Safety of low-molecular-weight heparin compared to unfractionated heparin in hemodialysis: a systematic review and meta-analysis

Background Low molecular weight heparins (LMWH) have been extensively studied and became the treatment of choice for several indications including pulmonary embolism. While their efficacy in hemodialysis is considered similar to unfractionated heparin (UFH), their safety remains controversial mainly due to a risk of bioaccumulation in patients with renal impairment. The aim of this systematic review was to evaluate the safety of LMWH when compared to UFH for extracorporeal circuit (ECC) anticoagulation. Methods We used Pubmed, Embase, Cochrane central register of controlled trials, Trip database and NICE to retrieve relevant studies with no language restriction. We looked for controlled experimental trials comparing LMWH to UFH for ECC anticoagulation among end-stage renal disease patients undergoing chronic hemodialysis. Studies were kept if they reported at least one of the following outcomes: bleeding, lipid profile, cardiovascular events, osteoporosis or heparin-induced thrombocytopenia. Two independent reviewers conducted studies selection, quality assessment and data extraction with discrepancies solved by a third reviewer. Relative risk and 95% CI was calculated for dichotomous outcomes and mean weighted difference (MWD) with 95% CI was used to pool continuous variables. Results Seventeen studies were selected as part of the systematic. The relative risk for total bleeding was 0.76 (95% CI 0.26–2.22). The WMD calculated for total cholesterol was −28.70 mg/dl (95% CI -51.43 to −5.98), a WMD for triglycerides of −55.57 mg/dl (95% CI -94.49 to −16.66) was estimated, and finally LDL-cholesterol had a WMD of −14.88 mg/dl (95% CI -36.27 to 6.51). Conclusions LMWH showed to be at least as safe as UFH for ECC anticoagulation in chronic hemodialysis. The limited number of studies reporting on osteoporosis and HIT does not allow any conclusion for these outcomes. Larger studies are needed to evaluate properly the safety of LMWH in chronic hemodialysis. Electronic supplementary material The online version of this article (doi:10.1186/s12882-017-0596-4) contains supplementary material, which is available to authorized users.


Background
Low molecular weight heparins (LMWH) are among the preferred anticoagulant for preventing and treating venous thrombosis [1]. A meta-analysis of randomized controlled trials showed similar efficacy between LMWH and unfractionated heparin (UFH) for acute deep venous thrombosis prevention, and no bleeding risk difference [2] with a reduced mortality rate in favor of LMWH. [2] LMWH are replacing UFH as the first line treatment for pulmonary embolism and unstable angina, a choice mainly due to their predictable effect and convenient use [3].
Despite the fact that the European Best Practice Guidelines for hemodialysis recommended the use of LWMH for the extracorporeal circuit (ECC) anticoagulation, UFH remains the most frequent choice for hemodialysis treatment in North America [4]. While cost may be the main argument for not using LMWH in hemodialysis, their safety remains a major concern. UFH is metabolized by both hepatic and renal pathways but LMWH are mainly cleared through the kidneys leading to a potential bioaccumulation and an increased risk of hemorrhage [5]. A meta-analysis conducted by Lim et al. evaluated the safety and efficacy of LMWH in hemodialysis, finding no difference between LMWH and UFH for both bleeding and thrombosis of the ECC. While the sample was large enough to evaluate efficacy (thrombosis), limited data was available to evaluate the risk of bleeding, leading to large confidence intervals (CI), and therefore limited conclusions [6].
UFH is known to modify the lipid profile, to induce osteoporosis, and to carry a risk of heparin-induced thrombocytopenia (HIT). However, the impact of LMWH on these outcomes in hemodialysis remains unclear [7].
The aim of this systematic review was to evaluate the safety of LMWH compared to UFH as an anticoagulant of the ECC among patients undergoing chronic hemodialysis. The primary outcomes were the risk of minor and major bleedings, cardiovascular events and osteoporotic fractures. Secondary outcomes were changes in lipid profile, osteoporosis and HIT.

Methods
Two independent reviewers (HHL and ER) conducted the study selection, validity assessment and data extraction with disagreement solved by a third reviewer (JPL).

Performed searches
Databases were screened to retrieve prospective experimental studies comparing LMWH to UFH for anticoagulation of the ECC during hemodialysis for patients with end-stage renal disease (ESRD). The search strategy was developed with a professional librarian. We searched Pubmed (from start up to January 2016), Embase (1974 to 2016 week 1), Cochrane central register of controlled trials (from start to January 2016), Trip database, the National Institute for Health and Care Excellence (NICE) and clinicaltrials.gov/ with no language restriction. Performed searches were conducted by one reviewer (HHL). Reference lists of selected studies were screened manually and authors were contacted when additional data was needed (HHL and ER) Additional file 1.

Study selection
Studies were selected if they fulfilled the following criteria: a) Patients were adults with ESRD undergoing chronic hemodialysis (incident or prevalent); b) LMWH was compared to UFH for anticoagulation of the ECC during hemodialysis; c) at least one of the outcomes of interest was reported (minor or major bleeding, lipid profile changes for total cholesterol, LDL-cholesterol, triglycerides, cardiovascular events, osteoporosis, osteoporotic fractures, HIT); d) the design was a prospective randomized or non-randomized cross-over or a parallel randomized study. A study was excluded if: a) it was a non-randomized parallel-design; b) it was using historical data; or c) the results were published more than once in which case only the most complete study was kept.

Quality assessment
Quality of each study was assessed using the Cochrane risk of bias tool [8] for randomized trials or the risk of bias in non-randomized studies (Robins-I) [9]. The Cochrane risk of bias tool evaluates the methodology of the study and the potential biases in the research question context. [8] There is no final score but results are reported as low, unclear or high risk of bias for different potential biases. The Robins-I uses a similar approach but is adapted to non-randomized trials.

Data extraction
A common form was used to ascertain extraction of the complete study characteristics and outcomes of interest as follows: a) bleeding events classified as major and minor when specified by the authors; b) cardiovascular events; c) LDL, total-cholesterol and triglycerides levels at start and end of study for both treatment and control; d) osteoporotic fractures; e) osteoporosis diagnostic; f ) bone density changes; g) OPG/RANKL biomarkers affecting bone density; and h) HIT.

Statistical analysis
The weighted kappa statistics was used to assess between reviewers agreement for study selection and quality assessment [10]. We calculated relative risks (RR) and 95% confidence intervals (CI) for dichotomous clinical data. A pooled weighted mean difference (WMD) and its 95% CI were used to compare continuous outcomes. The pooled overall effect for both RR and WMD was estimated using the method described by DerSimonian and Laird [11] with a random effect model for heterogeneous data. Heterogeneity was evaluated using the I 2 statistic using a 60% significance threshold. Publication bias was assessed visually using a funnel plot and using the Egger test [12]. All analyses were performed with Stata IC 11.0.

Sensitivity analyses
The pooled WMD and 95% CI were calculated based on the pre-and post-treatment measures using their standard deviations to derive the WMD's standard deviation [8]. In the main analysis, the correlation factor was fixed at 0.5 based on the assumption that repeated measures would be moderately correlated. We tested the robustness of this assumption with correlation factors of 0.3 and 0.8 as sensitivity analyses.

Characteristics of studies and patients
Characteristics of studies are listed in Table 1. The number of patients ranged from five to 110, and followup ranged between one and 60 months. Various types of LMWH were used for the experimental arm, including nadroparin, enoxaparin, tinzaparin, dalteparin, certoparin and fraxiparin. Both men and women were included with   Table 2.
The most common dialysis schedule was a session of four hours thrice weekly. Hemodiafiltration procedure was also used in three studies [15,26,27].

Quality assessment
Based on the analysis with the Cochrane risk of bias tool for randomized studies (Fig. 2), we observed a high rate of potential performance bias mainly caused by a lack of blinding and only few measures to ensure similar care to the patients independently from heparin type. Blinding, or lack of it, could also affect outcome detection (detection bias). We judged that there was a high risk of bias when the outcome of interest was subjective. Bleeding events were categorized as being minor or major in eight out of nine studies, but most of the time no clear definition was provided except for one study. [27] The lack of definition made the evaluation of bleeding events subjective. In randomized trials, concealment was not specified, leading to an imprecise risk of selection bias for these studies. Overall, attrition and reporting biases were adequately handled through the trials (Fig. 2). Similarly, non-randomized cross-over studies had a moderate to high risk of outcome measurement when bleeding events evaluation was based on subjective assessment due to lack of clear outcome definition. The Robins-I tool points also to a missing data problematic due to drop-outs. (Fig. 3) The weighted kappa was 0.72 for Cochrane risk of bias tool and 0.69 for Robins-I, indicating a substantial agreement between raters.

Synthesis of individual studies results
Not all studies had a predefined definition of the outcome of interest. Bleeding outcome was specified in the method section in only half of the studies [15,22,24,27,28]. All except one [14] reported bleeding events as minor or major. Only Stefoni et al. provided a clear definition of major (melena, hematemesis and arterial epistaxis) and minor bleeding (venous epistaxis, subconjuctival hemorrhage or prolonged bleeding from the cannulation site after removal of the dialysis needle) [27]. For Bramham et al., a bleeding event was defined as a prolonged compression time or any other bleeding episode reported during or between dialysis sessions by the patients, including epistaxis or conjunctival bleeding [15]. Among all bleeding events, only two were categorized by the authors as major, and both occurred in the LMWH group. When combining both major and minor bleedings, 24 (6.6%) events were recorded for LMWH compared to 31 (8.5%) for UFH. Bleeding outcomes are presented in Table 3.
No study evaluated the incidence of osteoporotic fractures. Osteoprotegerin (OPG) and receptor activator of nuclear factor-κB ligand (RANKL) levels were measured in two studies. The OPG/RANKL system is involved in bone metabolism. No significant difference between LMWH and UFH (Additional file 2: Table S3) was found [16,29]. Lai et al. [20] measured the bone densitometry with dual energy X-ray and observed an increase of bone mass density (BMD) in ward's triangle by 0.75% after switching to LMWH. This increase did not reach statistical significance (p = 0.11). The mean BMD in the same region decreased by 2.38% after patients were back on UFH [20]. They did not observe a similar trend for the other sites investigated.
Bramham et al. [15] reported that they did not observe any case of thrombocytopenia among their patients during the study follow-up for both LMWH and UFH.

Meta-analysis results
Schrader et al. [26] and Yang et al. [28] were excluded from the analysis since they did not have any bleeding event in both LMWH and UFH groups. The overall RR for total bleeding (minor and major) with LMWH compared to UFH was 0.76 (95% CI: 0.26, 2.22) using a random effect model. The I 2 statistic for heterogeneity was statistically significant (p = 0.018) (Fig. 4a). Compared to UFH, LMWH showed lower cholesterol level, with a pooled WMD of −28.70 mg/dl (95% CI: -51.43, −5.98) (Fig. 4c). A pooled WMD of −55.57 mg/dl (95% CI: -94.49, −16.66) was calculated for triglycerides showing lower levels in the LMWH group compared to UFH (Fig. 4d). A similar trend was observed with LDLcholesterolbut the results were not statistically significant (pooled WMD: −14.88 mg/dl (95% CI: -36.27, 6.51)) (Fig. 4b). Results were pooled using random effect model as the heterogeneity test was statistically significant. Because of the small number of studies, visual interpretation of the funnel plots to assess possible publication bias was inconclusive (Additional file 2). The Egger test results indicated a potential publication bias (data not presented).

Discussion
This systematic review reports on safety of LMWH compared to UFH when used for anticoagulation of the extracorporeal circuit during hemodialysis. There was no statistical difference for risk of bleeding in the LMWH compared to the UFH group. We also showed lower levels of total cholesterol and triglycerides when exposed to LMWH compared to UFH. LDL-cholesterol levels tended to be lower using LMWH but the difference did not reach statistical significance. Because of the small numbers of studies, we were not able to estimate the differential risk of cardiovascular events, osteoporotic fractures and HIT. A decade after the publication of the first metaanalysis published by Lim et al. [6], we conducted a new systematic review to re-evaluate LMWH bleeding risk in hemodialysis compared to UFH by including all new relevant studies. Lim et al. found similar overall bleeding risk (RR = 0.96, 95% CI: 0.27, 3.43), a result that is consistent with our study. Our updated meta-analysis adds to this earlier study by narrowing the CI and reducing the upper limit from three to two, supporting that, at least, LMWH does not carry a higher risk of bleeding. Our results are also consistent with observed bleeding risk among patients without renal failure using LMWH for their primary indication [32,33]. Renal-dependent clearance of LMWH remains a concern. A prospective observational cohort study evaluated the bioaccumulation of dalteparin when administered at the therapeutic dose twice a day. Patients with a glomerular filtration rate (GFR) inferior to 30 mL/min/1.73 m 2 showed a significant bioaccumulation of dalteparin at the end of the study [34]. On the other hand, two studies were conducted among critically ill patients with severe renal  Fig. 2 Cochrane risk of bias tool diagram insufficiency, and did not observe bioaccumulation of dalteparin using a prophylactic dose of 5000 IU once a day [35,36]. The later schedule is closer to the hemodialysis context. No prior systematic review compared LMWH and UFH for cardiovascular events and lipid profile in hemodialysis. UFH is known to activate lipoprotein lipase and hepatic lipase and therefore to affect lipid profil [7]. Unfortunately, data on cardiovascular events were limited. Although it was demonstrated in the SHARP study [37] and a meta-analysis [38] that, for patients with chronic kidney disease, lowering LDLcholesterol by 38.61 mg/dl (1 mmol/L) induced a reduction of 17% of major atherosclerotic events, in dialysis-dependent patients there is only a weak association between LDL-cholesterol level and the risk of cardiovascular event [39]. Indeed, low LDL-cholesterol level could be associated with a high risk of cardiovascular event in dialysis patients when there is coexistence of malnutrition and inflammation [40,41]. While our results were in favor of LMWH, it remains uncertain whether the decrease would translate into lower cardiovascular events among ESRD patients.
Our meta-analysis is the first to compare LMWH and UFH for bone metabolism in hemodialysis. The exact mechanism involved in osteoporosis induced by UFH  Studies excluded due to "zero cells" in both LMWH and UFH groups remains unclear. One hypothesis suggests a suppression of osteoblast formation and activation of osteoclasts promoting bone loss [5]. Unfortunately, no study compared osteoporotic fractures. Different studies presented the OPG/RANKL system as an actor of bone remodeling [42,43]. No difference was detected in the studies retrieved [16,29]. While it is reported that LMWH cause less bone loss among pregnant women, these results were not replicated among hemodialysis patients [7]. Uncertainty was raised in the Kidney Disease Improving Global Outcomes (KDIGO) clinical guidelines in 2009 about the reliability of BMD measured with dual energy X-ray [44]. However a systematic review and meta-analysis conducted by Bucur et al. [45] showed that BMD is lower in chronic kidney disease (CKD) patients with fracture compared to CKD patients without fracture, suggesting that BMD could still be clinically relevant to evaluate fracture risk in CKD. Thrombocytopenia is a rare but severe adverse reaction to heparin. Although Bramham et al. [15] did not observe any HIT; we cannot reach any conclusion regarding this outcome. The incidence of thrombocytopenia was reported between 1 and 3% with UFH, while it was less than 1% with LMWH [46]. UFH binds to the platelet factor 4 (PF4). Antibodies then recognize this complex, which leads to an autoimmune reaction and thrombocytopenia [47]. Because a long chain of saccharides is needed to bind to PF4, and that LMWH are composed of short chains, it is believed that LMWH is less likely to form such complexes [48].
Our review has some limitations. Except for bleeding and cardiovascular events, we could not obtain data on every one of our primary outcomes. Our analyses were based on the secondary outcomes, and they might not be effective surrogates to infer on the main outcomes. The bleeding risk with LMWH may have been misestimated as some of the studies included an exploratory phase to determine the right LMWH dosage, which is partly based on occurrence of bleeding events, while UFH treatment arm did not go through such phase since patients were already stable under that regimen.
The potential publication bias detected must be taken into consideration when interpreting the results. Although the use of objective searching methods we could not eliminate this bias. Because of the lack of clear and consistent definition of major and minor bleeding in the retrieved studies, we opted to evaluate the risk of total bleeding. However, it would have been interesting to measure the risk of hemorrhage according to its severity, where major bleeding has a greater impact on the patient but also on the type of care needed [49]. A standardized definition of major and minor bleeding needs to be used in future investigations; such definition is already provided by the subcommittee on control of anticoagulation of the scientific and standardization committee of the international society on thrombosis and haemostasis [49]. The random effect model cannot compensate for the entire heterogeneity found between studies. High heterogeneity could be caused by multiple factors. The inclusion of different study designs in our review. We believe that the crossover and parallel-design should not be different for the measured outcomes in this context. Because of the small number of trials, it was not possible to analyze LMWH separately. LMWH have different pharmacokinetics and kidney failure does not alter similarly their elimination. For instance, enoxaparin might seem safe when there is clear evidence of its bioaccumulation [50,51]. Finally, the trials used different doses and had variable follow-up time. The dosage, the use of low or high flux dialysis, types of membranes, hemodiafiltration and hemodialysis procedures, all these factors would contribute to the measured heterogeneity.

Conclusion
In conclusion, the available data does not allow to determine which heparin form is safer when used for anticoagulation of the extracorporeal circuit during hemodialysis. The risk of hemorrhage was not statistically significant when LMWH was compared to UFH, and lipid levels remained comparable or lower with LMWH. As for osteoporosis and thrombocytopenia, the data does not allow making a reliable comparison with UFH. The potential sources of bias discussed earlier and the quality level of the retrieved studies hinder the interpretation of observed results. LMWH have been used in hemodialysis for many years, but we are still in dire need of trials addressing the aforementioned limitations. Future studies need to use larger sample size during a sufficiently long follow-up and a clearer definition of the outcome measured to make their results usable and interpretable. Anticoagulation is mandatory for most patients undergoing hemodialysis treatments, and the choice of anticoagulation agent must be supported by stronger evidence.

Additional files
Additional file 1: Literature search strategies for Pubmed, Embase and Cochrane as used in this study are detailed in this document. (PDF 16 kb) Additional file 2: Supplementary material providing the meta-analysis results for cholesterol (Table S1.) and triglycerides (Tables S2.), OPG/RANKL ratios (Table S3.), sensitivity analysis results (Table S4.