The authors reply

    loading  Checking for direct PDF access through Ovid


Ishii et al (1) highlight the unexpectedly high proportion of patients with excessive hyperoxia in our cohort. They argue that this raises concern about the quality of the data as it was not consistent with the recommended guidelines for arterial oxygen saturation suggested by the Extracorporeal Life Support Organization (ELSO). Although this is one possible explanation, we believe that it is also plausible that extracorporeal life support (ECLS) patients are commonly exposed to hyperoxia for the following reasons:
Many of these limitations highlighted by Ishii et al (1) and in our article (5), recently published in Critical Care Medicine, are likely due to the use of registry data, and therefore, these findings are meant to be hypothesis generating and confirmed in prospective studies. However, Ishii et al (1) speculate that if these values are true, they highlight the medical staff’s unfamiliarity with ECLS. Although this is possible, the ELSO Registry reflects real-world ECLS use, and we did not have details surrounding institutional characteristics to explore whether center characteristics are associated with the frequency of hyperoxia.
They suggest an analysis excluding excessive hyperoxia PaO2 greater than 200 mm Hg in the VV ECMO cohort. In an exploratory analysis, we evaluated different thresholds of hyperoxia within the categories created in increments of 20 mm Hg; however, no single threshold was associated with outcome—possibly attributable to the small numbers in each category (Appendix Tables 4 and 6 in [5]).
Finally, we used logistic regression for our primary outcome of hospital mortality, as mortality at other time points was not available. The difference in mortality between a 28-day measure and “in-hospital” measure was assumed to not likely be remarkably different given the acuity of the patients and the high proportion of death within 28 days (i.e., it would be unlikely that patients on ECMO would be palliated and discharged from the ICU). It would also be unlikely for patients who are still admitted after 28 days to die due to hyperoxia after day 28. Furthermore, observational studies have demonstrated high correlation between in-hospital mortality and 30-day mortality across patients with acute medical conditions (6). A Cox proportional hazard model could also be an appropriate method to perform our analysis; however, we were less interested in time to death and more interested in whether death occurred. Future research, in addition to choosing a standardizable outcome, needs to also consider the time-varying nature of the exposure.
    loading  Loading Related Articles