Intraoperative hemoadsorption and cardiac surgery-associated acute kidney injury: an updated systematic review and meta-analysis with trial sequential analysis

Abstract

Background

Cardiac surgery-associated acute kidney injury (CSA-AKI) following cardiopulmonary bypass (CPB) remains a high-risk complication with limited effective management. Hemoadsorption is increasingly used as an adjunctive therapy due to its potent cytokines clearance in experimental settings, yet its clinical efficacy is debated. This study aimed to evaluate the effect of hemoadsorption versus standard care on CSA-AKI and other major outcomes in adult cardiac surgery patients.

Methods

An updated systematic review and meta-analysis of randomized controlled trials (RCTs) was conducted following PRISMA guidelines. PubMed, Medline, Embase, Web of Science, and the Cochrane Library were systematically searched from inception to 8 February 2025. Eligible RCTs enrolled adult patients undergoing cardiac surgery and compared intraoperative hemoadsorption with standard care, with reported outcomes including CSA-AKI and other major endpoints. Pooled estimates were synthesized using inverse-variance random-effects models, with heterogeneity quantified by I² statistics. Subgroup, sensitivity and trial sequential analyses (TSA) were further performed.

Results

Fifteen RCTs were included, of which nine reported CSA-AKI (947 patients). Hemoadsorption was not associated with a statistically significant reduction in CSA-AKI (RR 0.80, 95% CI 0.63–1.03, P = 0.08, I² = 40%, GRADE: very low). The finding was sensitive to model choice and the inclusion of two studies (Diab 2022 and Abou-Arab 2025). Subgroup analyses revealed no significant interaction by device type. TSA indicated that the required information size was not reached. No significant differences were observed for CSA-AKI Stage 1 (RR 0.72, 95% CI 0.49–1.05, P = 0.09, I² = 16%), Stage 2 (RR 0.66, 95% CI 0.30–1.43, P = 0.29, I² = 11%), Stage 3 (RR 0.43, 95% CI 0.17–1.05, P = 0.06, I² = 0%), or renal replacement therapy (RR 0.52, 95% CI 0.22–1.25, P = 0.15, I² = 0%). Mortality and other clinical endpoints were comparable between groups. Among exploratory outcomes, only an overall reduction in IL-8 was noted (MD -18.23, 95% CI -31.90 to -4.56, P = 0.009, I² = 40%).

Conclusions

Intraoperative hemoadsorption did not significantly reduce CSA-AKI or improve clinical outcomes in adult cardiac surgery. Although trends favored severe AKI, very low certainty evidence and insufficient information preclude definitive conclusions, warranting further large-scale RCTs.

Registration

PROSPERO identifier CRD420250651941.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-026-06099-2.

Introduction

Cardiac surgery-associated acute kidney injury (CSA-AKI) occurs in up to 30% of patients and severely impacts both short- and long-term outcomes [1]. The inflammatory response triggered by cardiopulmonary bypass (CPB) is initiated by contact activation and further amplified by ischemia-reperfusion injury and subsequent endotoxin release [2]. This cascade serves as a key pathogenic mechanism underlying CSA-AKI [3]. Accordingly, inflammation is now recognized as a vital target for therapeutic intervention [4].

Hemoadsorption is an adsorptive blood purification technology designed to eliminate inflammatory mediators and toxins, and may therefore reduce CSA-AKI risk and improve related clinical outcomes [5]. Commercially available devices, including CytoSorb, oXiris, and Jafron HA330/HA380, exhibit distinct adsorption profiles. CytoSorb targets small- to medium-sized hydrophobic molecules up to 60 kDa but does not remove endotoxins. By contrast, oXiris and Jafron provide additional endotoxin adsorption. Jafron covers molecules of 10–60 kDa, while oXiris is limited to molecules up to 35 kDa [5]. Initially applied in critically ill patients with hyperinflammatory conditions [6], hemoadsorption has been increasingly used in cardiac surgery in recent years [7].

Despite the theoretical rationale, clinical evidence regarding the effects of hemoadsorption on renal and other patient-centered outcomes in cardiac surgery remains limited and inconsistent. Most randomized controlled trials (RCTs) have focused on surrogate endpoints, such as plasma-free hemoglobin (pfHb) [8, 9], cytokine levels [10–15], norepinephrine requirements [16], or microcirculatory flow [17] as primary measures, rather than clinically important endpoints including CSA-AKI, renal replacement therapy (RRT), and mortality. Furthermore, the optimal device selection and most suitable surgical setting are still unclear.

To date, few systematic reviews and meta-analyses have been published in this field [18,19]. However, these earlier syntheses have not incorporated all previously published studies [9, 16], nor have they included all recently released RCTs [17, 20]. Given the evolving evidence base and persistent clinical uncertainty [4], we conducted an updated synthesis to evaluate the impact of intraoperative hemoadsorption on CSA-AKI and other major clinical outcomes in adult patients undergoing cardiac surgery with CPB, aiming to provide a more comprehensive understanding of its clinical efficacy.

Methods

Registration and protocol

This study was conducted in accordance with the Cochrane Handbook [21] and the PRISMA 2020 reporting guidelines [22]. The review protocol was prospectively registered with PROSPERO (CRD420250651941) on 24 February 2025, before data extraction and formal analysis commenced. Although the protocol initially considered both RCTs and observational studies, only RCTs utilizing intraoperative hemoadsorption were ultimately included to reduce confounding and ensure high evidence quality.

Eligibility criteria

Study eligibility was defined according to the PICOS framework: RCTs (S) of adult cardiac surgery patients (P) comparing intraoperative hemoadsorption (I) with standard care (C) and reporting CSA-AKI incidence or other major clinical outcomes (O) were included. Vascular surgeries involving isolated thoracoabdominal aortic aneurysm repair [23] were considered ineligible.

Search strategy

Systematic searches of PubMed, Medline, Embase, Web of Science (core collection), and the Cochrane Library were performed from inception to 8 February 2025. Comprehensive search terms related to “hemoadsorption” and “cardiac surgery” were used, with database-specific adaptations. Reference lists of relevant studies, reviews, and included articles were also manually screened to identify additional eligible trials. The full search strategies are provided in the Supplementary Methods S1.

Study selection

All retrieved records were imported into Endnote X9.3.3 for literature management and duplicate removal. Two independent reviewers (ML and DL) screened titles, keywords, and abstracts to identify potentially eligible trials. Full-text articles were then assessed independently against the predefined eligibility criteria. Any discrepancies were resolved by consensus, with a third reviewer (RZ) providing arbitration when necessary.

Data extraction

Two reviewers (ML and DL) independently extracted data from eligible studies using a pre-piloted standardized data collection form in Microsoft Excel. Extracted information included publication year, authors, country, study period, sample size, baseline patient characteristics, intervention details (device type, blood flow rate, circuit configuration, and treatment duration), intraoperative variables, predefined outcomes and study funding sources.

Outcomes

The primary outcome was the incidence of CSA-AKI diagnosed according to the Kidney Disease: Improving Global Outcomes (KDIGO) or Acute Kidney Injury Network (AKIN) criteria. Studies reporting only nonspecific worsening renal function were excluded from the primary analysis [15]. Secondary outcomes included the stage-specific CSA-AKI incidence (Stage 1, 2 and 3), requirement for RRT or dialysis, mortality (in-hospital, 28/30-day, and the longest available follow-up), duration of mechanical ventilation, length of intensive care unit (ICU) stay, length of hospital stay, postoperative duration of any vasopressor support, delirium, stroke, and arrhythmias. Several outcomes were considered exploratory due to heterogeneous measurement time points or a limited number of trials (fewer than three trials), including 90-day mortality, myocardial infarction, Sequential Organ Failure Assessment (SOFA) scores, other vasopressor/inotrope-related outcomes (vasopressor/inotrope utilization, cumulative doses), and circulating inflammatory markers (IL-10, TNF-α, IL-6, IL-8, C-reactive protein, and procalcitonin).

Risk of bias assessment

Two reviewers (ML and DL) independently assessed the risk of bias for each included RCT using the Cochrane Risk of Bias 2.0 (ROB-2) tool [24]. Disagreements were resolved by discussion or adjudicated by a third reviewer (RZ). Each trial was judged as low risk, some concerns, or high risk of bias.

Statistical analysis

All analyses were conducted using RevMan 5.4.1 (The Cochrane Collaboration, 2020). A two-sided P < 0.05 was regarded as statistically significant. For dichotomous outcomes, the pooled effects were estimated using risk ratios (RRs) with 95% confidence intervals (CIs). For continuous outcomes, mean differences (MDs) with 95% CIs were calculated. When data were reported on different scales, a validated online conversion tool [25] was applied following the methods [26, 27] described by Wan et al. and Luo et al. Studies with obvious data errors or insufficient raw data for quantitative synthesis were excluded. Considering anticipated clinical heterogeneity, all pooled estimates were derived using the inverse-variance random-effects model. Heterogeneity across trials was quantified by the I² statistic, with I² > 50% indicating substantial heterogeneity. Meta-analysis results were illustrated with forest plots.

Subgroup and sensitivity analysis

Predefined subgroup analyses were performed according to hemoadsorption device type. Formal surgical subgroup analysis was abandoned due to imbalanced group distribution and limited trial quantity. Interaction tests were conducted only when both subgroups contained no fewer than three trials; otherwise, results were reported as descriptive exploratory comparisons. To assess the robustness of pooled estimates, sensitivity analyses were undertaken by switching to a fixed-effect model, excluding small-sample trials (fewer than 30 participants per arm), and performing leave-one-out analysis (only for analyses with ≥ 3 RCTs) to identify influential individual studies.

Trial sequential analysis

Trial sequential analysis (TSA) was conducted using TSA Viewer (Copenhagen Trial Unit) [28] to control the risk of type I statistical error. The required information size (RIS) was estimated with a two-sided α of 5%, statistical power of 80%, and a predefined 20% relative risk reduction. Pooled event rates in control participants were used to define the baseline risk. Heterogeneity was adjusted using model-variance-based diversity (D²), and the O’Brien-Fleming α-spending function was applied to construct trial sequential monitoring boundaries.

Assessment of certainty of evidence and publication bias

The certainty of evidence for each outcome was rated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) framework [29]. Two reviewers (ML and DL) independently evaluated the evidence quality across five GRADE domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Given that the primary CSA-AKI analysis included only nine RCTs, publication bias was assessed only by visual inspection of funnel plot symmetry, as formal statistical tests are considered unreliable with fewer than ten studies [30].

Results

Study selection

The PRISMA flow diagram summarizing the study selection process is presented in Fig. 1. A total of 1564 records were identified through database searching, of which 647 duplicate entries were removed. The remaining 917 records underwent title and abstract screening, leading to the exclusion of 873 irrelevant citations. The subsequent 44 articles were assessed in full text, among which 29 were excluded for failing to satisfy the PICOS eligibility criteria. Exclusions included three trials previously identified in another meta-analysis [19]; these were ruled out due to enrolment of patients undergoing thoracoabdominal aortic aneurysm repair [23], use of a lipopolysaccharide-specific adsorber [31], or initiation of hemoadsorption intraoperatively with continuation into the postoperative period [32]. Ultimately, fifteen eligible RCTs comprising 1190 patients were included in the final analysis.

Fig. 1.

PRISMA flow diagram for study screening and selection

Study characteristics

Details of study and population characteristics are summarized in Table 1. The fifteen RCTs published between 2016 and 2025 included twelve single-center [8, 10–17, 33–35], one two-center [20], and two multi-center studies [9, 36], spanning 12 countries across Europe, North America, and Asia. Mean patient age ranged from 50 to 75 years, with male proportions varying from 40% to 100%. Median CPB duration exceeded 120 min in all included studies, ranging from 127 to 228 min. All enrolled patients were at high inflammatory risk. Ten trials recruited patients undergoing complex elective cardiac surgery with prolonged CPB [9–15, 17, 20, 33], two included those with infective endocarditis [16, 36], two involved aortic surgery cohorts [8, 35], and one enrolled heart transplant recipients [34]. Hemoadsorption was administered intraoperatively during CPB in all trials. One trial used two parallel-configured CytoSorb cartridges integrated into a CPB side-circuit, with a total flow rate of 600 mL/min initiated 1 h after CPB commencement [9]. The other fourteen trials adopted a single column (oXiris, CytoSorb, or Jafron) with flow rates set between 200 and 500 mL/min; notably, five trials did not report or monitor intraoperative device flow [14, 16, 33, 34, 36]. All control groups received standard care without adjunctive hemoadsorption.

Table 1.

Characteristics of included trials

Trial	Study design	Population	HA type and using time	HA blood flow rate, mL/min	N	Age, years	Male, n (%)	CPB time, min	Primary outcome
Abou-Arab 2025, France [17]	Single- center	Elective cardiac surgery with expected CPB duration > 90 min	oXiris, during CPB	450	HA: 33 Ctrl: 35	HA: 68 (61,71) Ctrl: 68 (60,74)	HA: 25 (76%) Ctrl: 27 (77%)	HA: 127 (97,154) Ctrl: 128 (97,154)	Microcirculation change from baseline at Day 1
Bao 2024, China [8]	Single- center	Sun’s procedure for acute type A aortic dissection	HA380 Jafron, during CPB	200–250	HA: 40 Ctrl: 44	HA: 51.13 ± 8.35 Ctrl: 54.75 ± 11.41	HA: 29 (72.5%) Ctrl: 32 (72.7%)	HA: 197.8 ± 51.87 Ctrl: 205.35 ± 39.70	Changes in pfHb from baseline to 48 h postoperatively
Condello 2024, Italy [33]	Single- center	Elective cardiac surgery with expected CPB duration > 120 min	HA380 Jafron, during CPB	NA	HA: 10 Ctrl: 10	HA: 61 ± 7 Ctrl: 65 ± 5	HA: 4 (40%) Ctrl: 6 (60%)	HA: 128 ± 6 Ctrl: 123 ± 5	Postoperative AKI incidence (AKIN)
Diab 2022, Germany [36]	Multi- center	Infective endocarditis, with EuroSCORE II > 3	CytoSorb, during CPB	NA	HA: 138 Ctrl: 144	HA: 68.5 (58,76) Ctrl: 69 (60,76)	HA: 98 (71%) Ctrl: 111 (77%)	HA: 128 (93,173) Ctrl: 120 (82,167)	∆SOFA
Gleason 2019, America [9]	Multi- center	Elective complex cardiac surgery with expected CPB duration ≥ 3 h	CytoSorb, 1 h after CPB initiation	600 (two cartridges, a total flow)	HA: 23 Ctrl: 23	HA: 66 ± 8 Ctrl: 61 ± 17	HA: 18 (78%) Ctrl: 13 (56%)	HA: 198 ± 66 Ctrl: 228 ± 78	pfHb change from presternotomy to end CPB
Holmén 2022, Sweden [16]	Single- center	Infective endocarditis	CytoSorb, during CPB	NA	HA: 10 Ctrl: 9	HA: 72 (62,74) Ctrl: 70 (64,76)	HA: 9 (90%) Ctrl: 8 (89%)	HA: 137 (88,184) Ctrl: 192 (93,277)	24 h and 48 h postoperative norepinephrine dose
Nemeth 2024, Hungary [34]	Single- center	Orthotopic heart transplantation	CytoSorb, during CPB	NA	HA: 30 Ctrl: 25	HA: 56 (47,61) Ctrl: 56 (48,60)	HA: 15 (50%) Ctrl: 15 (60%)	HA: 133 (116,154) Ctrl: 129 (104,169)	Early postoperative haemodynamic instability
Pérez- Fernández 2024, Spain [20]	Two- center	Elective cardiac surgery with expected CPB duration > 90 min	oXiris, during CPB	200–250	HA: 169 Ctrl: 174	HA: 68.8 ± 9 Ctrl: 68.6 ± 10	HA: 113 (67%) Ctrl: 111 (64%)	HA: 132 ± 46.7 Ctrl: 127 ± 39.2	CSA-AKI incidence (KIDGO)
Poli 2019, Switzerland [14]	Single- center	Elective cardiac surgery with expected CPB duration > 120 min	CytoSorb, during CPB	NA	HA: 15 Ctrl: 15	HA: 67 (44,76) Ctrl: 69 (49,80)	HA: 13 (86.7%) Ctrl: 11 (73.3%)	HA: 145 (130,183) Ctrl: 138 (87,207)	Perioperative cytokine changes
Bernardi 2016, Austria [10]	Single- center	Elective cardiac surgery with expected CPB duration > 120 min	CytoSorb, during CPB	200	HA: 19 Ctrl: 18	HA: 64 (30,81) Ctrl: 69 (51,81)	HA: 12 (63.2%) Ctrl: 14 (77.8%)	HA: 191 (112,288) Ctrl: 170 (83,274)	Differences in the evolution of cytokines
Garau 2019, Germany [11]	Single- center	Elective cardiac surgery with expected CPB duration > 120 min	CytoSorb, during CPB	300	HA: 20 Ctrl: 20	HA: 67.9 ± 12.7 Ctrl: 72.7 ± 9.2	NA	HA: 141.2 ± 41.1 Ctrl: 138.6 ± 40.1	Cytokines and procalcitonin levels before CPB, 0 h, 6 h and 24 h after CPB
Hohn 2024, Germany [13]	Single- center	Elective cardiac surgery with expected CPB duration > 90 min, age > 65 years	CytoSorb, during CPB	400	HA: 19 Ctrl: 19	HA: 74.1 ± 5.9 Ctrl: 75.1 ± 5.4	HA: 11 (57.9%) Ctrl: 13 (68.4%)	HA:161.8 ± 107.4 Ctrl: 163.4 ± 61.7	Differences in mean IL-6 levels at ICU admission and during ICU stay
He 2022, China [12]	Single- center	Elective combined valve replacement with expected CPB duration > 2 h	HA380 Jafron, during CPB	200–300	HA: 30 Ctrl: 30	HA: 54.37 ± 9.53 Ctrl: 55.43 ± 9.90	HA: 14 (46.7%) Ctrl: 18 (60.0%)	HA:155.87 ± 30.24 Ctrl: 159.43 ± 34.19	Efficacy on eliminating inflammatory factors
Taleska Stupica 2020, Slovenia [15]	Single- center	Elective complex cardiac surgery with expected CPB duration > 90 min	CytoSorb, during CPB	400	HA: 20 Ctrl: 20	HA: 70.5 (34,80) Ctrl: 71 (31,85)	HA: 14 (70%) Ctrl:14 (70%)	HA:146 (91,281) Ctrl: 127 (90,238)	Evolution of cytokine levels and complement C5a, expression of CD64 and CD163 markers
Wagner R 2019, Czech Republic [35]	Single- center	Aortic root operation (Ross and David)	CytoSorb, during CPB	300–500	HA: 15 Ctrl: 13	HA: 50 ± 10 Ctrl: 54 ± 15	HA: 13 (86.7%) Ctrl:13 (100%)	HA:198 ± 15 Ctrl: 182 ± 44	Myocardial, monocyte and vascular miRNAs plasma levels

The data are presented as mean ± standard deviation and/or median (interquartile range, Q1-Q3) as extracted from the article. All studies listed were RCTs, and control groups received standard care (CPB without HA). AKI, acute kidney injury; AKIN, AKI Network; CPB, cardiopulmonary bypass; CSA-AKI, cardiac surgery-associated AKI; Ctrl, control; HA, hemoadsorption; ICU, intensive care unit; KIDGO, Kidney Disease: Improving Global Outcomes; N, number of patients; NA, not available; pfHb, plasma-free hemoglobin; RCT, randomised controlled trial; SOFA: Sequential Organ Failure Assessment

Risk of bias assessments

Of the fifteen RCTs, three demonstrated a low overall risk of bias across all domains [13, 20, 36], six presented some concerns [11, 12, 14, 16, 17, 34], and six were rated as high risk [8–10, 15, 33, 35]. Studies categorized as some concerns primarily showed limitations in randomization procedures and selective reporting of outcomes. A high overall risk of bias was predominantly attributable to selective outcome reporting. Detailed domain-level risk of bias judgments for individual trials are provided in Supplementary Figures S1A–B.

Primary outcome: CSA-AKI

Nine trials involving 947 patients reported the incidence of CSA-AKI [8, 9, 14, 16, 17, 20, 33, 34, 36]. Pooled analysis under a random-effects model showed that intraoperative hemoadsorption was not associated with a significant reduction in CSA-AKI compared with standard care (RR 0.80, 95% CI 0.63–1.03, P = 0.08), with moderate heterogeneity (I² = 40%) (Fig. 2).

Fig. 2.

Forest plot for CSA-AKI

TSA was performed to quantify the statistical reliability and sufficiency of the accumulated evidence for CSA–AKI. The required information size was calculated to be 4335 participants, yet the cumulative sample size included in this meta-analysis (n = 947) remained substantially below this threshold, clearly indicating that the current body of evidence was statistically underpowered. The cumulative Z-curve did not cross the trial sequential monitoring boundaries, signifying that the existing dataset was insufficient to draw definitive conclusions (Fig. 3).

Fig. 3.

Trial sequential analysis for CSA-AKI

Subgroup analysis for CSA-AKI

Subgroup analysis stratified by device type revealed no significant between-subgroup difference in treatment effect. Outcomes were comparable for studies using CytoSorb (5 trials, RR 0.80, 95% CI 0.52–1.22, I² = 50%, P = 0.29) versus other devices (4 trials, RR 0.78, 95% CI 0.56–1.09, I² = 23%, P = 0.15), with no significant subgroup interaction (P = 0.95). By surgical type, formal statistical inference was not feasible for the corresponding stratified analysis; therefore, these results were considered exploratory. The pooled RR was 0.74 (95% CI 0.56–0.98, I² = 23%) for the non-infective endocarditis subgroup (7 trials) and 0.94 (95% CI 0.57–1.52, I² = 17%) for the infective endocarditis subgroup (2 trials).

Details are shown in Supplementary Figures S2A-B and Table S1. A predefined subgroup analysis based on treatment duration was not undertaken, as all included studies restricted intervention delivery exclusively to the intraoperative phase.

Sensitivity analysis for CSA-AKI

To assess the robustness of the primary findings, three sensitivity analyses were conducted. First, reanalysis using a fixed-effect model yielded a marginally significant reduction in CSA–AKI (RR 0.86, 95% CI 0.74–1.00, P = 0.05, I² = 40%). Second, excluding small trials with fewer than 30 participants per study arm generated a pooled effect consistent with the primary analysis (RR 0.89, 95% CI 0.70–1.14, P = 0.36, I² = 42%). Third, leave-one-out analysis indicated that the pooled estimate was influenced by two individual trials. Exclusion of either Diab 2022 or Abou-Arab 2025 shifted the pooled effect toward statistical significance. In addition, omitting Diab 2022 or Nemeth 2024 substantially reduced inter-study heterogeneity, with I² decreasing to 14% and 13%, respectively (Table S1).

Secondary outcomes

Formal meta-analysis was restricted to outcomes reported by at least three RCTs.

Renal-related outcomes

Forest plot summaries and TSA findings for secondary renal endpoints are presented in Fig. 4 and Supplementary Figures S3A-D. Three RCTs reported CSA–AKI incidence stratified by KDIGO stage [8, 20, 34], while five trials documented the need for RRT [8, 13, 14, 20, 34]. Pooled estimates showed no significant intergroup differences across AKI severity stages: Stage 1 (RR 0.72, 95% CI 0.49–1.05, P = 0.09, I² = 16%), Stage 2 (RR 0.66, 95% CI 0.30–1.43, P = 0.29, I² = 11%), and Stage 3 (RR 0.43, 95% CI 0.17–1.05, P = 0.06, I² = 0%). No significant difference was detected for RRT requirement (RR = 0.52, 95% CI 0.22–1.25, P = 0.15, I² = 0%). Although point estimates suggested a trend toward greater clinical benefit with advancing AKI severity, none of these renal-related outcomes reached statistical significance.

Fig. 4.

Forest plot for CSA-AKI risk by severity and RRT requirement

TSA for Stage 1 CSA-AKI failed to achieve the required information size of 4588 participants, and the cumulative Z-curve remained within the trial sequential monitoring boundaries. For Stage 2 CSA-AKI, Stage 3 CSA-AKI, and RRT, TSA modelling was not feasible owing to the limited number of available studies or low event rates.

Mortality

No significant between-group differences in mortality were observed across all follow-up time points (Supplementary Figure S4A). Pooled analyses yielded the following results: in-hospital mortality (3 RCTs, RR 1.27, 95% CI 0.22–7.36, P = 0.79, I² = 19%), 28/30-day mortality (7 RCTs, RR 1.00, 95% CI 0.67–1.50, P = 0.99, I² = 0%), and mortality at the longest available follow-up (11 RCTs, RR 1.04, 95% CI 0.71–1.52, P = 0.84, I² = 0%). Leave-one-out sensitivity analysis further confirmed that the pooled mortality estimates remained consistent and stable after sequential exclusion of each individual trial (Table 2).

Table 2.

Secondary outcomes of RCTs assessing the effect of hemoadsorption in patients undergoing cardiac surgery

Outcome	N.Of studies	Relative effect (95%CI)	I2 (%)	P value	Leave-one-out analysis
Mortality in hospital	128 (3 RCTs)	RR = 1.27 (0.22, 7.36)	19	0.79	Consistent1
Mortality at 28/30-days	871 (7 RCTs)	RR = 1.00 (0.67, 1.50)	0	0.99	Consistent1
Mortality at the longest follow-up available	1029 (11 RCTs)	RR = 1.04 (0.71, 1.52)	0	0.84	Consistent1
Mechanical ventilation time (h)	589 (9 RCTs)	MD = −1.08 (−7.99, 5.82)	71	0.76	Consistent1
Length of ICU stay (d)	1170 (14 RCTs)	MD = −0.44 (−1.04, 0.16)	69	0.15	Omitting Garau turns P (overall effect) to 0.04.
Length of hospital stay (d)	970 (10 RCTs)	MD = −0.14 (−1.00, 0.72)	0	0.75	Consistent1
Postoperative duration of any vasopressor support (h)2	394 (4 RCTs)	MD = −2.20 (−7.76, 3.35)	5	0.44	Consistent1
Delirium	418 (4 RCTs)	RR = 0.56 (0.26, 1.19)	5	0.13	Consistent1
Stroke	748 (5 RCTs)	RR = 0.80 (0.36, 1.78)	0	0.59	Consistent1
Arrhythmias	534 (6 RCTs)	RR = 0.98 (0.74, 1.31)	0	0.90	Consistent1

CI, confidence interval; ICU, intensive care unit; MD, mean difference; NA, not applicable; RCT, randomized controlled trial; RR, risk ratio; ¹ The results were consistent and not dependent on any individual study. Leave-one-out sensitivity analysis was performed only when the number of studies was ≥ 3; for outcomes with fewer than three studies, this analysis was not applicable. ² To facilitate meta-analysis, data reported as “postoperative 24 h,” “Day 1,” “ICU stay,” or “postoperative” were considered equivalent and labeled “postoperative.” Data for any vasopressor (unspecified) and noradrenaline-specific duration were included as reported in the original trials

Other clinically relevant outcomes

Additional clinical endpoints, including duration of mechanical ventilation, intensive care unit stay, hospital length of stay, duration of postoperative vasopressor support, delirium, stroke, and arrhythmias, were also evaluated. Pooled analyses revealed no significant clinical benefit of intraoperative hemoadsorption for any of these outcomes. Leave-one-out analysis for intensive care unit stay showed that excluding the trial by Garau shifted the pooled result toward statistical significance, whereas all other outcomes remained robust after sequential omission of individual trials (Table 2 and Supplementary Figures S4B–H).

Exploratory outcomes

Outcomes reported in only two RCTs or with substantial heterogeneity in assessment timing were defined as exploratory. Most relevant clinical and surrogate measures, including 90-day mortality, myocardial infarction, SOFA scores, vasopressor/inotrope-related parameters (vasopressor/inotrope utilization, cumulative doses), and circulating inflammatory markers (IL-10, TNF-α, IL-6, C-reactive protein and procalcitonin), showed no significant differences between groups. However, a significant overall reduction in IL-8 levels was observed (MD −18.23, 95% CI −31.90 to −4.56, P = 0.009, I² = 40%), although no individual time point subgroup reached statistical significance (all P > 0.05), with no evidence of subgroup interaction (P for interaction = 0.92). Full exploratory outcome data are provided in the Supplementary Materials (Figures S5A-J, Tables S2–S4).

Publication bias

Potential publication bias for the primary CSA-AKI endpoint was assessed by visual inspection of funnel plot symmetry. The funnel plot generated from the nine included trials showed an approximately symmetrical distribution of effect sizes around the pooled estimate (RR 0.80), with no obvious asymmetry indicating substantial publication bias (Supplementary Figure S6).

Quality of evidence assessment

The GRADE certainty assessment rated the overall evidence as very low for CSA–AKI and all secondary renal-related outcomes (Table S5). Evidence downgrading was attributed to serious risk of bias, serious indirectness, and serious imprecision across included trials, as well as concerns regarding potential reporting bias.

Discussion

Our updated meta-analysis suggests that intraoperative hemoadsorption during cardiac surgery is not associated with a significant reduction in CSA-AKI nor improvements in any clinically relevant endpoints. Nevertheless, TSA confirmed insufficient cumulative evidence, and sensitivity analyses revealed unstable pooled effect estimates. Accordingly, these findings do not support any definitive therapeutic conclusion regarding the use of intraoperative hemoadsorption.

The main finding indicated that intraoperative hemoadsorption did not reduce CSA-AKI relative to standard care, despite nominal trends in risk ratios suggesting potential benefit with advancing AKI severity. However, it must be emphasized that TSA, where applicable, failed to meet the required information size. This limitation may stem from the fact that most included RCTs (7/9) [8, 9, 14, 16, 17, 34, 36] were primarily designed around surrogate endpoints rather than AKI, rendering renal outcomes secondary, underpowered, or inconsistently assessed.

Substantial clinical heterogeneity across the included trials further limits the interpretability of a single pooled estimate, even under random-effects modeling. Leave-one-out sensitivity analyses identified two high-risk patient cohorts as major sources of heterogeneity: infective endocarditis (Diab 2022) [36] and heart transplantation (Nemeth 2024) [34]. Their omission reduced I² from 40% to 14% and 13%, respectively. To further explore whether treatment effects varied across clinical scenarios, we attempted subgroup analyses by device type and surgical context, including elective prolonged CPB, infective endocarditis, acute aortic dissection, and heart transplantation. These populations differ considerably in inflammatory burden, perioperative risk, and baseline susceptibility to AKI [3, 8, 34, 37]. For device-based subgroups, no significant difference was noted between CytoSorb and other devices. For surgical settings, however, such an analysis could not be performed because each category contained too few trials to generate reliable estimates [21]. Consequently, it remains unclear whether hemoadsorption confers no overall clinical benefit or whether certain surgical subgroups may still attain measurable therapeutic advantages.

Interestingly, hemoadsorption failed to yield the anticipated benefit in the infective endocarditis subgroup, despite its theoretical rationale. This may be attributable to low statistical power, with only two eligible trials available [16, 36], and to the fact that pathogen source control and antimicrobial therapy, rather than cytokines adsorption, remain the cornerstone of infective endocarditis management [37]. Moreover, hemoadsorption may act as a “double-edged sword” by increasing the clearance of certain antibiotics, such as fluconazole and vancomycin, thereby risking subtherapeutic drug concentrations [38]. Dose adjustment and therapeutic drug monitoring should be considered in this setting.

Sensitivity analyses further confirmed that the primary finding lacked statistical robustness. Although the fixed-effects model yielded a marginally significant result, the random-effects model provided more conservative estimates amid moderate heterogeneity. The overall effect was also sensitive to the inclusion of two individual studies: the largest neutral trial (Diab 2022) exerted the highest statistical weight (28.2%), while a smaller trial (Abou-Arab 2025) reported a contradictory estimate (RR 1.24). Exclusion of either trial rendered the pooled estimate statistically significant. As such, available evidence cannot confirm a clinically meaningful role of intraoperative hemoadsorption in CSA-AKI prophylaxis.

Our results are consistent with the RCT subgroup analysis reported by Salles et al. [19], which comprised six RCTs and detected no apparent benefit (RR 0.78, 95% CI 0.50–1.20). Such a trend contrasts with another meta-analysis that suggested a protective effect [18]. The discrepancy may be partly explained by methodological differences, specifically, we adopted a random-effects model, excluded a non-cardiac surgery trial [23], and incorporated two additional eligible RCTs [9, 17].

Compared with previous meta-analyses, the present study provides several complementary contributions to the existing literature. We updated the evidence base by including recently published trials not incorporated in earlier syntheses, restricted the analysis exclusively to RCTs to reduce selection bias, and adhered to a pre-specified protocol to avoid post-hoc data dredging. We also conducted comprehensive sensitivity and exploratory subgroup analyses to assess the stability, heterogeneity, and contextual variability of treatment effects across studies. The GRADE framework was used to evaluate the certainty of evidence for all renal endpoints, facilitating transparent interpretation and graded clinical recommendations. Critically, TSA was applied to assess the adequacy of cumulative sample size and mitigate type I error inflation associated with repeated meta-analytic testing, supporting more restrained and reliable conclusions than conventional meta-analysis alone.

The biological rationale for hemoadsorption in mitigating CSA-AKI is rooted in the interconnected pathogenic cascades triggered by CPB. Contact between blood components and synthetic bypass circuits initiates a marked systemic inflammatory activation, driving excessive release of pro-inflammatory mediators (TNF-α, IL-6, IL-8, HMGB1) and complement fragments (C3a, C5a). These inflammatory disturbances promote glycocalyx degradation, elevate microvascular permeability, and augment leukocyte adhesion [2]. Concurrently, oxidative stress induced by CPB and cardiac ischemia-reperfusion injury provokes mitochondrial dysfunction and tubular epithelial cell apoptosis. Intestinal barrier disruption further facilitates endotoxin translocation, which synergizes with circulating cytokines to worsen renal hypoperfusion [2, 3, 39]. Through hydrophobic, ionic, and size-exclusion properties of specialized sorbent resins, hemoadsorption non-selectively sequesters pro-inflammatory mediators, damage-associated molecular patterns, oxidative byproducts, and free hemoglobin [5, 8, 9, 14, 33, 35, 40, 41]. This intervention interrupts the self-perpetuating pathogenic cycle at multiple vulnerable nodes—a mechanism unattainable with conventional hemodynamic support or diuretic therapy. Furthermore, hemoadsorption devices can be directly integrated into the CPB circuit [5], allowing an earlier therapeutic window for real-time clearance of pathogenic mediators prior to the development of irreversible renal parenchymal injury.

Nevertheless, current evidence from our analysis does not support a clear patient-centered benefit of routine intraoperative hemoadsorption. Several factors account for the discrepancy between its theoretical rationale and clinical efficacy. First, CSA-AKI is a heterogeneous disorder driven by multiple non-inflammatory mechanisms that hemoadsorption cannot readily address, including hypotension, hypoperfusion, neurohormonal activation, and baseline kidney disease [3]. In high-risk patients with pre-existing renal impairment, inflammation plays only a secondary role, making cytokines adsorption insufficient to reverse established tubular injury. Notably, most included trials enrolled patients with CKD stages 1–3 without baseline renal function stratification, which may dilute potential subgroup-specific treatment effects. Second, broad-spectrum hemoadsorption sequesters both pro- and anti-inflammatory mediators, potentially disrupting immune homeostasis [40]. Although targeted devices such as polymyxin B-immobilized columns (PMX) improve mortality in endotoxic septic shock via potent endotoxin binding, as demonstrated in the Tigris trial [42], relevant controlled data in cardiac surgery and CSA-AKI populations remain lacking. Hence, PMX could not be included in our analysis. Third, adsorption efficiency is concentration-dependent [5] and time-limited. Intraoperative application during CPB may miss the postoperative inflammatory peak [13] and predispose to cytokine rebound, consistent with our exploratory findings of inconsistent cytokine suppression and occasional postoperative IL-6 elevation. Technical limitations including adsorber saturation, desorption and competitive binding by plasma proteins may further diminish net mediator clearance [13]. Furthermore, baseline cytokine levels were not uniformly reported across studies, and biomarker measurements varied in timing, hindering standardized comparative analysis. Finally, the lack of biomarker-guided enrollment across heterogeneous trial cohorts may obscure variable treatment responses shaped by individual inflammatory profiles and perioperative risk burden [43].

Among exploratory analyses of surrogate endpoints, only an overall reduction in IL-8 was identified. Although no single time point subgroup yielded a significant IL-8 reduction, the pooled overall estimate reached statistical significance, likely driven by enhanced statistical precision from integrating consistent directional trends across multiple time points with low between-subgroup heterogeneity. Several studies also reported modulatory effects on microcirculatory disturbances and endothelial injury [17, 41], yet the limited number of such studies precluded formal meta-analytic pooling. Overall, the present exploratory analyses could not establish a consistent association between intraoperative hemoadsorption and sustained cytokines clearance, nor could we demonstrate tangible improvements in other clinical endpoints.

Our study has several limitations. First, six studies were rated at high risk of bias, mainly owing to inadequate randomization and selective outcome reporting. Second, most trials had relatively small sample sizes, with marked clinical and methodological heterogeneity in surgical procedures, hemoadsorption device settings, AKI diagnostic criteria and flow rates. Such heterogeneity undermines the generalizability and robustness of the pooled findings. Third, reporting of secondary and exploratory outcomes was incomplete and measurement time points varied across trials, restricting further analysis and interpretation of these endpoints. Although a potential gradient effect across AKI severity strata was observed, the small number of studies per stage (three each) and overlapping confidence intervals rendered this finding merely hypothesis-generating. Fourth, several predefined subgroup analyses were underpowered due to insufficient studies per stratum, and planned dose-response analyses based on hemoadsorption duration were not feasible.

Conclusions

In conclusion, intraoperative hemoadsorption did not significantly reduce CSA-AKI or improve clinical outcomes in adult cardiac surgery. Although point estimates suggested a potential gradient of benefit with increasing AKI severity and an exploratory overall reduction in IL-8 levels, the available evidence remains insufficient to support definitive conclusions. All findings were derived from very low certainty evidence, with TSA verifying an insufficient cumulative sample size. Further large-scale, adequately powered RCTs are therefore warranted.

Abbreviations

AKIN

Acute Kidney Injury Network

CI

Confidence interval

CPB

Cardiopulmonary bypass

CSA-AKI

Cardiac surgery-associated acute kidney injury

GRADE

Grading of Recommendations Assessment, Development and Evaluation

ICU

Intensive care unit

KDIGO

Kidney Disease: Improving Global Outcomes

MD

Mean difference

pfHb

Plasma-free hemoglobin

PICOS

Population-Intervention-Comparison-Outcome-Study design

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

RCT

Randomized controlled trial

RIS

Required information size

ROB-2

The Cochrane Risk of Bias 2.0

RR

Risk ratios

RRT

Renal replacement therapy

SOFA

Sequential Organ Failure Assessment

TSA

Trial sequential analysis

Author contributions

ML and RZ conceived the study. ML and DL conducted the literature search and data extraction. ML performed the statistical analysis and drafted the manuscript. RZ critically revised the manuscript and provided supervision. All authors contributed to the interpretation of results, reviewed the manuscript, and approved the final version.

Funding

This work was supported by grants from the 1·3·5 Project for Disciplines of Excellence-Clinical Research Fund, West China Hospital, Sichuan University (2024HXFH019) and the Science & Technology Agency of Sichuan Province, China (2024YFFK0253).

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.