Hemoperfusion during extracorporeal membrane oxygenation: an updated systematic review and meta-analysis of 8,151 patients

Abstract

Background

This systematic review and meta-analysis evaluated the clinical efficacy and safety of hemoperfusion in adults receiving extracorporeal membrane oxygenation (ECMO).

Methods

We systematically searched PubMed, Embase, the Cochrane Library, CNKI, and CBM for studies comparing ECMO plus hemoperfusion (ECMO + HP) with ECMO alone in adult patients. The primary outcome was all-cause mortality. Prespecified subgroup analyses were conducted according to study design (RCT vs non-RCT), ECMO type (VA vs VV), and etiology (cardiogenic shock vs cardiac arrest). Risk of bias was assessed using RoB 2.0 and ROBINS-I, and Trial Sequential Analysis (TSA) was performed. The systematic review was registered with PROSPERO (CRD42025643149) and conducted in accordance with PRISMA guidelines.

Results

Thirteen studies involving 8,151 patients were included, comprising three RCTs and ten non-RCTs. Overall, ECMO + HP did not significantly reduce all-cause mortality compared with ECMO alone (OR 0.98, 95% CI 0.49–1.94; P = 0.95; I² = 76%). However, mortality was higher with ECMO + HP in RCTs (OR 4.96, 95% CI 1.67–14.77; P = 0.004; I² = 34%), whereas no significant difference was observed in non-RCTs (OR 0.58, 95% CI 0.25–1.33; P = 0.20; I² = 79%). Subgroup analyses by ECMO type showed no survival benefit with either VA-ECMO (OR 0.80, 95% CI 0.41–1.58; P = 0.53; I² = 47%) or VV-ECMO (OR 1.23, 95% CI 0.30–4.99; P = 0.77; I² = 77%). Among VA-ECMO patients, etiology-based analyses showed no mortality benefit in cardiogenic shock (OR 0.58, 95% CI 0.31–1.09; P = 0.09; I² = 22%) or cardiac arrest (OR 2.42, 95% CI 0.71–8.25; P = 0.16; I² = 26%). No significant differences were identified in other clinical outcomes or complications, and TSA indicated that the evidence remains insufficient to confirm a mortality benefit. Most studies were at moderate to serious risk of bias.

Conclusion

In patients receiving extracorporeal membrane oxygenation, hemoperfusion did not reduce mortality and may be associated with increased mortality in randomized controlled trials. Current evidence does not support routine use of ECMO + HP in ICU condition.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13054-026-06203-6.

Introduction

Extracorporeal membrane oxygenation (ECMO) is a life-saving therapy for patients with refractory cardiogenic shock or severe respiratory failure, providing temporary support for gas exchange and systemic circulation until recovery or definitive treatment is possible [1, 2]. However, ECMO is frequently associated with an excessive inflammatory response arising from both the underlying critical illness and the extracorporeal circuit itself [3, 4]. This hyperinflammatory state may lead to vasoplegia, capillary leak, and multiple organ dysfunction, all of which are associated with worse clinical outcomes [5, 6].

In response to this challenge, hemoperfusion has increasingly been used in patients receiving ECMO to modulate excessive inflammation. Several adsorption cartridges, including CytoSorb, HA330, HA380, and oXiris, are currently available for clinical use. Previous studies have suggested that these devices can be safely integrated into ECMO circuits and may reduce inflammatory mediators, vasopressor requirements, and hemodynamic instability, with the potential to improve clinical outcomes [7–10]. The rationale for hemoperfusion lies in its ability to remove endotoxins, proinflammatory cytokines, and other harmful mediators directly from the bloodstream, thereby attenuating their pathophysiologic effects [11, 12].

However, the clinical benefits of hemoperfusion remain controversial. Several studies have failed to demonstrate meaningful improvements in clinical outcomes, and high-quality evidence is still lacking [13–15]. Therefore, we conducted this systematic review and meta-analysis to synthesize the available evidence on hemoperfusion in patients receiving ECMO and to provide an updated evidence base for extracorporeal life support practice in critically ill adults.

Methods

Protocol and registration

This systematic review and meta-analysis was prospectively registered in PROSPERO (CRD42025643149) and conducted in accordance with the PRISMA 2020 statement [16].

Inclusion and exclusion criteria

Studies meeting the following PICOS (Population, Intervention, Comparator, Outcome, Study (design) criteria were included:

• P:Adult patients (≥18 years) receiving ECMO.

• I: Hemoperfusion.

• C: Treated with ECMO but not receiving hemoperfusion.

• O: All-cause mortality (30-day, in-hospital, ICU mortality).

• S: Randomized controlled trials (RCTs), propensity score matching (PSM) studies, and retrospective observational studies.

Exclusion criteria were case reports, animal studies, review articles, editorials, and conference abstracts. No restrictions were placed on publication date or language.

Literature search strategy

A comprehensive search of PubMed, Embase, the Cochrane Library, CNKI, and CBM was conducted from database inception to January 15, 2026. Search strategies combined keywords and MeSH/Emtree terms related to “extracorporeal membrane oxygenation” and “hemoperfusion” (see Supplementary Table S5). Reference lists of relevant reviews and included studies were also screened manually to identify additional eligible studies.

Study selection and data extraction

Two reviewers (XT and YCY) independently screened titles and abstracts, removed duplicates, and assessed full-text articles for eligibility. Discrepancies were resolved by consultation with a third reviewer (ZP).

Two reviewers (XT and YCY) independently extracted data, including first author, publication year, study design, study period, sample size, baseline characteristics, underlying cause, hemoperfusion cartridge, and treatment scheme. The primary outcome was all-cause mortality (30-day, in-hospital, or ICU mortality). Secondary outcomes included inflammatory biomarkers (IL-6, endotoxin, PCT, CRP, and lactate), vasoactive requirements (norepinephrine dose or vasoactive inotropic score [VIS]), other clinical outcomes (ECMO duration, ICU or hospital length of stay, and duration of mechanical ventilation), and complications (hemorrhage and infection).

Risk of bias assessment

Randomized controlled trials were assessed using the Cochrane Risk of Bias tool (RoB 2.0) [17], while non-randomized studies were evaluated using the Risk of Bias in Non-randomized Studies of Interventions tool (ROBINS-I) [18], with assessments performed independently by two reviewers.

Statistical analysis

All statistical analyses were performed using Stata/SE version 12.0 (StataCorp, College Station, TX, USA), RevMan 5.4, and Trial Sequential Analysis (TSA) software (version 0.9.5.10 beta; Copenhagen Trial Unit, Copenhagen, Denmark).

1. Effect Measures and Models

Dichotomous outcomes were analyzed using Odds Ratios (OR) with 95% Confidence Intervals (CI). Continuous outcomes were expressed as Mean Differences (MD) or Standardized Mean Differences (SMD). Heterogeneity was assessed using the Cochran Q test and the I² statistic. Given the anticipated clinical and methodological diversity, a random-effects model (DerSimonian-Laird method) was used for all pooled analyses.

• 2.Investigation of Heterogeneity (Meta-regression)

To explore potential sources of heterogeneity, univariable meta-regression analyses were conducted using the restricted maximum likelihood (REML) method with the Knapp-Hartung modification to reduce false-positive rates. The following covariates were pre-specified and tested:

Duration of hemoperfusion (hours); Publication year; Total sample size; Patient etiology (Respiratory/VV-ECMO vs Cardiac/VA-ECMO).

• 3.Sensitivity Analysis and Publication Bias

Sensitivity analyses were performed using the leave-one-out method to evaluate the stability of the pooled results. Potential publication bias was assessed visually using funnel plots and statistically using the Harbord’s modified test (preferred for binary outcomes with low event rates) instead of Egger’s test. Additionally, the non-parametric trim-and-fill method (Duval and Tweedie) was applied to estimate the number of missing studies and to calculate an adjusted effect size.

• 4.Trial Sequential Analysis (TSA)

To control the risk of random errors (Type I and Type II errors) associated with repetitive testing and sparse data, we performed Trial Sequential Analysis for the all-cause mortality [19]. We calculated the Required Information Size (RIS) assuming a two-sided Type I error α of 5%, a Type II error β of 20% (power = 80%), and a relative risk reduction (RRR) of 20% for the intervention. The cumulative Z-curve was constructed to assess whether the evidence was conclusive (crossing the monitoring boundaries) or futile (crossing the futility boundaries).

Results

Search results

A total of 1,093 records were identified through database searching, including 437 from PubMed, 498 from Embase, 46 from the Cochrane Library, 68 from CBM, and 44 from CNKI. After removal of duplicates (n = 175), 918 records were screened by title and abstract, of which 600 were excluded. Subsequently, 318 reports were sought for retrieval, and 23 full-text articles were assessed for eligibility. Among these, 10 studies were excluded because three did not report the primary outcome [20–22] and seven lacked an appropriate control group or did not meet the inclusion criteria [23–29]. Ultimately, 13 studies were included in the systematic review and meta-analysis [30–42]. The study selection process is shown in Fig. 1.

Fig. 1.

PRISMA flowdiagram of study identification for systematic reviews and meta analysis

Description of included studies

A total of 8,151 patients receiving ECMO were included. Five studies [35, 39–42] focused on VV-ECMO, whereas eight studies [30–34, 36–38] involved VA-ECMO. Among the included studies, three RCTs were published in 2021 [40], 2022 [36], and 2025 [30]. Of the 10 retrospective studies, three [37, 38, 41] used propensity score matching and were conducted between 2021 and 2022. One study [35] was based on a nationwide registry. The main indications for ECMO were cardiogenic shock, cardiac arrest requiring extracorporeal cardiopulmonary resuscitation (ECPR), and acute respiratory distress syndrome (ARDS). The cartridges used included HA380 [31, 32], HA330 [34], CytoSorb [33–42], and oXiris [30]. CytoSorb cartridges were generally replaced every 24 h for up to 72 h or until shock reversal. HA330 and HA380 cartridges were typically exchanged every 6 h for a total duration of 18 h, whereas oXiris was administered for 12 h. Adsorption devices were integrated into ECMO-integrated systems [33, 38, 40, 42], parallel ECMO circuits [31, 32, 34, 36, 37, 41], or CRRT-integrated setups [30, 39]. Anticoagulation was achieved with heparin or argatroban. All included studies were published after 2020. Detailed study characteristics are presented in Table 1.

Table 1.

Characteristics of included studies

Study	Country	Study design	Study period	ECMO type	Indication	Cartridge	Connection mode	HP scheme	Anticoagulation	ECMO + HP (n)	ECMO (n)
Ko 2025 [30]	Korea	Retrospective cohort	2018–2023	VA	Cardiogenic shock	oXiris	CRRT–ECMO integrated	Adsorptive CRRT initiated after ECMO cannulation	Heparin	20	20
Hao 2025 [31]	China	Retrospective cohort	2018–2023	VA	Cardiogenic shock	HA380	Parallel ECMO	HP initiated after ECMO; cartridge replaced q6 h	Heparin	39	42
Han 2024 [32]	China	Retrospective cohort	2019–2022	VA	Postcardiotomy cardiogenic shock	HA380	Parallel ECMO	HP within 24 h of ECMO; cartridge q6 h; total 18 h	Heparin	18	24
Lovrić 2024 [33]	Croatia	Retrospective cohort	2017–2018	VA	Cardiogenic shock	CytoSorb	ECMO integrated	Adsorber used during first 24 h of ECMO	NR	9	7
Lesbekov 2022 [34]	Kazakhstan	Retrospective cohort	2021–2022	VA	Postcardiotomy cardiogenic shock	CytoSorb / HA330	Parallel ECMO	Three adsorption sessions (24 h CytoSorb / 6 h HA330)	Heparin	20	10
Heidenreich 2022 [35]	Germany	Nationwide registry	2017–2019	VV	Severe respiratory failure	CytoSorb	NR	Hemoadsorption during ECMO (registry data)	NR	684	7015
Supady 2022 (1)(CYTER) [36]	Germany	Randomized controlled trial	2019–2020	VA	ECPR	CytoSorb	ECMO bypass	Adsorber replaced q24 h for 72 h	Heparin	22	19
Supady 2022(2) (registry) [37]	Germany	Retrospective cohort	2011–2019	VA	ECPR	CytoSorb	ECMO bypass	Adsorption ≥ 72 h; cartridge q24 h	Heparin	23	23
Soltesz 2022 [38]	Hungary	PSM cohort	2012–2020	VA	Refractory cardiogenic shock	CytoSorb	ECMO integrated	Continuous adsorption for 72 h	Heparin	29	29
Akil 2022 [39]	Germany	Retrospective cohort	2020–2021	VV	COVID-19 ARDS	CytoSorb	ECMO or CRRT circuit	Cartridge replaced q24 h; mean 6 sessions	Argatroban	16	10
Supady 2021 [40] (CYCOV)	Germany	Randomized controlled trial	2020	VV	COVID-19 ARDS	CytoSorb	ECMO integrated	Adsorber replaced q24 h for 72 h	NR	17	17
Rieder 2021 [41]	Germany	PSM cohort	2012–2018	VV	ARDS	CytoSorb	Parallel ECMO	Cartridge replaced q24 h; ≥ 72 h	NR	9	9
Akil 2020 [42]	Germany	Prospective cohort + historical control	2018	VV	Pneumogenic sepsis with ARDS	CytoSorb	ECMO integrated	Cartridge replaced q24 h; ≥ 2 sessions	Argatroban	13	7

ARDS, acute respiratory distress syndrome; CRRT, continuous renal replacement therapy; ECPR, extracorporeal cardiopulmonary resuscitation; ECMO, extracorporeal membrane oxygenation; HP, hemoperfusion; NR, not reported; PSM, propensity score matching

Risk of bias assessment

Among the three RCTs, Supady et al. [40] was evaluated to have a low risk of bias across all domains. The remaining two RCTs [30, 36] raised some concerns, primarily related to the randomization process and potential bias in the selection of reported outcomes. For the nine non-randomized studies, four studies [31, 37, 38, 41] were rated as moderate risk of bias. These studies employed propensity score matching to address key confounders (eg, age, severity scores, vasopressor requirements), which considerably mitigated bias arising from confounding and participant selection. The remaining observational studies were evaluated to have a serious risk of bias. The predominant issues in these studies included a lack of control for significant baseline confounders, particularly illness severity [32, 34, 35, 39], and the use of historical control groups [42], leading to significant selection bias. Notably, most studies, both RCTs and non-RCTs, were assessed as having low risk of bias in the domains of intervention classification, deviations from intended interventions, missing data, and outcome measurement. The body of evidence is characterized by a mix of study designs with varying methodological rigor, necessitating a cautious and nuanced interpretation of the pooled results (Figs. 2 and 3).

Fig. 2.

Risk of bias in non-randomized studies of interventions

Fig. 3.

Risk of bias in randomized trials

Primary outcome: all-cause mortality

ECMO combined with hemoperfusion did not show a significant reduction in the ECMO combined with hemoperfusion was not associated with a significant reduction in all-cause mortality compared with ECMO alone (OR 0.98, 95% CI 0.49–1.94; P = 0.95), with substantial heterogeneity (I² = 76%; P < 0.0001) (Fig. 4).

Fig. 4.

Forest plot of all-cause mortality stratified by study design (RCT vs. non-RCT)

Subgroup analysis

RCTs vs. non-RCTs

Three RCTs involving 115 patients demonstrated a significantly higher all-cause mortality with hemoperfusion compared to ECMO alone (OR 4.96, 95% CI 1.67–14.77; p = 0.004), with low heterogeneity (I² = 34%, p = 0.22). Ten non-RCT studies involving 8036 patients showed no significant difference in all-cause mortality (OR 0.58, 95% CI 0.25–1.33; p = 0.20), with high heterogeneity (I² = 79%, p < 0.0001) (Fig. 4).

VA-ECMO vs. VV-ECMO

Eight studies focused on VA-ECMO (354 patients) and five studies on VV-ECMO (7,797patients). In both subgroups, ECMO combined with hemoperfusion showed no significant reduction in the all-cause mortality rate for VA-ECMO (OR 0.80, 95% CI 0.41–1.58; p = 0.53) or VV-ECMO (OR 1.23, 95% CI 0.30–4.99; p = 0.77). Both subgroups showed moderate to high heterogeneity (VA-ECMO: I² = 47%, p = 0.07; VV-ECMO: I² = 77%, p = 0.001) (Fig. 5).

Fig. 5.

Forest plot of all-cause mortality stratified by ECMO type (VA-ECMO vs. VV-ECMO)

Cardiogenic shock vs. cardiac arrest

In VA-ECMO patients, subgroup analysis stratified by indication of cardiogenic shock (6 studies;n = 267) and cardiac arrest (2 studies; n = 87) showed no mortality benefit from hemoperfusion (cardiogenic shock: OR 0.58, 95% CI 0.31–1.09, p = 0.09; cardiac arrest: OR 2.42, 95% CI 0.71–8.25, p = 0.16). Stratification reduced substantial heterogeneity in both subgroups (I² = 22%, and I² = 26%, respectively) (Fig. 6).

Fig. 6.

Forest plot of all-cause mortality stratified by indication (cardiogenic shock vs. cardiac arrest)

Secondary outcomes

Changes in inflammatory biomarkers and vasoactive requirements

To rigorously assess the biological efficacy of hemoperfusion, we first conducted a quantitative meta-analysis focusing on the change in inflammatory biomarker levels from baseline to 72 h post-intervention. By employing the Cochrane methodology to impute missing variances for the change from baseline, we effectively adjusted for initial severity differences [43]. The pooled analysis using the standardized mean difference (Hedges’ g) revealed that, compared to the ECMO-only control group, the addition of hemoperfusion did not result in a significantly greater reduction in circulating IL-6 (p = 0.649), CRP (p = 0.694), or PCT (p = 0.462) over the 72-h period (Fig. 7).

Fig. 7.

Forest plot for changes in inflammatory biomarkers. A: Changes in IL-6. B: Changes in CRP. C: Changes in PCT

While our pooled 72-h data indicated no overall significant reduction, individual studies reported varying dynamic changes at different measurement time points (Table S4). For instance, Hao et al. [31] and Han et al. [32] reported a continuous and statistically significant decrease in IL-6 levels in the hemoperfusion group compared to the control group. Similarly, Ko et al. [30] revealed significant intra-group reductions in IL-6 from baseline to 24 h (p = 0.020) and to 7 days (p = 0.003), although median endotoxin levels at 48 h showed no significant difference between the intervention and control groups (p = 0.097). Conversely, and consistent with our pooled findings, studies by Supady et al. [36, 40] demonstrated that early use of hemoadsorption did not significantly reduce IL-6 levels compared to standard ECMO care.

Regarding hemodynamic stabilization, seven studies [30–33, 37–40] reported decreases in norepinephrine dose or vasoactive-inotropic score within the hemoperfusion group during the treatment period. However, in direct between-group comparisons, at least two studies [30, 35] found no significant differences in vasoactive requirements. Because of substantial clinical and methodological heterogeneity—such as variations in biomarker selection, inconsistent measurement time points beyond 72 h, and varying reporting formats—further quantitative meta-analysis of other time points and vasoactive agent use was not feasible. Extractable numerical data for these outcomes are summarized qualitatively in Supplementary Table S4.

Clinical outcomes

Six studies (n = 236) reported ECMO duration, seven studies (n = 234) reported ICU length of stay, four studies (n = 7,868) reported hospital length of stay, and three studies (n = 140) reported duration of mechanical ventilation. No significant differences were observed between ECMO combined with hemoperfusion and ECMO alone across all outcomes, including ECMO duration (SMD − 0.14, 95% CI − 0.69 to 0.40; p = 0.60; I² = 75%), ICU length of stay (SMD − 0.07, 95% CI − 0.33 to 0.19; p = 0.45; I² = 0%), hospital length of stay (SMD − 0.05, 95% CI − 0.13 to 0.24; p = 0.58; I² = 23%), and duration of mechanical ventilation (SMD − 0.02, 95% CI − 0.36 to 0.31; p = 0.89; I² = 0%) (Fig. 8).

Fig. 8.

Forest plot for clinical course outcomes. A: Forest plot for ECMO duration days. B: Forest plot for ICU stay days. C: Forest plot for hospital-stay days. D: Forest plot for Mechanical ventilation days

Complications (hemorrhage and infection)

Six studies (7,932 patients) reported bleeding events. ECMO combined with hemoperfusion did not significantly change the risk of bleeding compared with ECMO alone (OR, 0.99; 95% CI, 0.33–2.97; p = 0.98), with high heterogeneity (I² = 83%; p < 0.0001) (Fig. 9). Two studies (123 patients) reported infection events. ECMO combined with hemoperfusion did not significantly change the risk of infection compared with ECMO alone (OR, 0.38; 95% CI, 0.07–2.05; p = 0.26), with moderate heterogeneity (I² = 56%; p = 0.13) (Fig. 9).

Fig. 9.

Forest plot for complications. A: Forest plot for: hemorrhage complications. B: Forest plot for infection complications

Investigation of heterogeneity

Meta-regression analysis

Additional analyses are reported in the Supplementary Materials. Univariable meta-regression suggested that patient etiology may contribute to between-study heterogeneity, whereas hemoperfusion duration and publication year were not significantly associated with mortality estimates (Fig. S1-S3).

Sensitivity analysis for all-cause mortality

Leave-one-out sensitivity analysis did not materially alter the primary finding. While extreme estimate shifts occurred when excluding outlier study [35], the core finding of no significant mortality effect remained unchanged. This suggests that the observed heterogeneity originates from multiple sources rather than from a single influential study (Fig. S4).

Publication bias

The funnel plot showed asymmetry, which was statistically significant according to the Harbord test (p = 0.027). However, the trim-and-fill analysis did not impute missing studies (Fig. S5-S7).

Trial sequential analysis (TSA)

Trial Sequential Analysis (TSA) indicated that the cumulative Z-curve did not cross the trial sequential monitoring boundary for benefit or harm, nor did it reach the futility boundary. The required information size was estimated at approximately 19,000 patients. These findings suggest that, despite the relatively large pooled sample size, the current evidence remains statistically inconclusive and further adequately powered trials are needed (Fig. 10).

Fig. 10.

Trial sequential analysis for all-cause mortality

Discussion

Summary of main findings

This meta-analysis included 13 studies involving 8,151 patients, of whom 919 received ECMO plus hemoperfusion and 7,232 received ECMO alone. First, the overall analysis showed no significant reduction in all-cause mortality with hemoperfusion, although substantial heterogeneity was present (I² = 76%). Notably, subgroup analysis restricted to RCTs showed a significant increase in mortality in the intervention group. Second, our quantitative analysis indicated that hemoperfusion did not significantly reduce key inflammatory biomarkers (IL-6, CRP, and PCT) at 72 h. In addition, ECMO plus hemoperfusion did not shorten ECMO duration, ICU or hospital length of stay, or duration of mechanical ventilation. However, consistent with previous safety assessments, hemoperfusion during ECMO was not associated with an increased risk of complications.

The mechanism of hemoperfusion in ECMO patients

Patients receiving ECMO have reported mortality rates of 37%–40% [44, 45] and frequently develop systemic inflammatory responses characterized by excessive cytokine release. Hemoadsorption has therefore been proposed as a strategy to attenuate inflammation by removing circulating proinflammatory mediators. This concept provides the theoretical basis for hemoperfusion therapy. Several adsorption devices have been developed for this purpose, including HA330/380 [46] and CytoSorb [47], which provide nonselective removal of lipophilic molecules ranging from 100 to 55,000 Da. In addition, the oXiris filter is designed to remove endotoxins and cytokines [48].

Potential explanations for the lack of observed mortality benefit

Despite its strong theoretical rationale, hemoperfusion was not associated with improved overall survival. It has been debated whether insufficient treatment “dose” or duration may contribute to this lack of efficacy. However, our exploratory meta-regression did not identify treatment duration as a significant modifier of mortality estimates. One possible explanation is that cytokine adsorption may not translate into consistent systemic biomarker reduction or clinical benefit in heterogeneous ECMO populations. Our pooled biomarker analysis showed no significant between-group reduction in IL-6, CRP, or PCT at 72 h. This lack of biochemical effect may be related to rapid cartridge saturation [49]. In states of profound hyperinflammation, the continuous endogenous release of cytokines from the interstitial space [50] may rapidly exceed adsorptive capacity. Furthermore, currently available cartridges lack specificity and may remove anti-inflammatory mediators as well as therapeutic agents such as antibiotics and vasoactive drugs [51]. Such nonselective clearance may theoretically disrupt immune homeostasis or alter drug exposure, although these mechanisms remain speculative and require prospective validation.

Patient etiology and the need for more refined trial definitions

Patient etiology may be an important hypothesis-generating source of heterogeneity. While previous studies noted high heterogeneity, our exploratory meta-regression suggested that patient etiology may contribute to between-study heterogeneity. In patients with severe respiratory failure (predominantly receiving VV-ECMO), large-scale registry data [35] and rigorous RCT [40] consistently indicated no survival benefit and even potential harm. Conversely, smaller observational studies [33, 34] focusing on patients with cardiogenic shock (predominantly VA-ECMO) tended to report favorable outcomes. This apparent divergence may reflect confounding by etiology, study size, disease severity, ECMO mode, timing of treatment, and co-interventions. Rather than supporting a simple dichotomy between respiratory and cardiac indications, these findings highlight the need for future trials with tighter inclusion criteria. Patients may need to be selected according to more specific disease entities, inflammatory phenotypes, cytokine patterns, timing of hyperinflammation, or endotoxin burden, rather than ECMO mode alone.

Comparison with previous studies

Wenli Li et al. [52] conducted a systematic review of eight studies which found that hemoadsorption in ARDS patients requiring VV-ECMO did not significantly reduce inflammation or mortality. Similarly, Freiburger et al. [51] evaluated hemoadsorption during VA-ECMO, reporting no significant improvements in mortality. Our findings are consistent with these earlier studies but critically extend their scope by directly comparing these two distinct populations within a single, massive cohort of 8,151 patients. Furthermore, similar to another recent meta-analysis [53, 54], we confirmed that hemoadsorption therapy does not increase the risk of adverse complications, providing a comprehensive safety profile across various cartridges.

Limitations and robustness of evidence

This systematic review and meta-analysis has several limitations. There was substantial variation in sample size, cartridge type, and underlying disease across studies. Although exploratory meta-regression suggested that patient etiology may contribute to heterogeneity, this analysis was limited by the small number of studies and should be interpreted as hypothesis-generating. To further assess the robustness of the findings, we applied the nonparametric trim-and-fill method, which did not impute any missing studies, suggesting that publication bias was unlikely to fully explain the primary result. Finally, despite pooled large sample size, Trial Sequential Analysis showed that the cumulative evidence has not crossed the monitoring boundaries and that the required sample size has not been reached. This indicates that the current evidence remains statistically inconclusive. Ongoing prospective studies [55] with refined patient selection and standardized hemoperfusion protocols are needed to better define the role of hemoperfusion in patients receiving ECMO.

Conclusion

This systematic review and meta-analysis found no significant reduction in all-cause mortality with the addition of hemoperfusion to ECMO; mortality was higher in the RCT subgroup. No significant differences were observed in ECMO duration, ICU or hospital length of stay, duration of mechanical ventilation, or complication rates. Although several studies reported transient decreases in inflammatory biomarkers, these effects were inconsistent and did not translate into measurable clinical benefit. Overall, the available evidence does not demonstrate a consistent therapeutic advantage of hemoperfusion during ECMO. Future large, well-designed randomized controlled trials are needed to clarify the optimal timing, duration, and device selection for this intervention. Until such data are available, hemoperfusion should be considered an adjunctive and investigational strategy rather than a routine component of extracorporeal life support.

Abbreviations

ARDS

Acute respiratory distress syndrome

CBM

Chinese biomedical literature database

CI

Confidence interval

CNKI

China national knowledge infrastructure

CRRT

Continuous renal replacement therapy

ECMO

Extracorporeal membrane oxygenation

ECLS

Extracorporeal life support

ECPR

Extracorporeal cardiopulmonary resuscitation

HP

Hemoperfusion

ICU

Intensive care unit

IL-6

Interleukin-6

MD

Mean difference

OR

Odds ratio

PCT

Procalcitonin

PRISMA

Preferred reporting items for systematic reviews and meta-analyses

PSM

Propensity score matching

RCT

Randomized controlled trial

REML

Restricted maximum likelihood

ROBINS-I

Risk of bias in non-randomized studies of interventions

RoB 2.0

Risk of bias 2 tool

SMD

Standardized mean difference

TSA

Trial sequential analysis

VA-ECMO

Venoarterial extracorporeal membrane oxygenation

VIS

Vasoactive inotropic score

VV-ECMO

Venovenous extracorporeal membrane oxygenation

Author contributions

XT, DL, YXT and ZP conceived and designed the study. XT and YCY performed the literature search, study selection, and data extraction. YCY and TH conducted the statistical analysis. XT and YCY drafted the manuscript. ZP, JH and ZJ critically revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

This work was partially supported by the Sichuan Provincial Health Care Commission (No. ChuanJianYan ZH2025-201, awarded to Jun Zeng). The funder had no role in the study design, conduct, analysis, or manuscript preparation.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.