OBJECTIVES: We evaluate the association between pre-extracorporeal membrane oxygenation (ECMO) markers of impaired oxygen delivery, as quantified by the impaired oxygen delivery index (iDo2; Etiometry, Boston, MA) and neurologic outcomes in infants supported on ECMO. DESIGN: A single-center retrospective cohort study using demographic and clinical data along with iDo2 metrics calculated at 120-minute intervals before ECMO initiation. Primary outcomes included mortality, electroencephalography abnormalities, and head imaging findings. The secondary outcomes included brain MRI abnormalities and Functional Status Score (FSS) at discharge. SETTING: Pediatric ECMO center in North America. PATIENTS: We included infants younger than 1 year old who were supported on ECMO between 2013 and 2017. We excluded cases with congenital diaphragmatic hernia or postcardiac surgery ECMO. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We identified 47 infants meeting the inclusion criteria. Median (interquartile range IQR) age and weight at ECMO initiation were 16 days (IQR, 6-112 d) and 3.3 kg (IQR, 2.8-4.8 kg). The overall mortality was 26 of 47 (55% 95% CI, 41-69%). Nonsurvival compared with survival was associated with a greater proportion of patients with congenital heart disease (p = 0.03) and pre-ECMO cardiac arrest (p = 0.006). Time spent above iDo2 thresholds of 25%, 50%, and 75% increased closer to ECMO initiation. Higher iDo2 dose correlated with adverse neurologic outcomes, including electroencephalography abnormalities and abnormal imaging. Last, a logistic regression model for poor outcome (composite of death and FSS), including the time spent above the three iDo2 thresholds at the three time intervals pre-ECMO, showed classification model performance with an area under the receiver operating characteristic curve 0.81 (95% CI, 0.7-0.92; p = 0.0002). CONCLUSIONS: Markers of impaired oxygen delivery, such as iDo2 in the 6 hours before ECMO are associated with subsequent outcome. Incorporation of these metrics into clinical decision-support systems need to be prospectively tested and evaluated.
Rabinowitz et al. (Mon,) studied this question.