Data collection and analysis

Data were collected retrospectively from electronic records, including the IntelliVue Clinical Information Portfolio (Philips, Eindhoven, the Netherlands). Pre-ECMO data were obtained from ECMO referral systems [24], paper records or direct interview with members of retrieval teams. RESP score was calculated at the time of referral. Sequential Organ Failure Assessment score was calculated from pre-ECMO clinical parameters and day 0 ECMO laboratory results, with Glasgow Coma Scale presumed to be 15 unless deranged consciousness pre-intubation was confirmed. Data were collected until all patients had reached either death (nonsurvivors) or discharge from hospital (survivors). Quantitative variables are presented with median and interquartile range (IQR), and categorical variables are presented as frequencies and percentages. Missing data were not imputed and are recognised in tables by adjusted n-values. Comparison of nonparametric continuous variables used Mann–Whitney U-tests with significance at p<0.05 (Python: package SciPy v.1.4.1). The study had institutional approval and waiver of individual informed consent (reference no.10 796), qualifying as service evaluation defined by the UK National Health Service Health Research Authority (www.hra.nhs.uk).