self-fulfilling-prophecy bias cannot be fully excluded in any observational prognostication cohort, the marker profiles of WLST-excluded patients within the first four weeks matched or exceeded those of patients with documented poor outcomes and did not resemble those of recoverable patients, which reduces - though it does not completely rule it out - the likelihood that such bias is the primary reason for the observed associations. The best CRS-R score provided additional prognostic value, particularly in cases of uncertain prognosis, supporting its integration into multimodal assessment frameworks. Trial registration ClinicalTrials.gov: NCT02231060. Registered: August 2014. First patient enrolled: October 2014.