Continuous renal replacement therapy without anticoagulation in critically ill patients at high risk of bleeding: a systematic review and meta-analysis

Abstract Background: Continuous renal replacement therapy (CRRT) has been widely used in the critical care setting and anticoagulation is usually necessitated. However, critically ill patients are commonly at incremental risk of bleeding, which contributed to the hesitation of anticoagulant use for CRRT in clinical practice. The current guideline recommended CRRT proceed without anticoagulation in patients with contraindication to citrate and increased bleeding risk. Nevertheless, the ecacy of anticoagulation-free CRRT remains inconsistent. Therefore, the purpose of our present systematic review is to evaluate the ecacy and safety of anticoagulant-free CRRT based on the current literatures. Methods: We conducted a comprehensive search of PubMed (US National Library of Medicine, Bethesda, MD, USA), Cochrane Library databases and EMBASE from database inception to January 12, 2019 for potential candidate studies. Studies included adult critically ill (age > 18 years) patients with increased bleeding risk, and underwent CRRT without anticoagulation were considered for the inclusion. Results: Finally, 17 observational studies and 3 randomized controlled trials with 1615 patients were included in our present meta-analysis. There was no signicant difference in lter lifespan between the anticoagulation-free and systemic heparin group. The lter lifespan was signicantly prolonged in the citrate (WMD -23.01, 95%CI [-28.62, -17.39], P < 0.001; I 2 = 0%, P = 0.53) and nafamostat (WMD -8.4, 95%CI [-9.9, -6.9], P < 0.001; I 2 = 33.7%, P = 0.21) groups, compared with anticoagulation-free group. The averaged lter lifespan of the anticoagulation-free CRRT ranged from 10.2 to 52.5 hours. Conclusion: The lter lifespan in anticoagulation-free patients with increased bleeding risk was comparable to that in patients without increased bleeding risk underwent systemic heparin anticoagulation CRRT. Nafamostat was not recommended for CRRT anticoagulation due to its drawbacks. Currently, the optimal choice of anticoagulation strategy for critically ill patients without citrate contraindications at high risk of bleeding should be regional citrate anticoagulation. Further studies should focus on the special cut-off value of activated partial thromboplastin time (APTT), international normalized ratio (INR) and platelet (PLT) count, at which the anticoagulation-free CRRT would be benecial. anticoagulation CRRT. Compared to the anticoagulation-free protocol, could signicantly the lter lifespan without incremental bleeding and citrate-related metabolic advantage on lter lifespan were weighed out by its drawbacks and was not for anticoagulation of CRRT. Currently, the optimal choice of anticoagulation strategy for critically ill patients without citrate controindications at high risk of bleeding should be RCA. High bleeding risk patients with low APTT and INR, high PLT and bilirubin, or usage of MV requiring CRRT could have procoagulant tendency and should be treated with appropriate anticoagulation measures. Further studies should focus on the special cut-off value of APTT, INR and PLT, at which the anticoagulation-free CRRT would be feasible and benecial.

free protocol was not reported; (3) no su cient data on the endpoints and outcomes of the anticoagulation-free CRRT were available; (4) animal experiments; (5) intermittent RRT; (6) the following article types: review, case report, letter, conference abstract, and commentary.

Study quality assessment
The risk of bias in the randomized controlled trials were assessed by the use of "the Cochrane collaboration's tool for assessing risk of bias," [16] which including 6 items: (1) allocation concealment; (2) blinding of participants and personnel; (3) blinding of outcome assessment; (4) incomplete outcome data; (5) selective reporting and (6) other bias. For observational studies, a modi ed version of the Newcastle-Ottawa Scale [17] was employed for the study quality assessment (Additional le 1: Table S1). According to the 8 items of this scale, three aspects of study quality were evaluated: subject selection, comparability of cohorts, and assessment of outcomes. Two investigators (WZ and MB) independently evaluated the studies and disagreements were resolved through discussion and consensus.

Study selection and data extraction
Two reviewers (WZ and YY) screening the identi ed studies independently. The article type, title, and abstract were reviewed rst. Thereafter, the studies that passed initial screening underwent full-text review for the nal exclusion. Two of our authors (XLC and LJZ) independently extracted the following data using pre-determined forms: (1) characteristics of the included studies (e.g. rst author, publication year, study design, setting, exclusion criteria, sample size, interventions, and number of lters); (2) characteristics of the patients (e.g. demographic characteristics, severity, diagnosis on admission, baseline coagulation parameters); and (3) information related to the safety and e cacy of anticoagulation-free strategy. Disagreements were resolved by discussion and consensus. The observed outcomes included lter lifespan, bleeding complications, mortality, coagulation parameters, and renal and liver function.

Statistical analysis
For continuous variables, mean (95% CI [con dence interal]) was converted to mean ± standard deviation (SD) by using the calculator attacted in the Review Manager (RevMan) [Computer program]. Version 5.3. Copenhagen: The Nordic Cochrane Centre, the Cochrane Collaboration, 2014. The median (interquartile range [IQR] or range) was converted to mean ± SD by using the online calculator [18], which was programmed according to the methods reported by Wan, X. et al. [19] and Luo, D. et al. [20].
Weighted mean difference (WMD) and risk ratio (RR) (or odds ratio [OR] where appropriate) were pooled for continuous variables and categorical variables, respectively. Heterogeneity across the included studies was evaluated by the I 2 statistic and chi-square test. A I 2 value of more than 50% was considered as signi cant heterogeneity. The potential source of heterogeneity was explored by meta-regression and sensitivity analysis. Subgroup analysis was performed to minimize the heterogeneity among the studies with different characteristics. A P value less than 0.05 was considered as statistical signi cance. P value was assessed using random effects model for I 2 > 50% (P < 0.01) and xed effects model for I 2 ≤ 50% (P > 0.01) [21]. All meta-analyses were performed using the Stata software version 12.0 (Stata Corporation, College Station, TX, USA).

Evaluation of publication bias
The Begg's [22] and Egger's [23] tests were used to evaluate the potential publication bias for primary outcomes.
All studies included critically ill patients with acute kidney injury (AKI) requiring CRRT and a majority of the included patients had high risk of bleeding. Sepsis (43.8%, 381/870 patients) was the most common cause of AKI. The most common indications for CRRT were uid overload and electrolyte imbalance.
The de nitions of high risk of bleeding of the included studies were listed in Table 2. The most frequently used coagulation parameters for the identi cation of high bleeding risk were APTT, PLT, and INR. Eight studies [14,[26][27][28][29][30][31]35] reported the causes of increased risk of bleeding (Table 3). The reported tools for illness severity evaluation included Acute Physiology and Chronic Health Evaluation (APACHE) II, Sequential Organ Failure Assessment (SOFA), Simpli ed Acute Physiology Score (SAPS) II, and SAPS III score.

Quality of studies
The bias risk of the 3 randomized controlled trials [24,39,40] were illustrated in Figure 2. Allocation concealment and blinding were not performed in all of these studies, which perhaps were attributed to the nature of the intervention and the critical condition. The results of the methodological quality assessment of the 17 observational cohort studies using the Newcastle-Ottawa Scale were detailed in Additional le 3: Table S3. Eight studies [14, 30-32, 34-36, 38] were awarded 6 scores and were considered to be moderate quality study. The remaining 9 studies [25-29, 33, 37, 41, 42] earned 7 scores, which were considered to be high quality study.

Anticoagulation-free versus systemic heparin
Both the anticoagulation-free and heparin groups included critically ill patients requiring CRRT and were not signifcinatly different in age, gender, and sepsis.
However, the patients in the anticoagulation-free group had either developed bleeding tendency (i.e. prolonged coagulation parameters or thrombocytopenia) or already been on active bleeding before CRRT. Therefore, the comparability between anticoagulation-free and heparin group was poor.
Anticoagulation-free versus regional citrate anticoagulation (RCA) The baseline characteristics between anticoagulation-free and RCA groups were comparable in 3 [30,37,41] of the 5 related studies. The patient baseline characteristcs, especially the coagulation parameters, of the remaining 2 studies [29,32] were not comparable because the anticoagulation-free group included liver failure patients and the citrate group did not.
Only 1 study [35] reported the proportion of patients with liver disease, and the two groups was comparable at this parameter. The data of APTT and PLT were pooled in 2 [38,40] Figure 5C) between these two groups.

Risk factors of lter lifespan
The risk factors of lter lifespan were reported in 7 studies [14,27,31,32,36,38,40] and were summarized in Additional le 5: Table S5 and Additional le 6: Table S6. The most frequently included variables in the anlaysis of lter lifespan were coagulation parameters and CRRT protocol. In univariate analysis, low APTT, high PLT count, thrombelastogram (TEG) K value, and high bilirubin level were reported to be the risk factors of lter failure. Low APTT, high PLT, low circuit INR, low blood ow rate, high uid removal, high bilirubin level, and mechanical ventilation (MV) were reported to be the risk factors of lter failure in multivariate analysis. Of note, none of the aforementioned analyses were performed exclusively in anticoagulation-free group.

Publication bias
The baseline characteristics between anticoagulation-free and heparin groups were not comparable, therefore, the publication bias test, which otherwise was not effective, was not performed. Due to the small numbers of included studies, publication bias tests were not performed in studies that employed RCA or nafamostat protocol.

Discussion
Our present systematic review found out that (1) the most common causes of bleeding risk in critically ill patients who underwent anticoagulation-free CRRT were coagulopathy and post-surgery; (2) compared with systemic heparin anticoagulation in cricitcally ill patients without increased bleeding risk, anticoagulation-free CRRT in the patients with high bleeding risk yielded similar lter lifespan; (3) In patients at high risk of bleeding requiring CRRT, the lter lifespan was signi cantly prolonged in RCA group compared with anticoagulation-free group, the rate of citrate accumulation in RCA groups was relatively low and no citrate induced metabolic complications were reported; and (4) nafamostat was superior to anticoagulation-free protocol in terms of lter lifespan, however, this drug was used as anticoagulant for CRRT dominantly in Japan and Korea and was not recommended partly due to it's potential side effects.

Anticoagulation-free versus heparin
The overall mortality in the present systematic review was consistent with the results reported by Nash, D. M. et al [43], who summarized that the overall ICU mortality in critically ill patients underwent CRRT was more than 50%. Higher mortality in the anticoagulation-free group, most likely, was not directly caused by bleeding per se, which only re ected the pathophysiology of the underlying disease. Lauzier, F. et al. [44] reported that the ctitically ill patients with major bleeding compared with those without had a twofold increased risk of ICU mortality, despite the underlying pathophysiology leading to death remains unclear.
According to the inclusion criteria, a majority of patients in anticoagulation-free group had developed coagulopathy such as thrombocytopenia or prolonged APTT/INR before CRRT, which are non anti-coagulant factors with a positive association with lter lifespan [45]. In addition, the probable heparin resistance in the critically ill patients, especially Antithrombin III (AT III) de ciency, could attenuate the anticoagulant effect of heparin [46,47]. A combination of the aforementioned conditions might contribute to the similarity of lter lifespan between anticoagulation-free and heparin groups.

Anticoagulation-free versus citrate
In the KDIGO guideline [10], anticoagulation-free protocol was recommended for the critically ill patients requiring CRRT with increased bleeding risk and contraindications to citrate (i.e. liver failure and shock with skeletal muscle hypoperfusion).
Of the 5 studies that compared RCA with anticoagulation-free protocol in our systematic review, 2 studies [30,41] routinely employed anticoagulation-free CRRT in high bleeding risk patients until the availability of custom-made citrate-based replacement uid, 2 studies [29,32] employed anticoagulation-free protocol because of bleeding tendency or liver failure, and the remaining 1 study [37] included post-cardiac surgery patients with AKI and high risk of bleeding, who initially underwent anticoagulation-free CRRT and switched to RCA-CRRT if the lters clotted within 24 hours. In fact, most of these patients in the anticoagulation-free arms, except for a small number of liver failure patients, could theoretically tolerate and bene t from RCA-CRRT. Of note, all of the 5 studies were conducted before the issue of the 2012 KDIGO guideline. In addtion, new evidence suggested that, under strict protocol and colsely monitoring, RCA might be safe for CRRT in liver failure patients [48].
The averaged lter lifespan (range 33-49.8 hours) of RCA-CRRT in the present systematic review were consistent with the results (range 29.5-78 hours) in other studies [49][50][51][52][53]. The prolongation of lter lifespan and reduction of clotting events in the RCA-CRRT were also reported in previously published literatures [52,54,55]. In our systematic review, both the citrate and anticoagulation-free groups were with increased bleeding risk, the prolonged lter lifespan were most likely attributed to the use of anticoagulants [4].
Citrate accumulation was reported in 2 studies [30,37] and both of which did not exclude patients with liver failure, a pathological condition that could theoretically impair citrate metabolism and increase the risk of citrate accumulation [56,57]. The incidence of citrate accumulation in these 2 studies (5% and 3%, respectively) were slightly higher than that (2.99%) reported by Khadzhynov, D. et al. [58] in a large retrospective study which also did not exclude liver failure patients. However, patients with hypoperfusion shock, the other contraindication to citrate and a risk factor of citrate accumulation [59,60], were not excluded from the RCA groups in most of these 5 studies.

Anticoagulation-free versus nafamostat
Nafamostat, a inhibitor of serine proteases, can also act as an anticoagulant by inhibiting coagulation factors (including thrombin, Xa and XIIa). It has short duration of action in blood due to a 5-8 minutes elimination half-life and has the theoretical advantage of extracorporeal elimination coupled with reduced systemic anticoagulation [61]. This drug, dominantly available in Japan and Korea, had been safely used in the critically ill with high bleeding risk in some observational studies [62][63][64]. The reported lter lifespan of nafamostat-anticoagulated CRRT ranged from 22 to 25.5 hours and the bleeding incidence ranged from 4% to 6.6% [65,66]. In our systematic review, the lter lifespan was signi cantly prolonged in nafamostat group (range 19-31 hours) and the bleeding incidence was not increased, compared with anticoagulation-free group. However, because of the absence of antidotes and the potential sideeffects including agranulocytosis, anaphylaxis, hyperkalemia, and bone marrow suppression, nafamostat was not recommended for CRRT anticoagulation by the KDIGO guideline [10].

Risk factors of lter lifespan
The impact of blood ow rate on lter lifespan remained controversial in the previous studies [67,68]. The negative effect of high uid removal on lter lifespan has been demonstrated in 2 large multi-centre trials of RRT intensity [69,70]. Lower APTT and INR, higher PLT counts and bilirubin level, and the usage of MV had been reported as factors with negative association with lter lifespan in a meta-analysis [71], despite very low evidence level. These ndings suggested that, in order to maintain su cient circuit patency, the intensity of anticoagulation-free CRRT should be targeted at appropriate range, and in addition, patients with these risk factors who underwent anticoagulation-free CRRT should be monitored closely and switch to protocols with anticoagulation if the lter clotted within a short period of time.

Limitations
There are some limitations in our present systematic review. First, most of the included studies were observational study with potential selective bias and confounding factors. Second, the guidelines and clinical practice had changed greatly across the long time span of the included studies, so that there are great differences in patient population and CRRT protocols among these studies, which could contribute to the signi cant inter-study heterogenicity. Third, we tried to performe meta-regression analysis to indentify the cut-off value of coagulation parameters at which patients could bene t from anticoagulationfree CRRT, however, no signi cant results were obtained. Further studies should be carried out under the latest guideline and focus on the speci c cut-off point for coagulation factors that would indicate the possibility to perform CRRT without anticoagulation. Therefore, we are performing a retrospective cohort study to evaluate the effecacy of anticoagulation-free CRRT and and risk factors of lter lifespan in critically ill patients with high bleeding risk and try to nd out a model to predict the lter lifespan in patients who underwent anticoagulation-free CRRT.

Conclusions
Anticoagulation in critically ill patients requiring CRRT at high risk of bleeding is a challenging work for clinicians. The lter lifespan in anticoagulation-free patients with increased bleeding risk was comparable to that in patients without increased bleeding risk underwent systemic heparin anticoagulation CRRT.
Compared to the anticoagulation-free protocol, RCA could signi cantly prolong the lter lifespan without incremental bleeding risk and citrate-related metabolic complications. Nafamostat's advantage on lter lifespan were weighed out by its drawbacks and was not recommended for anticoagulation of

Consent for publication
Not applicable.

Availability of data and materials
All data generated or analyzed during this study are from published articles.

Competing interests
Page 8/22 The authors declare that they have no competing interests.

Funding
This work was supported by the National Natural Science Foundation of China (81700584).

Authors' contributions
WZ and MB contributed equally to this work. WZ, MB, and XMC conceived the study, participated in the design, collected the data, performed statistical analyses and drafted the manuscript. YY, XLC, and LJZ performed statistical analyses and helped to draft the manuscript. MB collected the data and revised the manuscript critically for important intellectual content. XMC collected the data, performed statistical analyses and helped to revise the manuscript critically for important intellectual content. All authors read and approved the nal manuscript.     The results of bias risk assessment of the 3 randomized controlled trials.