Prognostication of cardiac arrest survivors using low apparent diffusion coefficient cluster volume.

Introduction: We developed a new neuroprognostication method for cardiac arrest (CA) using the relative volume of the most dominant cluster of low apparent diffusion coefficient (ADC) voxels and tested its performance in a multicenter setting.

Methods: Adult (>15 years) out-of-hospital CA patients from three different facilities who underwent an MRI 12h after resuscitation were retrospectively analyzed. Patients with unknown long-term prognosis or poor baseline neurologic function were excluded. Average ADCs (mean and median), LADCV (relative volume of low-ADC voxels) and DC-LADCV (relative volume of most dominant cluster of low-ADC voxels) were extracted using different thresholds between 400 and 800 × 10(-6) mm(2) s(-1) at 10 × 10(-6) mm(2) s(-1) intervals. Area under the receiver operating characteristic curve (AUROC) and sensitivity for poor outcome (6-month cerebral performance category score >2) while maintaining 100% specificity were measured.

Results: 110 patients were analyzed. Average ADCs showed fair performance with an AUROC of 0.822 (95% confidence interval [CI], 0.744-0.900) for the mean and 0.799 (95% CI, 0.716-0.882) for the median. LADCV showed better performance with a higher AUROC (maximum, 0.925) in an ADC threshold range of 400 to 690 × 10(-6) mm(2) s(-1). DC-LADCV showed the best performance with a higher AUROC (maximum, 0.955) compared with LADCV in an ADC threshold range of 600 to 680 × 10(-6) mm(2) s(-1). DC-LADCV had a high sensitivity for poor outcomes (>80%) in a wide threshold range from 400 to 580 × 10(-6) mm(2) s(-1) with a maximum of 89.2%.

Conclusions: Quantitative analysis using DC-LADCV showed impressive performance in determining the prognosis of out-of-hospital CA patients in a multicenter setting.