Quantitative pupillometry thresholds for early neuroprognostication after cardiac arrest: A BOX-trial substudy

Abstract Funding Acknowledgements Type of funding sources: Private grant(s) and/or Sponsorship. Main funding source(s): Novo Nordisk Foundation Background Quantitative pupillometry as part of multimodal prognostication in comatose out-of-hospital cardiac arrest (OHCA) patients is promising, but the lack of validated thresholds for outcome presents a challenge for clinical use. Studies have suggested that a percentage reduction of pupillary size (qPLR, %) ≤3.8% and a neurological pupil index (NPi, 0-5) ≤2 is 100% specific for poor neurological outcome. However, this remains to be validated in a large consecutive population. Purpose To assess the proposed thresholds for qPLR and NPi for early identification of poor neurological outcome, defined as Cerebral Performance Category (CPC) 3–5 at 3-months follow-up after OHCA. Methods This predefined, prospective substudy from a large multicenter, randomized trial, obtained quantitative pupillometry measurements from consecutively enrolled patients at 0, 24, 48, 72, 96, and 120 hours after OHCA (t0-t120). However, pupillometry was not used to guide clinical decision-making. Specificity and sensitivity were calculated for the proposed thresholds of qPLR and NPi. Overall prognostic performances were assessed with logistic regression and area under the receiver operating characteristic (ROC) curves (AUC). Models were further adjusted for potential confounders. Results Of the 789 patients comprising the BOX-trial population, serial measurements of quantitative pupillometry were performed in a total of 710 (90%) patients. Patients had a mean age of 63±14 (range 18-93) years, predominantly of male sex (82%), with primary outcome occurring in 266 (37%) patients. For all time points, both median values of qPLR and NPi were significantly lower in patients with poor neurological outcome. The highest AUC was found at t48 for both qPLR and NPi. In the unadjusted model, AUC was 0.80 [95%CI 0.76-0.84] for qPLR and 0.79 [95%CI 0.75-0.83] for NPi (Fig 1). In the adjusted model AUC was 0.92 [95%CI: 0.87-0.96] for qPLR and 0.92 [95%CI: 0.88-0.97] for NPi (Fig 2). Both parameters had a significant contribution to the AUC in the adjusted model, with no significant difference in performance between qPLR and NPi in either model. At t48, specificity for CPC 3-5 was 99% [95%CI 97-100] with sensitivity at 20% [95%CI 15-26] for qPLR≤3.8% and 100% [95%CI 99-100] with sensitivity at 10% [95%CI 6-14] for NPi ≤2. At time points t0, t24, t72, t96, and t120 specificity was 91%, 96%, 99%, 100%, and 100%, respectively for qPLR≤3.8%, and 100% for all time points for NPi ≤2. Conclusion The predictive performance of qPLR and NPi was equally good with optimal timing of measurements at 48 hours after admission. The BOX-trial validates thresholds of NPi ≤2 with 100% specificity for outcome at all time points and of qPLR≤3.8% with 99-100% specificity at t48-t120. These findings support the clinical usefulness of quantitative pupillometry for multimodal neuroprognostication of comatose OHCA patients.