Endoscopic ultrasound-guided biliary drainage vs endoscopic retrograde cholangiopancreatography for malignant distal obstruction: Randomized-trial meta-analysis

Abstract

BACKGROUND

Randomized trials have compared primary transmural endoscopic ultrasound-guided biliary drainage with endoscopic retrograde cholangiopancreatography. However, differences in devices, expertise, patient selection, and outcome definitions limit broad conclusions.

AIM

To compare the efficacy and safety of primary ultrasound-guided vs retrograde biliary drainage.

METHODS

PubMed, MEDLINE, CENTRAL, ClinicalTrials.gov, and World Health Organization registry records indexed using CENTRAL were searched on August 6, 2026, without language restrictions. We included randomized trials comparing primary transmural ultrasound-guided vs retrograde drainage for malignant distal biliary obstruction. Risk ratios were calculated using Paule-Mandel random effects and modified Hartung-Knapp confidence intervals. Pancreatitis was examined using continuity-corrected risk ratio, Peto odds ratio, exact odds ratio, and absolute risk difference.

RESULTS

The six trials included 739 participants. Technical success did not differ significantly (risk ratio 1.08, 95% confidence interval 0.96-1.22; I² = 64%), and clinical success was comparable (risk ratio 1.01, 95% confidence interval 0.95-1.07; I² = 0%). Procedure-related adverse events were similar (risk ratio 0.83, 95% confidence interval 0.53-1.31). Pancreatitis occurred in 2/374 (0.5%) vs 24/365 (6.6%) participants (risk ratio 0.20, 95% confidence interval 0.05-0.77; Peto odds ratio 0.17, 95% confidence interval 0.08-0.36). The absolute risk difference was -6.0%, corresponding to a number needed to treat of 17. Stent dysfunction or re-intervention was numerically less frequent, with its confidence interval including no effect.

CONCLUSION

Ultrasound-guided drainage reduces pancreatitis and is a reasonable alternative for selected patients in expert centers; however, it is not a universal replacement for retrograde drainage.

Key Words: Endoscopic ultrasound-guided biliary drainage; Endoscopic retrograde cholangiopancreatography; Malignant distal biliary obstruction; Postprocedural pancreatitis; Randomized controlled trial; Meta-analysis

Core Tip: This randomized meta-analysis included six trials and 739 participants. Primary transmural ultrasound-guided and retrograde biliary drainage achieved comparable clinical success rates. Ultrasound-guided drainage reduced post-procedural pancreatitis from 6.6% to 0.5%, with concordant rare-event sensitivity analyses. Technical success varies across devices and trials, whereas its effects on other adverse events and repeat interventions remain uncertain. Evidence supports ultrasound-guided drainage as an alternative for selected patients in expert centers, and not as a universal replacement for retrograde drainage.

INTRODUCTION

Malignant distal biliary obstruction causes jaundice, pruritus, cholangitis, and impaired liver function and delays systemic anticancer treatment[1,2]. Endoscopic retrograde cholangiopancreatography (ERCP) with transpapillary stenting is the standard primary drainage approach[3]. Despite its effectiveness, cannulation and papillary instrumentation can cause pancreatitis, bleeding, infection, and perforation[4,5]. Tumor ingrowth or overgrowth can also compromise transpapillary stenting and prompt re-intervention.

Endoscopic ultrasound-guided biliary drainage (EUS-BD) creates a transmural route, most often a choledochoduodenostomy or hepaticogastrostomy, without traversing the papilla[6]. This anatomy provides a plausible reduction in pancreatitis and avoids stent positioning across the tumor. EUS-BD was initially used after failed ERCP and offers an internal alternative to percutaneous drainage[7].

Randomized trials have evaluated EUS-BD as a primary treatment rather than a rescue therapy[8-13]. Their results are not interchangeable; with older trials using tubular metal stents and including different transmural routes, whereas recent trials evaluated cautery-enhanced lumen-apposing metal stents in patients with a substantially dilated common bile duct. Operator experience, eligibility, pancreatitis prophylaxis, and follow-up differed between the groups. The randomized design improves internal validity, but its restriction to a small number of specialist-center trials reduces its statistical power and external validity.

We updated the evidence on August 6, 2026, retrieved the full report of the previously excluded Park trial and incorporated it along with the 2026 Anderloni trial. This review asked whether primary transmural EUS-BD and ERCP differ in terms of technical success, clinical success, procedure-related adverse events, postprocedural pancreatitis, and stent dysfunction or re-intervention.

MATERIALS AND METHODS

Reporting and protocol

This review follows the PRISMA 2020 guidelines[14]. The original review was not prospectively registered, and no public protocol was deposited before conducting the study. The refined eligibility criteria and analytical changes introduced during the revision are documented in the Supplementary material. The absence of prospective registration is considered a limitation.

Search strategy

PubMed/MEDLINE, CENTRAL, and ClinicalTrials.gov were searched from their inceptions until August 6, 2026. The World Health Organization (WHO) International Clinical Trials Registry Platform records were identified using the ICTRP source indexed in CENTRAL. Reference lists of eligible trials, recent systematic reviews, and 2026 individual-participant-data meta-analyses were screened. No date or language restrictions were imposed. The exact Boolean strings, field tags, search dates, result counts, and registry queries are listed in Supplementary Table 1. Direct subscription searches of EMBASE, Scopus, and Web of Science were unavailable; EMBASE-sourced records indexed in CENTRAL were screened, and this incomplete direct database coverage is a limitation.

For reports without accessible full text, the DOI, publisher site, PubMed Central, trial registry, institutional repository, and author-sharing pages were checked. We retrieved and included the complete report of Park et al[10]. The 2025 Altonbary Conference Report and several ongoing registries lacked full reports with extractable outcome definitions and were not pooled. The authors were not contacted for this study. Ongoing and unpublished data are shown in Supplementary Table 2[15].

Eligibility and study selection

We included randomized trials that enrolled adults with malignant distal biliary obstruction who underwent primary transmural EUS-BD and compared them with primary ERCP-guided transpapillary drainage. Trials were required to report at least one pre-specified outcome. We excluded rescue EUS-BD following failed ERCP, non-randomized comparisons, single-arm studies, protocols without results, duplicate reports, and non-transmural routes. The Zhao et al[16] trial was excluded because its experimental procedure was an antegrade transpapillary endoscopic ultrasound (EUS) route rather than transmural drainage. Multiple publications were grouped by trial: The most complete report supplied the primary data, and companion publications were used only to resolve definitions or follow-up.

Two reviewers independently screened the PubMed database. One reviewer conducted the August 6, 2026, update and checked new eligibility decisions, event counts against primary reports, and recent individual-participant-data publications. The update was not duplicated independently, and a departure from optimal systematic review practice was reported as a limitation.

Data extraction and outcomes

A standardized form captured the trial design, randomized and analyzed sample sizes, malignancy, EUS route, stent type, operator setting, outcome definitions, follow-up, and arm-level event counts. Technical success was defined as successful stent placement. Clinical success followed the definition in each trial, usually a pre-specified bilirubin reduction within 2-4 weeks; the exact denominator and threshold for each trial are shown in Supplementary Tables 3 and 4. Procedure-related adverse events were extracted at the reported early or 30-day window, when available. Pancreatitis was considered a separate outcome. The final durability outcome combined trial-defined stent dysfunction or re-intervention because the reports used related but non-identical definitions and time horizons; this heterogeneity was considered when interpreting the pooled estimate.

Risk of bias and certainty

The risk of bias was assessed using the revised Cochrane risk-of-bias tool for randomized trials (RoB 2) across randomization, deviations from intended interventions, missing outcome data, outcome measurements, and selective reporting[17]. Open-label procedures were not automatically rated as high risk; judgment depended on the objectivity of each outcome and adherence to the intention-to-treat analysis. Certainty was summarized using the Grading of Recommendations Assessment, Development and Evaluation domains of risk of bias, inconsistency, indirectness, imprecision, and publication bias[18]. The domain-level judgments are presented in Supplementary Tables 5 and 6.

Statistical analysis

For technical success, clinical success, adverse events, pancreatitis, and stent dysfunction or re-intervention, study-level risk ratios (RRs) were pooled on a log scale. A 0.5 continuity correction was only applied to studies with zero cells. Between-study variance was estimated using the Paule-Mandel method[19]; random-effects confidence intervals (CI) used a modified Hartung-Knapp procedure as only six trials contributed and their sizes varied[20]. Heterogeneity was summarized with I² and τ². Formal funnel plots or regression tests were not performed because < 10 trials were available.

Pancreatitis was uncommon, and several trial arms experienced no adverse events. The continuity-corrected random-effects RR was the primary relative measure. Sensitivity analyses used the Peto odds ratio, which performs reasonably well with sparse balanced trials when the effects are not extreme[21] and a conditional exact odds ratio from the pooled 2 × 2 table. The absolute effect was expressed as the Newcombe risk difference and number of treatments required. These measures answer different questions and are not combined; RR and odds ratio describe the relative effects, whereas risk difference describes the absolute change in event probability. Analyses were independently implemented in Python 3.12 and checked against published trial totals.

RESULTS

Study selection

A total of 466 databases and registry records were identified. After removing 55 cross-source and report-level duplicates, 411 titles and abstracts were screened; 17 reports underwent full-text or registry assessment and 11 were excluded for documented reasons. Six randomized trials, including 739 participants (374 EUS-BD and 365 ERCP), entered the qualitative and quantitative syntheses (Figure 1). The principal exclusion criteria were nonrandomized designs, different EUS routes or comparators, and protocols or registries without usable results; the reasons for this are listed in Supplementary Table 7.

Figure 1 PRISMA 2020 flow diagram. Searches were completed on 6 August 2026; report-level exclusions and exact strategies appear in the Supplementary material.

The six trials were published from 2018 to 2026 (Table 1)[8-13]. Three older trials used tubular metal stents and three new trials used lumen-apposing metal stents for choledochoduodenostomy. The Park trial randomized 30 participants, but one participant in each arm subsequently underwent surgery; 28 participants remained in the published analysis. The Anderloni trial included 220 participants, and was the only trial in which pancreatitis occurred after EUS-BD. The pooled outcomes are summarized in Table 2.

Table 1 Characteristics of the included randomized trials.

Ref.	Setting	EUS/ERCP	Population	Ultrasound-guided route	Stents	Follow-up
Bang et al[9], 2018	United States	33/34	Pancreatic cancer	Choledochoduodenostomy	Tubular covered metal/covered metal	At least 6 months
Park et al[10], 2018	South Korea	14/14 analyzed	Predominantly pancreatic cancer	Transmural drainage	Partially covered tubular metal/metal	Median 95/147 days
Paik et al[8], 2018	South Korea; multicenter	64/61	Unresectable distal obstruction	Choledochoduodenostomy or hepaticogastrostomy	Tubular covered metal/covered metal	To death or dysfunction
Chen et al[11], 2023	Canada/France; multicenter	73/71	Borderline, locally advanced, or unresectable	Choledochoduodenostomy	Lumen-apposing metal/metal	1 year
Teoh et al[12], 2023	International; multicenter	79/76	Unresectable distal obstruction	Choledochoduodenostomy	Lumen-apposing metal/covered metal	1 year
Anderloni et al[13], 2026	Italy; multicenter	111/109	Distal obstruction; common bile duct at least 15 mm	Choledochoduodenostomy	Lumen-apposing metal/covered metal	6 months

EUS: Endoscopic ultrasound; ERCP: Endoscopic retrograde cholangiopancreatography.

Table 2 Summary of pooled outcomes.

Outcome	EUS-guided	ERCP	RR (95%CI)	I2	Interpretation
Technical success	350/374	304/365	1.08 (0.96-1.22)	64	No clear difference; inconsistent
Clinical success	331/369	322/359	1.01 (0.95-1.07)	0	Comparable
Procedure-related adverse events	52/374	63/365	0.83 (0.53-1.31)	1	Imprecise
Postprocedural pancreatitis	2/374	24/365	0.20 (0.05-0.77)	0	Lower with ultrasound-guided drainage
Stent dysfunction or reintervention	35/374	56/365	0.60 (0.36-1.01)	0	Possible reduction; uncertain

Risk ratios use Paule-Mandel random effects and modified Hartung-Knapp confidence intervals. CI: Confidence interval; EUS: Endoscopic ultrasound; ERCP: Endoscopic retrograde cholangiopancreatography; RR: Risk ratio.

Risk of bias

Newer multicenter trials were judged to have a low overall risk of the primary outcomes. Bang, Paik, and Park were rated as having some concerns as treatment and outcome assessment were open-label, reporting details were incomplete for at least one RoB 2 domain, and the small Park trial excluded one randomized participant per arm after surgery. No trial was judged to be at high risk for any of the outcomes. These judgments do not eliminate the substantial indirectness caused by expert center practice, device variation, and eligibility based on bile duct dilation.

Technical and clinical success

Technical success was achieved in 350/374 (93.6%) and 304/365 (83.3%) patients who underwent EUS-BD and ERCP, respectively. The pooled result did not establish a difference (RR: 1.08, 95%CI: 0.96-1.22; I² = 64%; Figure 2A). The three tubular-stent trials were homogeneous and centered on no effect (RR: 1.00, 95%CI: 0.85-1.18), whereas the lumen-apposing-stent trials favored EUS-BD numerically (RR: 1.18, 95%CI: 0.99-1.41). This pattern supports device and trial setting heterogeneity rather than a single universal technical effect.

Figure 2 Forest plot. A: Risk ratios use Paule-Mandel random effects and modified Hartung-Knapp confidence intervals; B: Clinical success according to each trial’s prespecified bilirubin-response definition; C: Procedure-related adverse events at the trial-reported early or 30-day window; D: Postprocedural pancreatitis. The plotted risk ratio uses a 0.5 continuity correction for zero cells; Peto and exact sensitivity estimates are reported in the text; E: Trial-defined stent dysfunction or reintervention. Definitions and follow-up horizons differed and are detailed in Supplementary Table 3.

Clinical success was reported in 331/369 (89.7%) and 322/359 (89.7%) patients who underwent EUS-BD and ERCP, respectively. The pooled RR was 1.01 (95%CI: 0.95-1.07; I² = 0%; Figure 2B). The definitions and denominators differed slightly, but the trial effects were consistent.

Procedure-related adverse events

Procedure-related adverse events occurred in 52/374 (13.9%) and 63/365 (17.3%) patients who underwent EUS-BD and ERCP, respectively. The pooled RR was 0.83 (95%CI: 0.53-1.31; I² = 1%; Figure 2C). The wide interval is compatible with benefits or harms, and the apparent statistical consistency does not overcome the differences in ascertainment windows and event definitions.

Postprocedural pancreatitis

Pancreatitis occurred in 2/374 (0.5%) EUS-BD and 24/365 (6.6%) ERCP participants, respectively. The random-effects RR was 0.20 (95%CI: 0.05-0.77; I² = 0%; Figure 2D). Sensitivity analyses were concordant (Supplementary Table 8): Peto odds ratio 0.17 (95%CI: 0.08-0.36) and conditional exact pooled odds ratio 0.08 (95%CI: 0.01-0.31; Fisher exact P < 0.001). The absolute risk difference was -6.0% (95%CI: -9.1% to -3.5%), corresponding to approximately one pancreatitis event avoided for every 17 patients treated with primary EUS-BD rather than ERCP. This estimate applies to the population and expertise of the trials.

Stent dysfunction or re-intervention

Stent dysfunction and re-intervention occurred in 35/374 (9.4%) and 56/365 (15.3%) patients who underwent EUS-BD and ERCP, respectively. The pooled RR was 0.60 (95%CI: 0.36-1.01; I² = 0%; Figure 2E). The point estimate suggests fewer repeat procedures after EUS-BD; however, the interval narrowly includes no effect, and the pooled outcome combines different definitions and follow-up periods.

DISCUSSION

In six randomized trials, primary transmural EUS-BD and ERCP achieved essentially the same clinical success. EUS-BD reduced the incidence of postprocedural pancreatitis from 6.6% to 0.5%, and this result remained under the three rare-event methods. However, the evidence does not show that EUS-BD reduces all procedure-related adverse events. Although the point estimate is favorable, it does not yet establish fewer stent dysfunctions or re-interventions.

The technical success results require more qualified reading than the original manuscript. Older tubular stent trials showed no advantage, whereas newer lumen-apposing stent trials tended to favor EUS-BD. The overall CI crossed no effect and heterogeneity was substantial. These differences may reflect the device design, bile duct diameter, case selection, local rescue algorithms, and a striking imbalance between extensive ERCP experience as well as the more recent EUS-BD experience in some trials. Therefore, the technical performance of these studies cannot be generalized to centers without interventional EUS expertise.

The finding of pancreatitis is biologically consistent because the transmural route avoids papillary cannulation and pancreatic duct instrumentation. However, it is not accurate to describe pancreatitis as eliminated. The 2026 Anderloni trial recorded events in both groups, although substantially fewer events occurred after EUS-guided choledochoduodenostomy[13]. The absolute estimate of approximately six fewer events per 100 procedures is more clinically informative than a relative measure alone. It also clarifies that the benefit must be weighed against EUS-specific complications, such as misdeployment, bile leak, bleeding, food impaction, or sump syndrome.

Our findings correspond with those of earlier randomized trial meta-analyses that reported comparable clinical success and uncertain overall adverse-event differences[22,23]. The 2026 individual-participant-data analysis of the three lumen-apposing-stent trials found higher technical success but similar 30-day adverse events, clinical success, and stent dysfunction[24]. The present review addresses a different question by combining all six eligible transmural randomized trials, including older tubular stent evidence and a newly published pancreatitis-focused trial. This broader scope increases the sample size along with clinical heterogeneity.

The most important clinical interpretation is the selection. Primary EUS-BD is a reasonable alternative when the distal duct is sufficiently dilated, an experienced interventional endoscopist and appropriate rescue pathways are available, and the patient anatomy favors a transmural route. The trials did not support the replacement of ERCP in any patient. Evidence is particularly limited for minimally dilated ducts, resectable diseases, altered anatomy, gastric outlet obstruction, community settings, and operators early in their learning curve.

Strengths and limitations

The revision resolves the prior full-text exclusion by adding Park, incorporating the 2026 Anderloni trial, restricting the intervention to anatomically comparable transmural drainage, reporting exact trial definitions, and using multiple rare-event analyses. The randomized-only design reduced confounding factors compared to mixed observational syntheses.

This study had several limitations. Six trials had limited power, particularly for uncommon harm and durability. The techniques, stents, operator experience, pancreatitis prophylaxis, adverse event definitions, and follow-ups differed. Some clinical success denominators included only technically successful procedures. The trials were concentrated in high-volume referral centers, and the lumen-apposing-stent trials required substantial bile duct dilation. This review was not prospectively registered. The update was not independently duplicated; direct subscription searches of EMBASE, Scopus, and Web of Science were unavailable; and unpublished investigators were not contacted. CENTRAL captured records sourced from EMBASE and the WHO ICTRP. Reference checking against recent comprehensive reviews did not identify another full randomized report with extractable data; however, residual publication or retrieval bias remains possible.

CONCLUSION

Primary transmural EUS-BD and ERCP provide comparable clinical success rates for malignant distal biliary obstructions. EUS-BD reduces the risk of pancreatitis; however, its technical advantages are device- and setting-dependent, and its effects on other adverse events and re-intervention remain uncertain. Current randomized evidence supports EUS-BD as an alternative to universal first-line replacement for ERCP in selected patients in expert centers.