FormalPara Take-home message

The use of extracorporeal liver support devices in patients with ALF and ACLF may improve mortality and hepatic encephalopathy. Future studies are needed to confirm these results and to determine which modality is most effective.

Introduction

Liver failure may occur with or without underlying liver disease. Acute liver failure (ALF) occurs without underlying chronic liver disease, and usually causes jaundice, coagulopathy, encephalopathy, and can progress to multi-organ failure and death [1, 2]. While some patients recover with supportive care, the definitive treatment for those who do not recover is liver transplantation, which is expensive and limited by the availability of organs. Extracorporeal liver support (ECLS) offers a potential option for bridging to transplantation or allowing longer time for recovery [3]. The concept behind the use of ECLS is to remove the hepatotoxic substances such as cytokines, vasoactive substances, endotoxins from gut flora, and low molecular weight toxins [2]. However, the contradicting results from previous literature have limited its use [3,4,5,6,7,8,9,10,11]. Although it can be used as a bridging therapy to transplant, it is unclear if ECLS improves survival among patients with ALF who are not candidates for liver transplantation.

ECLS systems are based on dialysis techniques to remove toxic substances such as nitric oxide, prostaglandins, reactive oxygen species, and pathogen-associated molecular patterns that may play a role in liver failure pathogenesis. Artificial systems use cell-free techniques for plasma filtration either by dialysis or exposure to an exchange medium such as charcoal [12]. Commonly used artificial systems include Molecular Adsorbent Recirculating System (MARS, Gambro, Lund, Sweden) and fractionated plasma separation and adsorption (SEPAD; Prometheus, Fresenius Medical Care GmbH, Bad Homburg, Germany), hemofiltration and plasma exchange [12]. On the other hand, bio-artificial systems use either human-based liver cells (e.g., ELAD, Vital Therapies Inc., San Diego, California, USA) or porcine liver cells (e.g., HepatAssist, Arbios, formerly Circe, Waltham, Massachusetts, USA). Besides detoxification of aforementioned substances, bio-artificial systems may have an additional benefit by supporting metabolic and synthetic liver function [13]. However, none of these modalities is designed to assist in the other major liver function of immune modulation [14].

ALF is defined as hepatic encephalopathy (HE) that occurs within 8–28 days from the onset of jaundice, with a high incidence of cerebral edema and a poor prognosis without liver transplantation [15]. Acute on chronic liver failure (ACLF), is distinct from ALF in which patients have pre-existing chronic liver diseases. The Asian Pacific Association for the Study of the Liver defines ACLF as “an acute hepatic insult manifesting as jaundice (serum bilirubin ≥ 5 mg/dl (85 micromol/l) and coagulopathy (INR ≥ 1.5 or prothrombin activity < 40%) complicated within 4 weeks by clinical ascites and/or encephalopathy in a patient with previously diagnosed or undiagnosed chronic liver disease/cirrhosis” [16]. The European Association for the Study of the Liver and the American Association for the Study of Liver Diseases define ACLF as “acute deterioration of pre-existing, chronic liver disease, usually related to a precipitating event and associated with increased mortality at 3 months due to multi-system organ failure” [17].

Several factors can affect the prognosis of patients with ALF or ACLF. For those listed for liver transplantation, the mortality rate is 29% for patients with ALF, and up to 48% for patients with ACLF [18]. In the North American Consortium for the Study of End-stage Liver Disease (NASCELD) study, the mortality was 40% in ACLF patients, and was as high as 77% in those with additional organ failures [19, 20]. The clinical course of ACLF is variable, spontaneous resolution can be as high as 50% in the absence of organ failure, and only 15% in patients with multi-organ failure [21].

The impact of ECLS on clinical outcomes of patients with ALF or ACLF is unclear. Therefore, we conducted a systematic review and meta-analysis of randomized trials to determine the efficacy and safety of artificial or bio-artificial ECLS modalities in patients with liver failure [22].

Methods

Study protocol

We registered the study protocol with the International Prospective Register of Systematic Reviews (PROSPERO; ID CRD42018080201). We followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines [23].

Study selection

Eligible studies met the following criteria: (1) the study design was a randomized controlled trial (RCT); (2) the population were adults with ALF or ACLF; (3) the interventions were any form of artificial or bio-artificial ECLS; (4) the control group received supportive care not including ECLS; (5) the outcomes were all-cause mortality or liver-related mortality, bridging to liver transplant, improvement of HE and adverse events such as hypotension, bleeding, thrombocytopenia, line infection, and citrate toxicity. All outcomes were assessed at the longest follow-up reported in the studies.

Search strategy and data extraction

We searched MEDLINE, EMBASE and Cochrane Central Register of Controlled Trials (CENTRAL) from inception through March 13, 2019 [Electronic Supplemental Material (ESM) Tables 1, 2)]. We assessed citations for eligibility without language, date or type of publication restrictions. In addition, we screened references of relevant articles to identify additional citations. Two reviewers (FA and EB) independently and in duplicate screened titles and abstracts for full-text review and evaluated the full-text articles for eligibility. Two reviewers (FA and BA) also, independently and in duplicate, extracted relevant data from eligible studies using a standardized form. We attempted to contact study authors to obtain missing data. Disagreements were resolved through discussion or a third arbitrator.

Risk of bias assessment

Two reviewers (JD and KA), independently and in duplicate, assessed the risk of bias of individual trials using the Cochrane Collaboration Risk of Bias tool [24]. Reviewers judged trials to be at low, unclear or high risk of bias for each domain. Reviewers deemed the overall risk of bias for individual trials low if all domains were at low risk, unclear if at least one domain was unclear, but no domain was at high risk of bias, and high risk if any domain was at high risk of bias.

Statistical analysis

We used RevMan software (Review Manager, version 5.3. Copenhagen: The Nordic Cochrane Center, The Cochrane Collaboration, 2014) for data analysis. We used the DerSimonian and Laird random-effects model to pool the weighted estimates across studies [25] and the inverse variance method to estimate study weights. For dichotomous outcomes, we report pooled relative risk (RR) with corresponding 95% confidence interval (CI). We defined significant statistical heterogeneity using Chi2P < 0.10 or I2 > 50% [26].

We used the Cochrane Collaboration method to calculate the number needed to treat (NNT) [27]. Based on recent observational studies, we used an assumed control risk (ACR) of 25% and 40% for mortality in ALF and ACLF, respectively [19, 20, 28]. For outcomes with over ten studies we inspected funnel plots visually to assess for publication bias and used Egger’s test to assess for publication bias [29].

We performed predetermined subgroup analyses to explore whether specific factors influenced treatment effects. Pre-specified subgroup analyses were artificial versus bio-artificial treatment modalities and low versus high and unclear risk of bias studies. In addition, post hoc subgroup analyses by type of liver failure (ALF versus ACLF) and funding source were performed. We performed a post hoc sensitivity analysis excluding trials published as abstracts only. For subgroup analyses, we tested for interaction using a χ2 significance test [30].

Finally, we performed a post hoc trial sequential analysis (TSA) to explore the risk of random errors in cumulative meta-analyses [31,32,33,34]. Trial sequential monitoring boundaries adjust the Z score (P value) for significance each time a trial is added to the meta-analysis (i.e., accounting for multiple testing and accrued information). We considered a cumulative Z curve that is greater than the trial sequential boundary a significant effect. Thus, if cumulative Z curve crossed trial sequential significance boundary, we inferred that the intervention is superior to control, even if sample size did not reach required meta-analysis sample size. We aimed to maintain an overall 5% risk of a type I error and a power of 80%. For the required information size (RIS) calculations we used a relative risk reduction (RRR) of 20%, and user-defined incidence rates estimated from all included trials in the conventional meta-analyses for mortality (45.95%) and HE (45.6%) outcomes.

Assessment of quality of evidence

We used the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach to classify the certainty of evidence into high, moderate, low, or very low for each outcome [35]. Well-conducted RCTs provide high certainty but can be downgraded based on the following five domains: risk of bias, inconsistency, indirectness, imprecision and reporting bias.

Results

Our search identified 1068 records. After removing duplicates, 944 records remained. Of those, we excluded 873 irrelevant records. We assessed the remaining 71 full-text articles and further excluded 46 articles. We included 25 studies (enrolling 1796 patients) that met our eligibility criteria (ESM Fig. S1) [36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60].

Thirteen RCTs enrolled patients with ALF [37,38,39, 41,42,43, 45, 48, 52, 53, 56, 57, 59] and 13 RCTs enrolled patients with ACLF [36, 40, 44,45,46,47, 49, 50, 54, 55, 58,59,60]. The average age across all studies was 44 years, and males constituted 59% of all patients. The most common etiologies for ALF were alcohol, viral hepatitis and acetaminophen toxicity. Nineteen trials used artificial ECLS [36,37,38, 41, 42, 44,45,46, 48,49,50, 52,53,54,55,56,57,58,59] and only five trials used bio-artificial ECLS [39, 40, 43, 47, 60]. Trials were mainly from USA, Europe, and Asia. Among artificial systems, MARS (Teraklin AG, Rostock, Germany) [36, 41, 44, 46, 51, 54, 57, 58] was the most commonly used followed by Biologic-DT (HemoCleanse, Inc., West Lafayette, IN, USA) [42, 48, 49, 53, 59], FPSA (Prometheus, Fresenius Medical Care Deutschland GmbH 61346 Bad Homburg v. d. H. Germany) [50, 51], plasma exchange with hemoperfusion [45], whole blood exchange [56] and charcoal hemoperfusion [37]. Bio-artificial modalities included extracorporeal liver assist device (ELAD, Vital Therapies Inc., San Diego, CA, USA) [40, 47, 60] and HepatAssist (Circe Biomedical Inc., Lexington, MA, USA) [39]. Funding was from a combination of academia and industry in 16 trials [36,37,38, 41, 43, 45, 46, 49, 51,52,53,54,55,56,57,58] and from industrial sources in 9 trials [39, 40, 42, 44, 47, 48, 59, 60]. We present characteristics of included trials in Table 1.

Table 1 Characteristics of included studies
Full size table

Risk of bias assessment

Of the 25 studies, 21 were published as full articles and 4 as abstracts. We did not consider lack of blinding of participants as high risk of bias since it is impossible to ensure blinding and the outcomes were objective; therefore, less likely to be affected by lack of blinding [61]. Fourteen studies were adjudicated as overall low risk of bias [36, 37, 39, 44, 46, 49,50,51,52, 54, 55, 57, 58, 60], 10 were adjudicated as overall unclear risk of bias [40,41,42,43, 45, 47, 48, 53, 56, 59] and 1 adjudicated as overall high risk of bias [38]. We present the details of risk of bias assessment in ESM Fig. S2 and Table 4.

Assessment of quality of the evidence

We present the details of our assessment of the certainty of evidence for each outcome according to the GRADE approach in Table 2.

Table 2 GRADE evidence profile
Full size table

Main outcomes

Mortality

Twenty-four RCTs enrolling 1778 patients reported on mortality [36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60]. The use of ECLS probably reduces mortality (RR 0.84; 95% CI 0.74, 0.96, P = 0.01, I2=33%, moderate certainty) (Fig. 1). Publication bias was not detected by visually inspecting funnel plot and by Egger’s test (P = 0.417) (ESM Fig. S3).

Fig. 1
figure 1

Forest plot for mortality outcome

Full size image

Bridging to liver transplant

The data reported in individual trials were either incomplete or not reported; therefore, we were not able to perform a meta-analysis for this outcome.

Hepatic encephalopathy

Twelve RCTs enrolling 417 patients reported on HE [36, 38, 42,43,44,45,46, 48, 49, 53, 58, 59]. The use of ECLS may improve HE compared to usual care (RR 0.71; 95% CI 0.60, 0.84, P < 0.0001, I2 0%, low certainty) (Fig. 2). We downgraded the certainty evidence by one point as publication bias was suspected by visually inspecting funnel plot and by Egger’s test (P = 0.041) (ESM Fig. S4).

Fig. 2
figure 2

Forest plot for hepatic encephalopathy outcome

Full size image

Adverse events

Nine RCTs enrolling 748 patients reported on hypotension [39, 42,43,44, 46, 49, 51, 55, 60]. The effect of ECLS on the risk of hypotension was uncertain [RR 1.46; 95% CI (0.98, 2.2), P = 0.07, I2 = 15%, low certainty]. Eleven RCTs enrolling 1031 patients reported on bleeding, with little to no difference between the two groups (RR 1.21; 95% CI 0.88, 1.66, P = 0.25, I2 = 31%, moderate certainty) [36, 42,43,44, 46, 48,49,50, 55, 57, 60]. Five RCTs enrolling 564 patients reported on thrombocytopenia [39, 44, 51, 57, 60]; the use of ECLS was associated with increased risk of thrombocytopenia (RR 1.62; 95% CI 1.0, 2.64, P = 0.05, I2 = 62%, very low certainty). Only one RCT with 16 patients reported on line infections (RR 1.92; 95% CI 0.11, 33.44, P = 0.65, low certainty) [46] (Fig. 3). None of the included trials reported on citrate toxicity.

Fig. 3
figure 3

Forest plot for adverse events

Full size image

Subgroup analyses

We conducted four subgroup analyses, the first was by type of liver failure (ALF versus ACLF). Thirteen RCTs enrolled 738 patients with ALF (RR 0.90; 95% CI 0.75, 1.08, P = 0.27, I2= 25%) [37,38,39, 41,42,43, 45, 48, 52, 53, 56, 57, 59] and 13 RCTs enrolled 1040 patients with ACLF (RR 0.78; 95% CI 0.66, 0.93, P = 0.006, I2= 30%) [36, 40, 44,45,46,47, 49, 50, 54, 55, 58,59,60]. Although the estimates for ALF subgroup were imprecise, the interaction test did not suggest a subgroup difference (P = 0.28) for mortality (ESM Fig. S5). The second analysis was by type of ECLS (artificial versus bio-artificial liver support). Nineteen trials (1308 patients) (1308 patients) [36,37,38, 41, 42, 44,45,46, 48,49,50, 52,53,54,55,56,57,58,59] and five trials (470 patients) [39, 40, 43, 47, 60] used artificial and bio-artificial liver support systems, respectively. We did not find a subgroup difference for mortality (P = 0.55) (ESM Fig. S6). The third subgroup analysis was by risk of bias (low versus unclear and high). Eleven trials (1096 patients) were at low risk of bias [36, 37, 39, 44, 46, 49, 54, 55, 57, 58, 60] and 13 trials (682 patients) were at unclear risk of bias [38, 40,41,42,43, 45, 47, 48, 50, 52, 53, 56, 59]. The risk of bias did not significantly influence risk of death (P = 0.80) (ESM Fig. S7). The fourth subgroup analysis was by funding source (industry versus academic), this analysis was decided post hoc. Nine trials (692 patients) [39, 40, 42, 44, 47, 48, 50, 59, 60] were industry funded, and 15 trials (1086 patients) [36,37,38, 41, 43, 45, 46, 49, 52,53,54,55,56,57,58] were funded by academic sources. We observed no significant subgroup differences (ESM Fig. S8). We present all subgroup analyses in the supplement.

Sensitivity analysis

A sensitivity analysis excluding four studies published in abstract form [40, 41, 47, 53] yielded similar results as the primary analyses for mortality [RR 0.87; 95% CI (0.75, 1.00), P = 0.05, I2=37%, moderate certainty] (ESM Fig. S9), and HE (RR 0.70; 95% CI 0.58, 0.84, P = 0.0006, I2= 0%, low certainty) outcomes (ESM Fig. S10). We performed a post hoc sensitivity excluding the study by Zhou et al. as the control group composed of patients who declined consent to any of the two intervention arms (plasma exchange versus plasma exchange with albumin dialysis) [38]. The results remained similar to primary analyses for both mortality and hepatic encephalopathy (ESM Figs. S11, S12).

Trial sequential analysis

For mortality outcome; post hoc TSA concurs with the conventional analysis and provides a reliable estimate that the use of ECLS is associated with a reduced mortality risk compared to control (TSA-adjusted RR 0.84, 95% CI 0.73, 0.97), and that the RIS has been achieved (ESM Fig S13). Whereas, for HE outcome, post hoc TSA showed that the cumulative Z score crossed the adjusted boundaries for benefit (Z > 1.96), the RIS has not been reached (39%), indicating inconclusive benefit for ECLS in the reduction of HE using (TSA-adjusted RR 0.71 95% CI 0.57, 0.89) (ESM Fig. S14). Power of 80% and RRR of 20% were used for TSA.

Discussion

This systematic review and meta-analysis of 25 RCTs provides moderate certainty evidence on reduction of mortality with ECLS. In ALF and ACLF patients, ECLS may reduce mortality by 16%, which translates into 74 fewer deaths per 1000 patients, and an NNT of 22 and 16 in ALF and ACLF population, respectively. The effect on mortality was more prominent with artificial devices than with bio-artificial devices, and in ACLF than in ALF population. In addition, our results show that ECLS may reduce HE in patients with liver failure.

Although we included all types of liver failure in this review, there are differences in pathophysiology, causes, and prognosis within this population. Unlike ACLF patients, ALF patients have no pre-existing liver disease [21, 28]. Most common causes of ALF include acetaminophen toxicity (46.3%), indeterminate (12.2%) and other drugs (10.8%) in a recent report from the United States [28], compared to bacterial infection (39.1%) and alcohol (22.9%) among 417 cases of ACLF in Europe [21]. In addition, liver transplantation might not be an option for many patients with ACLF due to advanced age, active alcohol intake, other comorbidities and associated organ failure [21]. ALF patients are more likely to get liver transplants sooner than ACLF patients [57, 62]. Furthermore, ACLF patients had higher overall mortality than ALF patients (260 versus 188 per 100 waitlist-years) [62]. This could explain the larger mortality reduction in ACLF population, as they have a higher baseline risk of death, less likely to receive definitive therapy with liver transplant and a smaller chance of spontaneous resolution (15–50%) [21].

The use of ECLS devices appeared to be safe, we did not observe a significant increase in the risk of adverse events. However, thrombocytopenia was more common with ECLS, but the certainty of data was very low. Finally, no studies reported on citrate toxicity.

Although there are several published meta-analyses on this topic, we included more trials (25 total, 24 for mortality) than any previous meta-analyses [3,4,5,6,7,8,9,10,11, 63] (ranging between 4 and 19), which improved the precision of our findings.

Our findings have important limitations. The duration of follow-up for mortality outcome varied between studies (Table 1), although the statistical heterogeneity was below our pre-specified threshold, it is a potential source of clinical heterogeneity. The results of post hoc TSA (power 80% and RRR 20%) revealed potential imprecision with wider adjusted 95% CI for HE outcome as the RIS has not been achieved. In addition, several studies were industry funded, raising concerns about potential bias. Although our post hoc subgroup analysis did not support this possibility, subgroup analyses are often underpowered. Reporting of some outcomes such as liver transplant was not clear. It was not possible to determine the number of patients listed for liver transplant in each of studies. In addition, the population was heterogeneous with different causes of liver failure, and individual patient data were not available for various subgroups. Further studies are needed to identify which subgroups would benefit the most from ECLS. Lastly, ECLS devices are expensive and are not available at most centers, given the additional direct costs of the intervention, policymakers need to better understand the cost-effectiveness of ECLS in liver failure before making it available for routine use in practice. Hessel et al. studied the cost-effectiveness in a cohort of 149 patients with ACLF, of which 67 (44.9%) were treated with MARS and found that the incremental cost per life year gained was 30% less for MARS [64]. However, data on cost-effectiveness of other modalities are lacking.

Conclusions

Our meta-analysis shows that ECLS may reduce death and improve HE in patients with liver failure. Before ECLS can be routinely used in practice, future RCTs are needed to determine the magnitude of effect, the most effective modality, and the subgroup that would benefit the most from ECLS.