Inclusion criteria: (1) study design: the study was original research from prospective or retrospective studies; (2) participants: adults and children; (3) exposure: refractory septic shock treated with ECMO; (4) comparator: survival at discharge; and (5) outcome: death or survival. Exclusion criteria: the review, commentaries, opinions, guidelines, pathological studies. Two investigators independently searched PubMed, Embase, and Cochrane Library databases for studies that enrolled patients with refractory septic shock treated with ECMO, which were published before July 9, 2021 and restricted to English. The search phrases for the three databases included Boolean terms “AND” and “OR” with the following keywords in various possible combinations: “ECMO,” “extracorporeal membrane oxygenation,” “Infant,” “newborn,” “neonatal,” “pediatric,” “adults,” and “refractory septic shock” (details of the search strategy are included in the Supplementary Material). Two authors screened titles, abstracts, or full texts and determined their eligibility. In addition, a manual search of all relevant studies and their citation lists was performed to identify additional articles for inclusion. And no restrictions were placed on study type (prospective or retrospective) because we have used the Joanna Briggs Institute Checklist to assess all the studies and put it in the Supplementary Data.

Meta-analysis was performed using R software. Data including study design, outcomes, patient characteristics, and interventions were extracted independently. Survival to discharge was the outcome measure in our meta-analysis. Subgroup analysis was also performed to analyze the differences between the adult and pediatric groups. Briefly, for the meta-analysis of proportions, the exact confidence interval (CI) for each proportion was computed using the Clopper-Pearson method (10). I²-tests were performed to assess the heterogeneity of the summary rates. Statistical heterogeneity between studies was identified using I² statistics, where I² ≤ 40%, between 30 and 60%, between 50 and 75%, and ≥75% indicated low, moderate, substantial, and considerable heterogeneity, respectively. P-values for I² statistics were derived from the chi-square distribution of Cochran's Q-test.

The results of the study selection process are presented in Figure 1. A total of 293 records were found in PubMed, Embase, and the Cochrane Library databases. And we excluded 36 records and 243 irrelevant records (review, commentaries, opinions, guidelines, pathological) in this meta-analysis. After excluding duplicates, irrelevant studies, and studies without corresponding data, we eventually identified 14 studies that met the inclusion criteria.

Of the 293 articles screened, 14 were identified for systematic review and meta-analysis. The cumulative estimate of survival (14 studies, 535 patients) in the cohort was 39% (95% CI: 27–51%) (Figure 2). An assessment of the funnel plots of all the patients is shown in Figure 3. The 14 studies included in the systemic review and meta-analysis that reported on children with refractory septic shock requiring ECMO are shown in Table 1 (11–24). And the details of the cohort studies are summarized in Table 2. Furthermore, the compilation of adverse events associated with the use of ECMO are provided in Table 3.

During the subgroup analysis, the cumulative estimate of survival in the adult group (6 studies, 276 patients) in the cohort was 18% (95% CI: 10–27%) (Figure 4), and an assessment of the funnel plot of adults is shown in Figure 5. The cumulative estimate of survival in the pediatric group (8 studies, 259 patients) in the cohort was 53% (95% CI: 47–59%) (Figure 6), and the assessment of the funnel plot of children is shown in Figure 7.

There was no evidence of publication bias, and the methodological quality of all the studies included in our analysis scored high using the JBI critical appraisal tool (Supplementary Table 1). The GRADE analysis demonstrated moderate to high certainty in the evidence presented in this paper.

There was great heterogeneity in the forest and funnels plots of all patients (Figures 2, 3). Consequently, we selected the random-effects model to assess the overall survival rate of patients treated with ECMO, which was 39%. We divided the patients into two subgroups: adults, and children and neonates.

The adult group with refractory septic shock treated with ECMO had significantly lower survival rates (18%) than pediatric group. In addition, the heterogeneity in the forest and funnel plots of adults was lower than that of the overall patients (Figures 4, 5). The pediatric and neonate group with refractory septic shock treated with ECMO had significantly higher survival rates than the adult groups. In addition, the heterogeneity in the forest and funnel plots of pediatrics was lower than that of the overall patients (Figures 6, 7). This indicates that our subgroup analysis is of great significance in reducing heterogeneity.

Huang et al. have concluded that for patients with refractory septic shock treated with ECMO, the older the adult, the worse the prognosis. All 20 adults aged over 60 years died despite the use of ECMO, which indicates that the outcomes of these patients remain unsatisfactory. There are three possible explanations for this phenomenon: First, refractory septic shock has a variety of hemodynamic presentations. Left ventricular dysfunction with reduced cardiac output is commonly seen in infants and children; however, distributive shock, a hyperdynamic state with high cardiac output, usually manifests in adults (28). ECMO is used primarily for cardiac or cardiopulmonary support and is intuitively more beneficial for patients with ventricular dysfunction than for those with profound vasodilation. Thus, the performance of ECMO is better in pediatric than adult refractory septic shock. Second, among adults, older patients have more comorbid diseases (such as diabetes and hypertension), which are more difficult to recover from after ECMO for life support. Thus, the adult survival rates declined at discharge. According to our experience, although some children with refractory septic shock who need ECMO in PICU also have chronic comorbidity such as type 1 diabetes, congenital heart disease, chronic lung disease, the incidence rate is lower than that of adults, and the exposed time to chronic diseases is shorter than adults. Third, cardiovascular diseases increase dramatically with age in human. While it is clear that advanced age allows more time for individuals to be exposed to risk factors in general, there is strong evidence that age itself is a major independent risk factor for death (29). These three points may be the cause of the significant differences between the adult and pediatric groups.

Han et al. (22) have reported that the shock-to-ECMO interval for adults at 12, 18, and 24 h during ECMO between the survival and death groups was significantly different. They found that shock-to-ECMO interval before ECMO placement in the survival group were significantly lower than those in the death group (23.5 vs. 42.2 h, P = 0.037). Cheng et al. (17) have found that better outcomes were associated with in ECMO patients with door-to-ECMO times of 96 h or less. Solé et al. (21) have reported 21 refractory septic shock patients treated with ECMO, 9 were pediatric and 12 were newborns. And they were diagnosed with septic shock for a median duration of 29.5 h before ECMO was started (IQR, 20–46), with significant differences between the survival and non-survival groups (P = 0.009). These three studies indicate that the shorter the time from refractory septic shock to ECMO, the higher the survival rate at discharge.

First, this study lacked long-term survival rate for patients with refractory septic shock treated with ECMO; however, ECMO is a supportive method, not a treatment. The most important points in patient management are the control of the focus of infection and the early initiation of adequate antibiotics. Second, the number of the cases in some studies included in this meta-analysis are small. Thus, further multicenter studies conducted with larger sample sizes are needed to confirm our findings.