The guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and the Meta-Analysis of Observational Studies in Epidemiology (MOOSE) for reporting systematic reviews and meta-analyses were followed based on an established protocol—the PRISMA and MOOSE checklists are provided as Supplementary materials 1 and 2 [15,16]. Also, this study was submitted to PROSPERO, an international database of prospectively registered systematic reviews in health and social care. Individualized search strategies were developed for the following databases from inception through November 30, 2018: PubMed, EMBASE, Web of Science, and Cochrane databases. The medical subject heading (MeSH) terms included: liver biopsy. For articles related to liver biopsy, subject heading search terms and title and abstract were reviewed for the following: endoscopic ultrasound (EUS), percutaneous, transvenous, or transjugular.

All relevant articles, regardless of type, year of publication, or status (full-text manuscripts and published abstracts), were included. Based on the initial search results, the titles and abstracts of all potential studies were screened for eligibility, with references manually reviewed by cross-checking bibliographies of retrieved articles. Two reviewers (TRM and ANB) independently performed the screening process of titles and abstracts according to predefined inclusion and exclusion criteria. Any differences between TRM and ANB were resolved by mutual agreement and in consultation with a third reviewer (BN). For studies with incomplete information, primary contact was attempted with the principal authors to obtain missing data.

Randomized controlled trials and observational studies evaluating all three treatment cohorts (EUSLB, PCLB, and TJLB) were included in this analysis. For comparative review, we included studies with adults patients aged 18 and older, and studies that used all three techniques (EUSLB, PCLB, and TJLB) for hepatic sampling. Included studies were required to report at least one of the primary outcomes for study analysis. Non-human studies, or studies that only compared two modalities, were excluded. Multiple published works from similar authors were evaluated for overlapping enrollment times to preserve independence of observations, with contact of principal authors performed when needed. Studies with insufficient data, as well as study types that did not report primary or independent data, were excluded from this systematic review and meta-analysis. Case series were also not included in effort to reduce selection bias.

All EUSLB needle types were included for cumulative analysis. However, on subgroup analysis, only needles used in routine clinical practice were considered for inclusion. The available EUSLB needle types that were considered acceptable for subgroup analysis included the 19 G fine needle aspiration (FNA) or the fine needle biopsy (FNB)—including 19 G SharkCore FNB (Medtronic Co., Minneapolis, MN, USA), 19 G Expect FNA (Boston Scientific, Natick, MA, USA), 22 G SharkCore FNB needle, and the 19 G, 22 G, and 25 G Acquire EUS FNB needle (Boston Scientific). Based on previous comparator studies, older or first generation EUS core needles, including needles that are no longer available in clinical practice, were excluded to simulate real-world clinical practice [17-20]. The excluded needle types comprised the 19 G or 22 G ProCore, 18 G QuickCore (Cook Medical, Bloomington, IN, USA), and Coaxial Temno needle (Care-Fusion, McGraw Park, IL, USA).

The primary outcomes of this structured systematic review and meta-analysis were adequacy of biopsy specimens for pathology review (i.e., complete portal triads [CPT], total specimen length [TSL] in mm, and length of longest piece [LLP] in mm), and rate of adverse events. The adequacy of a liver biopsy has conventionally been defined in terms of the minimum length of the tissue cores and the presence of a minimum number of complete portal tracts (defined as at least 6 to 11 complete portal tracts for adequate diagnosis) [1,21,22]. Previous systematic reviews and meta-analyses of percutaneous and transjugular routes to obtain hepatic tissue have defined adequacy of liver biopsy as 7.5 CPT, with TSL of 17.7 mm, and 6.5 CPT, with TSL of 12 mm, respectively [10,23]. Currently, an optimal definition of specimen adequacy for EUSLB has not yet been established. Adverse events were primarily concerned with bleeding; however, direct procedure-related complications, such as significant abdominal pain, were also included.

The systematic review and meta-analysis were performed using pooled proportions. Subsequently, individual study proportions were transformed into a quantity using the Freeman–Tukey variant of the arcsine square root transformed proportion. Next, pooled proportions were calculated as the back transform of the weighted mean of the transformed proportion using DerSimonian–Laird weights for the random effects model [24-27]. The rates were estimated using a random effects model and then presented as point estimates (rates) with 95% confidence intervals (CIs) [28-30].

Subgroup analyses were also performed based upon the comparative effectiveness of varying biopsy strategies limited to studies with EUS biopsy needles used in current clinical practice. If there were <3 studies included, pooled proportions using a random effects model could not be calculated. In this case, non-pooled outcome measures were used to compare outcomes between different sampling strategies using two-sample t-tests for binomial proportions. All calculated p-values were 2-sided, and a p-value of <0.05 was considered statistically significant. Tabular and graphical analyses were performing using Comprehensive Meta-Analysis software, version 3 (BioStat, Englewood, NJ, USA). Combined weighted proportions were determined by use of the Stata 15.0 software package (StataCorp LP, College Station, TX, USA).

For observational studies, the risk of bias and quality were evaluated using the Newcastle-Ottawa Quality Assessment Scale. For randomized controlled trials, the JADAD score was used to determine quality [31,32]. High-quality studies were defined as a Newcastle–Ottawa Quality Assessment Scale score of ≥5 or a JADAD score of ≥4. For observational studies, scores were determined by study design, full manuscript or abstract article type, cohort size, specimen collection details, histologic examination, adverse event reporting, and attrition rate similar to previously published literature [20]. Risk of bias and quality of studies were assessed independently by two authors (TRM and ANB), with disagreements resolved by consultation with the third reviewer (BN).

Heterogeneity was assessed using the chi squared test and the I² statistic [33]. Significant heterogeneity was defined as p<0.05 using the Cochran Q test or I² >50%, with values >50% indicating substantial heterogeneity. Based on the use of the random effects model to estimate average effect size, a 95% prediction interval was calculated to determine the dispersion of effects and clearly illustrate heterogeneity in the calculated effect size [28,34-36].

To assess publication bias, Egger regression testing was used and a funnel plot was created and visually inspected for asymmetry [37,38]. The trim and fill method was used to correct for funnel plot asymmetry and to provide an adjusted effect [39]. The classic fail-safe test was also applied to assess risk of bias across studies.

Five comparator studies were included in this meta-analysis and a total of 656 patients were enrolled—EUSLB (n=301); PCLB (n=176); and TJLB (n=179) [14,40-43]. A PRISMA flow chart of search results is shown in Fig. 1. Three abstracts and two full-text manuscripts published between 2015 and 2018 were included in this analysis. Among the included studies, four were retrospective in design and one was prospective. The cumulative data of all studies and patient characteristics are highlighted in Table 1. The mean age of all included patients was 51.62±2.81 years and 43.93% of patients were male. The adequacy rate of biopsy specimens for EUSLB was 93.51% (95% CI, 60.41 to 99.27); I²=88.40%; prediction interval –1.00 to 1.00, range 2.00; Meanwhile, for PCLB, the adequacy rate was 98.27% (95% CI, 93.31 to 99.57); I²=0.00%; prediction interval 0.68 to 1.00, range 0.32. Finally, for TJLB, the adequacy rate was 97.61% (95% CI, 93.39 to 99.16); I²=0.00%; prediction interval 0.78 to 1.00, range 0.22 (Fig. 2A-C). The cumulative data for each modality is highlighted in Supplementary Table 1.

In order to compare the three modalities, only EUS biopsy needles utilized in current clinical practice were considered. Indeed, the study by Nakanishi et al. was excluded given inclusion of needle types not routinely utilized in clinical practice [14]. Regarding adequacy of biopsy specimens, EUSLB was comparable to PCLB (98.53% vs. 100.00%, p=0.410) and was similar to TJLB (98.53% vs. 100.0%, p=0.351). Meanwhile, adverse events with EUSLB were similar to PCLB (16.46% vs. 12.24%, p=0.516) and to TJLB (EUS: 16.46% vs. TJLB: 4.88%, p=0.071).

Analysis of specimen adequacy also involved comparison of CPT, TSL, and LLP among the three modalities. A comparison of EUSLB and PCLB resulted in histologic samples with no difference in CPT (odds ratio [OR], 4.70; 95% CI, –0.54 to 9.92; I²=79.79%; p=0.079; Fig. 3A), no difference in LLP (OR, –3.34; 95% CI, –7.14 to 0.46; I²=79.63%; p=0.085; Fig. 3B), and longer TSL (OR, 11.35; 95% CI, 6.77 to 15.93; I²=0.00%; p<0.001; Fig. 3C). Meanwhile, a comparison of EUSLB and TJLB resulted in no difference in LLP (–1.92 [95% CI, –5.94 to 2.11]; I²=82.86%; p=0.351; Fig. 4A), smaller number of CPT –1.67 (95% CI, –3.28 to –0.06; I²=0.00%; p=0.042; Fig. 4B), and longer TSL (OR, 6.28; 95% CI, 1.93 to 10.63; I²=3.94%; p=0.005; Fig. 4C).

All 5 studies were assessed using the Newcastle–Ottawa Quality Assessment Scale, which assigned a high-quality grade (i.e., score ≥5) to the five studies (four retrospective and one prospective). Authors’ judgements each included study is highlighted in Table 1. Publication bias was also assessed. Visual inspection of the funnel plot for studies assessing adequacy demonstrated that smaller and statistically insignificant studies appeared to be missing likely due to publication bias (Supplementary Fig. 1A). Adequacy of EUSLB was decreased from 93.51% (95% CI, 60.41 to 99.27) to 81.94% (95% CI, 47.49 to 95.79) with Duval and Tweedie’s trim and fill method (Supplementary Fig. 1B). Using the classic fail-safe test to assess for publication bias, it was determined it would take 43 non-significant studies to nullify the results of this analysis.

Despite comparable adequacy and complication rates, the EUS technique affords many advantages over percutaneous and transjugular approaches. While PCLB and TJLB have been associated with significant variation in sample yield, EUS strategies provide high-resolution imaging and access to sampling of both lobes of the liver, which may provide a more accurate representation of liver histology and reduced sampling error [10,20,44-46]. Endosonography with Doppler capability allows for the avoidance of intrahepatic vessels and real-time imaging to navigate and avoid the bowel, gallbladder, and pleural space [12-14,44]. Most importantly, EUSLB is able to achieve these results regardless of body habitus, whereas the percutaneous and, to a lesser extent, the transjugular routes, may be limited in obese patients [47]. Furthermore, benefits of an EUS-guided approach include concomitant endoscopic esophageal variceal screening, gastric biopsy for Helicobacter pylori, and EUS imaging of the gallbladder and bile duct for evaluation of choledocholithiasis as a potential etiology of abnormal liver chemistries, in addition to the exclusion of ampullary and pancreas lesions. Despite these benefits, EUS may be limited by operator variability, requirement for deep sedation, procedure length, and associated monetary costs [20,40]. However, the use of deep sedation may also beneficial, and may be appealing to the patient, reducing any morbidity, pain, or anxiety associated with the procedure.

The first published cases of EUSLB were described in 2007 using a novel Tru-Cut core biopsy needle (QuickCore; Cook Medical, Winston Salem, NC, USA) [18,48]. Since the introduction of the first EUSLB, many studies have been performed to determine the preferred needle. Schulman et al. sought to assess 6 needle types: 19 G (SharkCore FNB [Medtronic], Expect FNA [Boston Scientific], and ProCore FNB [Echo Tip HD ProCore; Cook Medical]), 22 G (SharkCore FNB needle [Medtronic]), and percutaneous 18 G (QuickCore [Cook Medical] and Coaxial Temno Needle [Care-Fusion, Waukegan, IL, USA]) [17]. This previous study found that the 19 G SharkCore FNB needle provided superior histologic yield compared to existing percutaneous needles. Therefore, we excluded a study that utilized the QuickCore and ProCore percutaneous needle from our comparative analysis based on the evidence of the reduced quality of these needles, current availability, and consideration of the needles utilized in routine clinical practice [14].

Currently, there are a variety of guidelines and recommendations that outline parameters for hepatic tissue adequacy regarding liver biopsy—each dependent upon the strategy or route of tissue acquisition. Cholongitas et al. previously performed a systematic review and meta-analysis of over 10,000 PCLBs and reported that a mean of 7.5 CPT and TSL of 17.7 mm were required for pathologic review [23]. When obtained via the transjugular route, adequacy is defined as 6.5 CPT and TSL of 12 mm [10]. Although an optimal definition of specimen adequacy for EUSLB has not yet been established, the American Association for the Study of Liver Diseases (AASLD) recommends ≥11 CPT as the definition of adequacy, regardless of route of sampling [8]. Perhaps more importantly, the AASLD guideline also utilizes TSL >15 mm to define adequacy, albeit it considers that the ideal size is 30 mm [8,18]. Although all routes of tissue sampling in our study achieved ≥11 CPTs and TSL >15 mm, only EUSLB achieved the ideal TSL of ≥30 mm.

There are several limitations to this study. Although we assessed for heterogeneity as well as risk of bias, specific limitations include differences in patient population across multiple included modalities, as well as familiarity of the individual provider with the service (i.e., interventional radiology and gastroenterology) and with various techniques. Although each study evaluated all sampling techniques, there was significant heterogeneity among the included procedures (i.e., I²>50%) and the variety of needle types utilized. Most importantly, the lack of a formal definition of EUSLB adequacy precludes consistency across centers and pathology providers. Similarly, while our cumulative analysis included EUS needle types no longer utilized in clinical practice, the use of a standard needle type is not verified across all institutions. Additionally, due to a paucity of data, the indication for liver biopsy could not be confirmed for individual cases, thereby limiting a subgroup analysis to evaluate for changes in efficacy or safety between different indications.

Furthermore, no randomized studies were included in this analysis and only two full-text manuscripts were included, thereby potentially increasing the selection bias. As most included studies were retrospective in design, there may have been a reason for EUSLB to be performed as opposed to other sampling routes—perhaps due to body habitus or other confounders that may artificially reduce or increase the efficacy or safety of the procedure. Publication bias was present in this meta-analysis; however, with the trim and fill method, the adequacy of EUSLB decreased from 93.51% to 81.94%. Additionally, given the lack of current comparative data and expertise needed to perform EUSLB, the learning curve, or clinical acumen needed to perform an effective procedure, is unknown— thereby potentially limiting the generalizability of our findings to centers with less familiarity or proven proficiency.

Despite these limitations, this study has several strengths. Most importantly, our meta-analysis using a random effects model to methodologically summarize all available data to evaluate feasibility, efficacy, and safety of EUSLB compared to PCLB and TJLB, with an emphasis on subgroup analysis analyzing needles only commonly used in current clinical practice. Although reporting bias of negative studies cannot be verified, small case series were eliminated to minimize inherent selection bias. Additionally, this systematic review and meta-analysis evaluated objective data, including biopsy adequacy, CPT, TSL, LLP, and adverse events for all included modalities. These findings, as highlighted in our systematic review and meta-analysis, are similar to previous cumulative systematic reviews and meta-analyses of EUSLB. Mohan et al. reported a histologic diagnosis rate of 93.9% and adverse event rate of 2.3% for EUSLB [20]. These authors concluded that a 19 G FNA needle provided optimal biopsy specimens. While recent published data by Diehl et al. suggests that heparin priming of EUS-FNA or FNB needles does not negatively affect cytologic or histologic interpretation, additional studies by similar authors have shown tissue adequacy to be higher for 19 G FNA versus 22 G FNB needle among patients with chronic liver disease [49,50].

To date, our study is the only meta-analysis to directly compare EUSLB, PCLB, and TJLB, including data from studies that evaluated all common approaches to obtain hepatic tissue. While TJLB provides an ability to simultaneously obtain portal pressure measures, EUS may be able to do the same in the near future [51,52]. In fact, human studies have already been performed, and demonstrated a technical success rate of 100% with no serious adverse events reported for EUS [53,54]. Moreover, this systematic review and meta-analysis specifically aimed to provide real-world data to simulate clinical practice through subgroup analysis that included only needle types that are currently used in clinical practice. Despite the significant learning curve and inter-operator variability required to perform EUS, we are hopeful that the subgroup analysis will improve the generalizability of our results. While significant heterogeneity was noted in our meta-analysis and was not surprising given the cumulative nature of reporting results, the use of a prediction interval was employed to demonstrate the variability within real world clinical practice [28,55].

In conclusions, EUSLB is a minimally invasive procedure that appears to be a safe and effective approach to obtain hepatic tissue. Although adequate needle size and tissue acquisition techniques for optimal EUS-FNA require further study, current data suggests that EUSLB is comparable to the alternative strategies of PCLB and TJLB. A cost-utility study to determine which modality is most cost-effective may be helpful to guide future decisions regarding diagnostic strategies. Future well-designed, prospective comparator studies are needed to determine which subset of patients may benefit most from EUSLB.