Results
Our electronic search retrieved 1647 citations, 34 of which were selected for full-text review (Figure 1). Twelve studies with a combined population of 194 patients fulfilled the inclusion criteria (Table 1). 136 patients reached an endpoint of weaning from ECMO or death while the rest were still on ECMO. Overall study validity was acceptable, with a median score of 6 on the Newcastle Ottawa scale NOS appraising the quality of observational studies, without being opposed by their non-randomized design. The median Berlin score for ARDS prior to starting ECMO was III. Patients received mechanical ventilation before ECMO implementation for a median of four days and ECMO was maintained for a median of 13 days (Table 2,3)
Random-effect pooled estimates suggested an overall in-hospital mortality risk ratio of 0.49 (95% confidence interval 0.259 to 0.721; I2 = 94%) (Figure 2). Most of the preliminary studies were from China (seven studies with 41 patients having endpoints). Larger studies then followed from the USA, Japan and France (Five studies with 95 patients with endpoints).
To investigate the overall inter-study heterogenicity, a subgroup analysis was performed according to the country of origin of each study (Figure 3) This showed persistent heterogeneity only in the 7 Chinese studies with pooled mortality risk ratio of 0.66 (I2 = 87%) (95% CI = 0.39-0.93), while the later larger studies coming from the USA showed pooled estimate mortality risk ratio of 0.41 (95% CI 0.28-0.53) with homogeneity (p=0.67) similar to France with a pooled mortality risk ratio of 0.26 (95% CI 0.08-0.43) with homogeneity (p=0.86)
In four of our studies, there was a control group who received mechanical ventilation for severe ARDS. The mortality rate was 87.5% in the ECMO patients and 69.2% in conventional therapy patients. The pooled odds of mortality in ECMO versus conventional therapy were not significantly different (p=0.273, 95%CI: 0.06–1.111). There was no observable heterogeneity (I2 = 0%, Cochran’s Q, p-value = 0.57 ( (Figure 4)
Moderators tested were age, Pre ECMO-Berlin score of ARDS, ventilation days before ECMO, and duration of days on ECMO. Exploratory meta-regression identified age as a significant negative moderator of mortality (P =0.02) with younger age patients at a higher risk of death. No other factors demonstrated a significant moderator but this lack of a statistical significance for likely or established prognostic factors should be viewed with consideration of the limited statistical power of meta-regression when applied to a limited data-set. Publication bias was excluded by visualizing the funnel plot of standard error (Figure 5). The funnel plot is symmetrical with only 2 studies outside the threshold (20,23). Exclusion of bias was also proven with an Egger’s test value p=0.566 and Begg&Mazumdar test value with p=0.373