Blood Purification and Mortality in Sepsis and Septic Shock

A Systematic Review and Meta-analysis of Randomized Trials

Alessandro Putzu, M.D.; Raoul Schorer, M.D.; Juan Carlos Lopez-Delgado, M.D., Ph.D.; Tiziano Cassina, M.D.; Giovanni Landoni, M.D.


Anesthesiology. 2019;131(3):580-593. 

In This Article

Materials and Methods

The current systematic review was conducted in compliance with the PRISMA (Preferred Reporting Items Systematic Reviews and Meta-Analysis) guidelines[18] (Supplemental Digital Content, table S1, and Cochrane methodology[19] and according to a prepublished protocol (PROSPERO database, CRD42018104643).

Search Strategy

Two investigators (A.P. and R.S.) independently searched PubMed, the Cochrane Central Register of clinical trials, and Embase up to January 1, 2019, for relevant articles (Supplemental Digital Content, table S2, The search strategy aimed to include any randomized study performed with any type of extracorporeal blood purification technique compared to conventional therapy in adult critically ill patients with sepsis and septic shock. Abstracts from recent international conferences were searched for additional studies. In addition, we hand-scanned references of retrieved articles and pertinent reviews to identify other eligible trials (backward snowballing).

Study Selection

References obtained from searches were first independently examined at the abstract level by two authors (A.P. and R.S.) and then collected as full-text articles if potentially relevant. Eligible studies met the following PICOS criteria: (1) Population: adult critically ill patients with sepsis with or without septic shock; (2) Intervention: any extracorporeal blood purification technique (hemoperfusion, renal replacement therapy techniques, plasmapheresis); (3) Comparison intervention: conventional therapy; (4) Outcome: mortality at longest follow-up available; and (5) Study design: randomized controlled trial. The exclusion criteria were blood purification for renal failure indication at randomization, trials with overlapping populations with a previously included article (e.g., manuscripts with different follow-up or subanalyses of a previously published trial), and pediatric studies. Two authors (A.P. and R.S.) independently assessed selected studies for the final analysis, with disagreements resolved by consensus with a third author (G.L.). If the article did not include data on mortality or was not full-text, the corresponding author was contacted for further data. No language restrictions were imposed.

Data Abstraction

One author (A.P.) extracted relevant information from each selected study. These data were checked by another author (R.S.). Disagreement was resolved by consensus with a third author (G.L.). We specifically extracted potential sources of significant clinical heterogeneity (e.g., study design, clinical setting, inclusion and exclusion criteria, blood purification regimen).

The primary endpoint of this review was mortality at the longest follow-up available, and the secondary endpoint was mortality at 28 to 30 days.

Quality Assessment

Two authors (A.P. and R.S.) independently assessed the internal validity of each included trial according to the Cochrane Collaboration methods.[19,20] We assessed the risk of bias associated with the random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, completeness of outcome data, selective reporting, and other bias. The other bias domain included the classic items reported by the "Cochrane Handbook for Systematic Reviews of Interventions"[19] but also the presence of an intention-to-treat analysis, sample size calculation, and ethical approval of the trial. If one or more of the domains were judged as having a high or unclear risk of bias, we classified the trial as having a high risk of bias. Due to the nature of the intervention, blinding of participants and personnel seemed difficult and was therefore not judged as crucial for bias assessment. We evaluated the potential risk of bias by applying a rating of "Low," "High," or "Unclear" to each study.

Two authors (A.P. and R.S.) independently reviewed the presence of authors' possible conflict of interest and the funding source for each study, then rated each trial as of "Low," "High," or "Unclear" risk regarding those specific points.

The certainty of the body of evidence was assessed using the grading of recommendations assessment, development, and evaluation framework.[21,22] The grading of recommendations assessment, development, and evaluation framework characterizes the certainty of a body of evidence on the basis of study limitations, imprecision, inconsistency, indirectness, and other considerations.

Statistical Analysis

Individual trial and summary results were reported as relative risk with 95% CI. We used a random-effects model except in cases where few trials dominated the available evidence or where significant publication bias was present, as random-effects meta-analysis applied in these contexts may give inappropriately high weight to smaller studies. Statistical heterogeneity was explored by the Cochran Q statistic and characterized using the I2 metric. Publication bias was assessed by visually inspecting the funnel plot for the primary outcome. Statistical significance was set at P = 0.05. The meta-analysis was performed using Review Manager (RevMan, version 5.3; The Nordic Cochrane Center, The Cochrane Collaboration, Copenhagen, Denmark, 2014).

The primary analysis was stratified by blood purification technique: hemoperfusion, hemofiltration, hemoperfusion combined with hemofiltration, or plasmapheresis. Hemoperfusion subgroup analyses including trials on polymyxin B immobilized fiber column hemoperfusion or hemoperfusion with other devices were carried out. To explore the sources of heterogeneity, we performed some subgroup analyses: (1) low risk of bias versus unclear/high risk of bias trials; (2) trials conducted in Asia versus Europe and America; (3) trials from the Nakamura group versus other trials; and (4) trials published after 2010 versus older trials.

To explore the relationship between treatment effect and disease severity, we performed various analyses: (1) a random-effects meta-regression on the APACHE II (Acute Physiology, Age, Chronic Health Evaluation II) score,[23] sepsis-related organ failure assessment score,[24] and control group mortality;[14] (2) subgroup analyses according to conventional therapy group mortality: low-risk group (mortality rate less than 30%), intermediate-risk group (30 to 60%), and high-risk group (greater than 60%).[14] We also performed a meta-regression for age to investigate a possible influence on outcome estimates. Finally, sensitivity analyses were performed by analyzing the data with a fixed or random effects model and using other summary statistics.

We performed a predefined random-effects trial sequential analysis,[25–27] with the intent of maintaining an overall 5% risk of type I error and a 10% risk of type II error. We assumed a relative risk reduction of 15% and derived the control event proportion from the actual dataset. The resulting required information size was further diversity (D2)-adjusted. In case of D2 = 0 we performed a sensitivity analysis assuming a D2 = 25%. We used the trial sequential analysis software (TSA Viewer [Computer program], version Beta, Copenhagen Trial Unit, Center for Clinical Intervention Research, Rigshospitalet, Copenhagen, Denmark, 2016). Deviations from the initial protocol are reported in the supplement (Supplemental Digital Content, eMethods 1,