A systematic search was performed in Medline, Embase and Cochrane Library. We searched for English-written studies from January 2010 to December 2021. This systematic review and meta-analysis was registered in the International prospective register of systematic reviews PROSPERO under the registration number: CRD42022295819. We adopted the search strategy following keywords and their variations; (1) population 1: “Child”, OR “Pediatrics” OR “Infant” OR “Neonate” and others, (2) population 2: “Cardiac Surgical Procedures” OR “Thoracic Surgery” OR “Cardiopulmonary Bypass” and others, (3) Intervention: “Extracorporeal Membrane Oxygenation” and others (Supplementary File 1).

We searched and enrolled English articles that present the outcome of PC-ECMO in the neonate and/or pediatric patients, but only those which met the Population, Interventions, Comparison and Outcomes (PICO) criteria (Supplementary File 2) were included in the analysis. The authors followed the guidelines for Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) (8). All three independent reviewers reviewed all articles (HJC, RH, ISJ) to screen titles and abstracts based on predefined inclusion criteria. The full text of the eligible articles was reviewed. Three reviewers extracted and compared information of all included articles and evaluated all the studies. When they disagreed, disagreements were either discussed to reach a consensus between the two reviewers or decided by a third reviewer. Selected studies also had to fulfill the following inclusion criteria: (1) provide data on neonates and/or pediatric patients who required PC-ECMO; (2) include patients aged 18 years or younger; (3) report data on in-hospital mortality; (4) be a prospective or retrospective observational study; (5) published in the English language as a full article; (6) include at least ten patients; (7) published since 2011. Studies were excluded if they (1) did not report information on the use of PC-ECMO or (2) did not provide data on the outcome (3) included only adult patients. Studies published in languages other than English (4) and those published before 2010 (5), were omitted. Review articles, animal studies, conferences, abstracts, meetings, studies with <5 in sample size, studies with unclear information and data from registry were also excluded (6) (Figure 1).

The retrieved articles were independently collected and checked by three researchers (HJC, RH, ISJ). The involvement of a third investigator settled any disagreement on collected data to help reach the consensus. There was no attempt to obtain specific or missing data from the authors. The following data were extracted: first author, year of publication, the overall number of cardiac surgery procedures performed during the study period, type of intervention, number of patients supported with PC-ECMO, gender, significant outcomes. The quality of the included studies was assessed by using RoBANS (Risk of Bias Assessment tool for Non-randomized Studies) (9). The RoBANS contains six domains, including (1) the selection of participants, (2) confounding variables, (3) measurement of intervention (exposure), (4) blinding of outcome assessment, (5) incomplete outcome data, and (6) selective outcome reporting. The primary outcome of this analysis was pooled proportional survival to hospital discharge in this cohort. The secondary outcomes were pooled survival of subgroup by each indication for PC-ECMO, pooled survival of neonatal ECMO, pooled survival differences between the single ventricle and biventricular physiology, pooled survival difference between genetic and non-genetic conditions (Supplementary File 3).

We analyzed the data using R (Version 3.6.3, R Foundation for Statistical Computing, Vienna, Austria) and Rex software (Version 3.6.3, RexSoft Inc., Seoul, South Korea). We showed different pooled proportional rate or risk ratio (RR) with 95% confidence interval (CI), using a forest plot with a fixed and random-effects model. Additionally, the possibility of publication bias was checked through a funnel plot. Cochran Q tests and I² were used to evaluate the possible heterogeneity. I² < 25% was considered as of low heterogeneity, 25–70% was of mediate heterogeneity and >70% was of high heterogeneity. If p-value > 0.1 and I² < 50%, the fixed effects model was selected; otherwise, the heterogeneity was assessed to determine whether the random effects model could be used. p-value < 0.05 was considered statistically significant.

Nine studies reported on the outcome of CPB weaning failure. The pooled survival rate of patients with CPB weaning failure was 44.7% (95% CI: 42.6–46.7; I²:0%). Nine studies reported on the outcome of LCOS. The pooled survival rate was 47.3% (95% CI: 39.9–54.7%; I²: 5%). Twelves studies reported on the outcome of PC-ECMO for cardiac arrest. The pooled survival rate was 37.6% (95% CI: 31.0–44.3%; I²: 32%). Four studies reported survival rate after PC-ECMO for respiratory failure. The pooled survival rate was 47.7% (95% CI: 32.5–63.1%; I²: 0%). But there was no significant difference in pooled survival in subgroup analysis, based on each indication of PC-ECMO support (interaction between groups. P = 0.19) (Figure 4).

Survival from PC-ECMO for single ventricle or biventricular physiology was reported by 12 studies. Patients with single ventricle physiology had a significantly lower survival probability, compared to patients with biventricular physiology (RR = 0.74; 95% CI: 0.63–0.86; I² = 40%, P < 0.001) (Figure 5).

Eight studies reported the survival rate of PC-ECMO in patients with genetic conditions. Patients with genetic condition had a similar survival probability, compared to patients without genetic condition (RR = 0.93, 95% CI: 0.53–1.65, I²: 65%, P = 0.812) (Figure 6).

Our meta-analyses indicate that the survival rate varies according to the conditions before PC-ECMO application. While 46–47% of patients who failed to wean from CPB and deteriorated in ICU due to low cardiac output syndrome and respiratory failure survived, those, who required ECMO for cardiac arrest showed the lowest survival rate (39%). In neonatal and pediatric patients with LCOS refractory to medical management, earlier initiation of ECLS facilitates unloading of myocardial burden and reduces the risk of cardiovascular collapse (3). While there is no consensus on the timing for ECMO initiation in the perioperative period, our meta-analyses showed that neonates and pediatric patients with LCOS supported with ECMO had lower mortality than patients with ECMO after cardiac arrest (ECPR).

Respiratory failure overall has a survival rate of 59–72%, according to the ELSO registry. However, the survival of PC-ECMO for respiratory failure in our meta-analysis was observed lower survival. Despite advancement in surgical technique, 2–30% of pediatric patients required ECMO support for hypoxia or respiratory distress (5, 7, 35). Although there is limited data regarding PC-ECMO for respiratory failure, four studies with 46 patients were included in our meta-analyses with a survival rate of 48%. Likely neonates and pediatric patients recovering from surgery for CHD have higher disease severity than the entire cohort of respiratory failure patients supported with respiratory ECMO.

The complexity of the underlying CHD and cardiac surgery for CHD is an essential determinant of survival to hospital discharge. As the complexity varies, the survival hospital discharge of those patients requiring ECMO also varies widely (20, 30, 36). Our meta-analysis included 560 patients with biventricular physiology reported by 12 studies. Single ventricle physiology was associated with a lower pooled survival rate than biventricular physiology (35.9 vs. 49.5%). Mascio et al. demonstrated analysis of children with perioperative mechanical circulatory support (MCS) by the Society of Thoracic Surgeons Congenital Heart Surgery Database. The operations with the highest MCS rate were the Norwood procedure (17%), and 53.2% of the MCS patients did not survive hospital discharge after cardiac surgery (37). Misfeldt et al. have reported admission rate of single ventricle patients requiring ECMO was 2.3% of all hospitalization, and the mortality rate of this population was 57.1%, with no change over time (27).

Recognizable syndromes, extracardiac malformations, and chromosomal anomalies associated with genetic conditions occur in ~20–30% of children with CHD (38). Over time, the exclusion strategy of genetic abnormalities from cardiac surgical palliation has changed (39); previously, physicians were more reluctant to provide ECMO to this population, but the view has been changed over time that with changing outcome and even ECMO in patients with Trisomy 13 or 18 is has been reported in the recent era (39, 40). In our meta-analysis, eight studies with data of genetic conditions were included. The neonates and pediatric patients with genetic conditions tend to be a similar survival outcome with a RR of 0.93 compared to those without genetic condition. Although limited literature was found regarding genetic abnormalities and CHD requiring ECMO, Furlong-Dillard JM et al. have reported 327 patients with genetic abnormalities who required ECMO after CHD surgery with the mortality rate of 2–3% (18). Alsoufi et al. reported that despite adjustment for prematurity and low birth weight, children with genetic abnormalities had a significantly increased risk of death after cardiac surgery compared to children without genetic conditions (41–44).

There are several limitations in our meta-analysis. First, this analysis included articles published during the last 10 years in an attempt to mitigate advancements in technology. However, there is a possibility that we could have missed some valuable and historical data from earlier studies. Second, we did not include studies with cohorts of a bridge to transplantation and preoperative ECMO for CHD. Third, we didn't collect the specific indications and conditions, such as timing/severity of illness, individual cardiac anomalies, and detailed genetic information. Lastly, all included articles are retrospective observational studies, while randomized control studies could have yielded better quality of our findings.