Microbiology and antimicrobial resistance in biliary tract infections including acute and post-ERCP cholangitis: a systematic review and meta-analysis

Abstract

BACKGROUND:

Organisms cultured from bile in biliary tract infection, and their resistance, vary widely between settings, and pooled estimates to guide empirical therapy are lacking.

OBJECTIVE:

Estimate bile-culture yield, organism distribution, and resistance.

DESIGN:

Systematic review and random-effects meta-analysis of PubMed and reference lists

SETTING:

Reports spanned nine countries across Asia, Europe, Africa, and North America

METHODS:

We extracted ductal-bile culture data from observational studies published from 2015 to 2025 of biliary infection/cholangitis; noninfectious, gall bladder-only, microbiome-only, and nonextractable reports were excluded. DerSimonian–Laird models, subgroup analysis, and JBI appraisal were used.

MAIN OUTCOME MEASURE:

Culture yield; secondary organism proportions and resistance.

SAMPLE SIZE:

5147 patients (97% retrospective; 3% prospective); organisms: 35 267 observations.

RESULTS:

Yield was 77.3% (95% CI, 67.8–84.6); Gram-negative organisms 69.8% (65.6–73.7), E. coli, 27.4% (23.5–31.7), K. pneumoniae, 14.1% (12.3–16.2).

HETEROGENEITY:

I²=86.0%–97.7%; all P<.001.

RISK OF BIAS:

Fourteen reports were low risk, two moderate; exclusions minimally changed estimates.

CONCLUSION:

Bile-culture yield was high and Gram-negative organisms, particularly E. coli and K. pneumoniae, predominated, but estimates were highly heterogeneous and resistance could not be pooled; culture-directed therapy guided by local antibiograms is therefore preferable to a universal empirical regimen.

LIMITATIONS:

PubMed-only/open-access search, retrospective designs, inconsistent definitions, one dominant dataset.

REGISTRATION:

Retrospectively registered on OSF (89wqs; DOI: 10.17605/OSF.IO/89WQS).

INTRODUCTION

Endoscopic retrograde cholangiopancreatography (ERCP) is fundamental to the treatment of biliary obstruction. Infective complications are among its most important adverse events and include acute or post-procedural cholangitis, bacteremia, and sepsis.¹ Acute cholangitis is a potentially life-threatening infection of the biliary tree in which drainage and anti-microbial therapy are complementary rather than interchangeable interventions.²

For much of the twentieth century, normal bile was considered sterile. Culture-independent studies now detect low-biomass microbial signals in bile, but whether these always represent stable resident communities rather than transient colonization or contamination remains debated.³ Bile acids exert antimicrobial pressure; obstruction, reflux, and instrumentation can nevertheless impair ductal clearance and permit clinically important bacteriobilia.

In this review, biliary tract infection is an operational umbrella for suspected or clinically defined infection involving the bile duct and sampled through ERCP, percutaneous transhepatic cholangiodrainage (PTCD), or an equivalent ductal route. Acute cholangitis denotes infection with biliary obstruction and systemic inflammation; post-ERCP cholangitis occurs after instrumentation. Choledocholithiasis alone is not an infection, and gall bladder-only cholecystitis was excluded. Community-acquired cases were distinguished conceptually from healthcare-associated or post-intervention infection because prior hospitalization, antibiotics, stents, sphincterotomy, and drainage can alter resistance risk.⁴

Enterobacterales are Gram-negative bacilli that include Escherichia coli and Klebsiella pneumoniae; enterococci are Gram-positive cocci with different taxonomy, intrinsic susceptibility, and clinical significance. Recent non-biliary studies of ESKAPE organisms and K. pneumoniae illustrate that biofilm formation, efflux activity, reduced outer-membrane permeability, and β-lactamase production can coexist with multidrug, extensive-drug, or pan-drug resistance.^5,6 These studies provide mechanistic context rather than biliary prevalence estimates.

Individual biliary studies report widely varying organism and resistance proportions because geography, study period, case definition, healthcare exposure, and sampling route differ. We therefore estimated patient-level bile-culture yield and pooled proportions of Gram-negative bacteria, E. coli, K. pneumoniae, and polymicrobial cultures; summarized Enterococcus and resistance phenotypes when compatible denominators were unavailable; and tested the robustness of organism estimates to the dominant surveillance dataset.

METHODS

Protocol, registration, and reporting

This review was conducted and reported in accordance with PRISMA 2020.7 The review was retrospectively registered on the Open Science Framework (OSF 89wqs; DOI: 10.17605/OSF.IO/89WQS).

Operational definitions and eligibility

Eligible reports included retrospective or prospective observational cohorts and surveillance datasets that described bacteria cultured from ductal bile in patients with biliary tract infection, acute cholangitis, post-ERCP cholangitis, or suspected infection. Acute cholecystitis limited to gallbladder bile, uncomplicated choledocholithiasis without suspected infection, preoperative colonization before pancreatoduodenectomy, primary-sclerosing-cholangitis microbiome studies, pyogenic liver abscess, animal studies, reviews, guidelines, culture-independent 16S rRNA sequencing without culture data, and reports without extractable counts were excluded.

At least one extractable numerator and denominator was required for culture yield, organism distribution, polymicrobial culture, or a resistance phenotype. Only retrievable open-access full texts were eligible because subscription access was unavailable. This accessibility criterion was a pragmatic restriction, not a marker of scientific quality, and could systematically alter the geography, period, and institutional composition of the evidence base.

Information sources and search

PubMed was searched from 1 January 2005 to 2 July 2026 using bile/biliary, culture/microbiology, cholangitis/biliary-infection/post-ERCP, and antimicrobial-resistance term blocks. Reference lists of included reports and relevant reviews were hand-searched. The verbatim historical query export was not retained; Supplementary Appendix S1 provides the dated, field-tagged Boolean reconstruction and notes that it may not reproduce the historical result count exactly.

Study selection and data extraction

Records were screened by title and abstract, and retrievable reports were assessed against the eligibility criteria. Screening of titles, abstracts, and full texts was performed by A.S. M.A.S. and A.A. independently verified the inclusion and exclusion of reports against the eligibility criteria, with any disagreements resolved through discussion with A.E. We extracted country, design, clinical population, healthcare exposure, sampling route, numbers of distinct individuals, procedures or samples, positive cultures, isolates, Gram-negative bacteria, E. coli, K. pneumoniae, Enterococcus, polymicrobial cultures, and reported resistance phenotypes. Procedure and sample counts were kept separate from patient counts. Reports from the same centre and period were assessed for cohort overlap and counted once in totals.

Resistance categories were extended-spectrum β-lactamase (ESBL) production, production, carbapenem resistance (CRE), fluoroquinolone resistance, thirdgeneration-cephalosporin resistance, vancomycin-resistant Enterococcus (VRE), multidrug-resistant (MDR), extensively drug-resistant (XDR), and pandrug-resistant (PDR). International definitions describe MDR as non-susceptibility to at least one agent in at least three antimicrobial categories, XDR as nonsusceptibility to at least one agent in all but two or fewer categories, and PDR as nonsusceptibility to all agents in all categories.⁸ We retained each article's actual definition and denominator rather than reclassifying incomplete panels. Values reported only as percentages remain labelled as such; back-calculated counts are identified and are not presented as exact source-reported numerators.

Risk-of-bias assessment

Methodological quality of each prevalence estimate was evaluated with the nine-domain Joanna Briggs Institute (JBI) checklist for prevalence studies.⁹ The checklist was selected because this synthesis pooled cross-sectional proportions from each report, irrespective of the parent cohort or surveillance design, and did not estimate causal exposure effects. Its domains address the sampling frame, coverage, measurement, denominators, and statistical analysis relevant to these proportions; it should not be interpreted as a complete causal-bias assessment of every parent study. Eight or nine yes responses indicated low risk, six or seven moderate risk, and fewer than six high risk.

Statistical analysis

Single proportions were logit-transformed and pooled with DerSimonian–Laird random-effects models, then back-transformed.¹⁰ Freeman–Tukey double-arcsine models were sensitivity analyses.¹¹ We report pooled proportions with 95% confidence intervals, Cochran Q, I², τ², and 95% prediction intervals. Organism proportions used total isolates; polymicrobial proportions used culture-positive samples. Exact polymicrobial numerators were used when available; numerators reconstructed from a reported rounded percentage were marked approximate and tested in an exact-count-only sensitivity analysis. Babekir's mutually exclusive Results categories summed to 33 organism-positive cases and were used for culture yield, GNB, and E. coli; a sensitivity analysis excluded the report because its Methods sentence said all 50 cultures were positive. The primary culture-yield model required a distinct patient denominator, so the Gromski procedure-level series was excluded from that model and retained descriptively.

We repeated the Gram-negative, E. coli, and K. pneumoniae models after excluding Li et al, the 25 573-isolate surveillance dataset. Exploratory subgroup models compared clinically defined infection with broader or unselected biliary sampling for culture yield and East Asia with other regions when at least two studies contributed to each group. Setting and sampling-route meta-regression was not performed because categories were incomplete, overlapping, and too sparse. MANOVA was not applicable to aggregate single-proportion data without patient-level multivariate observations. Funnel plots and asymmetry tests were not prespecified and were not used because their interpretation is unreliable for highly heterogeneous single-proportion models under a known open-access selection restriction, even when 10–12 studies contributed. A post hoc risk-of-bias sensitivity repeated each primary model after excluding the two moderate-risk reports, Kar and Babekir; for endpoints to which Babekir did not contribute, this removed Kar only.

RESULTS

Study selection

The search identified 2544 PubMed records and 7 additional records. No duplicate record was logged; 2551 records were screened and 2485 were excluded by title/abstract. Sixty-six reports were sought, 18 could not be retrieved under the open-access criterion, and 48 were assessed in full. Thirty-two were excluded (reviews/guidelines, 8; wrong population, 12; 16S rRNA sequencing without culture, 2; no extractable counts, 5; intervention trial without organism distribution, 1; other prespecified eligibility reasons, 4), leaving 16 included reports (Figure 1; Supplementary Appendix S2).

Figure 1.

PRISMA 2020 flow diagram. Of 2551 screened records, 2485 were excluded at title/abstract screening. Sixtysix reports were sought, 18 were not retrieved, 48 were assessed, 32 were excluded with stated reasons, and 16 reports were included; 15 contributed to at least one quantitative synthesis.

Study characteristics

The 16 reports^12-27 were published from 2015 to 2025 and represented nine countries: China, the United States, Romania, Germany, India, South Korea, Egypt, France, and Sudan. Thirteen reports supplied source-resolved organism totals, comprising 35 267 organism observations; Li et al contributed 25 573 (72.5%) and no patient denominator. A combined patient total was not calculated because the reports mixed full cohorts, restricted infection strata, positive-culture-only series, repeated procedures, and studies without a person count. For example, Gromski reported 721 procedures in 615 people, Xing reported 80 infection cases within a 528-person cohort, Lee cultured 138 of 182 people, and the two Miutescu reports described the same 488-person cohort (Table 1). Study-frame and denominator reconciliation is detailed in Supplementary Appendix S3.

Table 1.

Characteristics and principal microbiological findings of the 16 included reports.

Study	Country	Design/setting	Participants/samples	Culture yield	Organism data	AMR / polymicrobial data
Gromski 202212	USA	Retrospective ERCP series	615 people; 721 aspirates	662/721 procedures (91.8%); not in patient model	E. coli 269/662 and Klebsiella spp. 295/662 positive cultures (K. pneumoniae 201; K. oxytoca 94); not isolate proportions	High-concern MDRO composite 155/662 (23.4%); ESBL 7.9% and CRE 3.6% of Enterobacteriaceae (numerators NR); polymicrobial ≈540/662 (81.6%)
Gu 202013	China	Retrospective; mixed biliary disease	1339 people; 826 isolates	738/1339 (55.1%)	GNB 619/826; E. coli 312/826; Enterococcus 109/826; K. pneumoniae 85/826	High ampicillin/ceftriaxone resistance; exact phenotype denominators incomplete; polymicrobial 86/738 (11.7%)
Zhao 202214	China	Retrospective BTI	277 people; 280 isolates	267/277 (96.4%)	GNB 215/280; E. coli 85/280; K. pneumoniae 47/280	ESBL E. coli 12/85 (14.1%); ESBL K. pneumoniae 3/47 (6.4%); polymicrobial 9/267 (3.4%)
Zhang 202415	China	Retrospective; choledocholithiasis + BTI	880 people; 1216 isolates	795/880 (90.3%)	GNB 685/1,216; E. coli 370/1216; K. pneumoniae 149/1216; Enterococcus 317/1216	ESBL E. coli 163/370 (44.1%); ESBL K. pneumoniae 58/149 (38.9%); polymicrobial 330/795 positive cultures (41.5%)c
Li 202416	China	Provincial surveillance	25 573 isolates; no patient N	NR	GNB 18,134/25 573; E. coli 8181/25 573; K. pneumoniae 3247/25 573	Carbapenem-resistant E. coli 1.1% (numerator NR); cephalosporin resistance reported
Chen 202417	China	Retrospective clinical BTI	1556 people; 3490 isolates	1125/1556 (72.3%)	GNB 2,340/3,490; E. coli 875/3490; K. pneumoniae 535/3490; Enterococcus ≥677/3490	ESBL E. coli 37.21% and K. pneumoniae 17.56% (exact numerators NR); polymicrobial 1,035/1,125 (92.0%)
Hassan 202518	Egypt	Prospective acute cholangitis	105 people; 61 bile microorganisms	50/105 (47.6%)	GNB 47/61; E. coli 11/61; K. pneumoniae 21/61; fungi 5/61	MDR 29/56 bacteria; ESBL 25/47 GNB; carbapenem resistance 22/47 GNB; polymicrobial ≈8/50 (16.0%)
Kar 202319	India	Retrospective biliary samples	234 people; 209 bacterial isolatesd	163/234 (69.7%)	GNB 168/209; E. coli 83/209; K. pneumoniae 28/209	MDR 105/209; resistance to ESBL antibiotics 142/209; carbapenem resistance 96/209; polymicrobial 44/163 (27.0%)
Kruis 202020	Germany	Retrospective acute cholangitis	348 people; 423 included positive cultures; 1066 isolates	NR for all sampled people	Enterococcus ≈330/1,066 (31%); Enterobacterales ≈314/1066 (29.5%)	Polymicrobial ≈292/423 (69%); empirical coverage 51%
Reuken 201721	Germany	Retrospective acute cholangitis	83-person clinical subset; 531 positive cultures; 1764 isolates	NR for all sampled people	E. coli 282/1764; Enterococcaceae 440/1764	MDR pathogen 24/83 people (28.9%); VRE reported; polymicrobial ≈414/531 (78%)
Miutescu-L 202322	Romania	Retrospective; same 488-person cohort as next row	488 people/samplesa	340/488 (69.7%)	Organism presence reported per sample, not as an isolate distribution	MDR 97/488; ESBL 60/488; CRE 22/488; polymicrobial 138/340 positive cultures (40.6%)
Miutescu-D 202423	Romania	Retrospective; overlapping cohorta	Same 488 peoplea	Not counted again	Overlapping sample-presence data; not pooled again	Polymicrobial 150/340 conflicts with 138/340 in the earlier report; same overall MDR 97/488, ESBL 60/488, and CRE 22/488
Cozma 202524	Romania/France	Prospective multicentre	205 people (143 Results/62); 257 positive cultures; 435 component isolatesb	Total sampled denominator NR	E. coli 96/435; Klebsiella spp. 54/435 across both centres	MDR 51/241 Romania and 15/194 France; polymicrobial 97/257 positive cultures (37.7%)
Babekir 202325	Sudan	Prospective obstructive jaundice	50 people; 33 organism-positive casese	Results: 33/50 (66.0%); Methods says all positivee	GNB 32/33; E. coli 18/33; Klebsiella spp. 7/33; P. aeruginosa 6/33; C. freundii 1/33; S. aureus 1/33	Drug-specific susceptibility reported graphically; tested denominators NR; GNB/E. coli pooled with Babekir-excluded sensitivities
Xing 202226	China	Retrospective malignant obstruction after PTCD	528-person cohort; 80 infection cases; 93 isolates	58/80 infection cases (72.5%)	GNB 51/93; E. coli 15/93; K. pneumoniae 12/93	Gram-positive cocci susceptible to vancomycin/linezolid; exact resistance counts NR
Lee 201527	South Korea	Retrospective cholangitis	182-person cohort; 138 sampled; 221 isolates	130/138 sampled people (94.2%)	GNB 141/221; E. coli 47/221; Klebsiella spp. 26/221	Polymicrobial ≈64/130 positive cultures (49.2%); resistance phenotype denominator NR

N, denominator; AMR, antimicrobial resistance; BTI, biliary tract infection; CRE, carbapenem-resistant Enterobacterales; ERCP, endoscopic retrograde cholangiopancreatography; ESBL, extended-spectrum β-lactamase; GNB, Gram-negative bacteria; GPC, Gram-positive cocci; MDR, multidrug resistance; NR, not reported or not extractable; PTCD, percutaneous transhepatic cholangiodrainage; VRE, vancomycin-resistant enterococci. Approximate signs identify counts reconstructed from rounded percentages. Organism values are isolate proportions unless a culture- or sample-occurrence unit is stated.

^aMiutescu-L 2023 and Miutescu-D 2024 describe the same 488-person cohort and were counted once.

^bCozma reported 143 Romanian participants in results (144 in the abstract) and 241+194=435 component isolates while also stating 436 total etiologic agents; the internally reconcilable Results values are displayed.

^cZhang reported 330 polymicrobial samples among all 880 samples; the synthesis uses the same exact numerator among 795 culture-positive samples.

^dKar gives conflicting GNB class totals in its results and resistance sections; the microbiology result of 168/209 was retained.

^eBabekir's Methods says all cultures were positive, whereas Results reports 17/50 with no growth. The mutually exclusive results categories sum to 33 organism-positive cases; these were retained for culture yield, GNB, and E. coli and tested by excluding the report. Klebsiella spp. was not relabelled as K. pneumoniae. Chen's stated 3490-strain denominator exceeds its displayed class components by 12 and was retained as reported.

Patient-level bile-culture yield

Ten studies with distinct patient denominators contributed 3699 positive cultures among 5147 sampled people. The pooled bile-culture yield was 77.3% (95% CI 67.8–84.6; I²=97.7%; τ²=0.549; 95% prediction interval 36.0–95.4%). This is a sampling yield, not the prevalence of clinical infection. In an exploratory source-based classification, yield was 82.1% (71.4–89.3%; 2,765/3,524; 7 studies) in clinically defined infection and 63.3% (51.4–73.7%; 934/1,623; 3 studies) in broader or unselected sampling (interaction P=.014). The classification was post hoc and the subgroups remained highly heterogeneous. Gromski et al reported a procedure-level yield of 662/721 (91.8%) and was excluded from the patient model.

Organism distribution

Gram-negative bacteria comprised 69.8% of isolates (22 432/32 002; 95% CI 65.6–73.7; 10 studies; I²=94.9%).^13-19,25-27 E. coli accounted for 27.4% (10 375/34 201; 23.5–31.7; 12 studies; I²=96.5%),^{13-19,21,24-27} and K. pneumoniae for 14.1% (4124/31 748; 12.3–16.2; 8 studies; I²=86.0%).^13-19,26 Babekir's mutually exclusive results categories reconstructed 33 organism-positive cases and were used for the Gram-negative and E. coli endpoints, with source-exclusion sensitivities because the Methods incorrectly stated that all 50 cultures were positive. Its genus-level Klebsiella count was not used for the species-specific K. pneumoniae endpoint; other reports limited to Klebsiella spp. were also excluded from that endpoint. These pooled estimates are shown in Figure 2.

Figure 2.

Forest plots of patient-level bile-culture yield and the proportions of Gram-negative bacteria, Escherichia coli, Klebsiella pneumoniae, and polymicrobial cultures. Study rows show events/denominator and 95% confidence intervals; blue diamonds show random-effects pooled estimates. Symbols identify rounded-percentage reconstructions, a numerator re-expressed per positive culture, the internally inconsistent Babekir denominator, and the earlier of two overlapping Miutescu reports. Enterococcus was not pooled because reports mixed genus- and species-level numerators.

Enterococcus was not pooled because the available reports mixed genus totals with E. faecium-only counts. Comparable descriptive genus-level proportions were 13.2% in Gu et al, 26.1% in Zhang et al, at least 19.4% in Chen et al, and approximately 31.0% in Kruis et al. The variation may be clinically relevant, but it does not support a single pooled genus estimate without a complete harmonized re-extraction.

Polymicrobial cultures

The 12-study random-effects summary was 43.7% (3057/5481; 95% CI 26.4–62.7; I²=99.2%; τ²=1.833; prediction interval 3.2–94.7%).^{12-15,17-22,24,27} Study-specific estimates ranged from 3.4% to 92.0%. Exact counts were available in seven reports; five additional numerators were reconstructed from reported rounded percentages and are marked approximate. Zhang's exact numerator of 330 was re-expressed from all 880 samples (37.5%) to the 795 culture-positive samples (41.5%). To avoid double counting, the earlier Miutescu report supplied the primary 138/340 count; substituting the overlapping later report's conflicting 150/340 count changed the pooled estimate to 44.0%. Restricting the model to exact numerators gave 32.7% (13.7–59.8; seven studies). These estimates are descriptive summaries of markedly heterogeneous definitions and settings, not a universal prevalence.

Antimicrobial resistance

Resistance data could not be validly pooled. Studies varied in organism denominator, drug panel, laboratory breakpoint, patient-versus-isolate unit, and MDR definition; several supplied only percentages. Supplementary Table S1 therefore reports the exact numerator and denominator when available, otherwise NR with the reported percentage and denominator type. For example, Zhao et al reported ESBL in 12/85 E. coli isolates (14.1%) and 3/47 K. pneumoniae isolates (6.4%). Reuken et al reported MDR pathogens in 24/83 patients (28.9%). No included report supplied an extractable XDR numerator and denominator under a reproducible definition.

Sensitivity and exploratory subgroup analyses

Excluding Li et al changed the pooled Gram-negative estimate from 69.8% to 70.2% (95% CI 64.2–75.6), E. coli from 27.4% to 26.9% (22.1–32.3), and K. pneumoniae from 14.1% to 14.8% (12.1–18.0). These nested analyses show modest point-estimate changes; they are not tests of statistical significance. Source-exclusion and overlap-substitution sensitivities are reported in Supplementary Table S2. Exploratory region and infection-definition contrasts were post hoc, small, and remained heterogeneous; they are presented as uncertainty analyses rather than evidence of regional equivalence or causality. Freeman–Tukey estimates were close to the primary models (Table 2; Supplementary Table S2). A risk-of-bias sensitivity excluding both moderate-risk reports yielded 79.3% culture yield (95% CI, 68.6–87.1), 67.9% Gram-negative bacteria (63.5–72.1), 25.1% E. coli (21.2–29.5), 14.2% K. pneumoniae (12.3–16.5), and 45.4% polymicrobial cultures (27.0–65.2); substantial heterogeneity persisted (Supplementary Table S2).

Table 2.

Random-effects estimates and sensitivity analyses.

Outcome	k	Events/N	Pooled % (95% CI)	Q (df)	I2 (%)	τ2	95% PI (%)	FT (%)
Primary: patient-level bile-culture yield	10	3699/5147	77.3 (67.8–84.6)	394.2 (9)	97.7	0.549	36.0–95.4	75.4
Primary: Gram-negative bacteria	10	22 432/32 002	69.8 (65.6–73.7)	177.6 (9)	94.9	0.071	54.6–81.6	70.8
Primary: Escherichia coli	12	10 375/34 201	27.4 (23.5–31.7)	313.2 (11)	96.5	0.110	14.8–45.0	27.6
Primary: Klebsiella pneumoniae	8	4124/31 748	14.1 (12.3–16.2)	50.0 (7)	86.0	0.034	9.1–21.2	13.9
Primary: polymicrobial culture	12	3057/5481	43.7 (26.4–62.7)	1388.4 (11)	99.2	1.833	3.2–94.7	45.0
Sensitivity excluding Li: Gram-negative bacteria	9	4298/6429	70.2 (64.2–75.6)	129.1 (8)	93.8	0.137	48.0–85.8	71.5
Sensitivity excluding Li: Escherichia coli	11	2194/8628	26.9 (22.1–32.3)	208.5 (10)	95.2	0.163	12.4–49.1	27.2
Sensitivity excluding Li: Klebsiella pneumoniae	7	877/6175	14.8 (12.1–18.0)	37.6 (6)	84.1	0.072	7.6–27.0	14.7
Sensitivity: exact polymicrobial numerators only	7	1739/3685	32.7 (13.7–59.8)	948.3 (6)	99.4	2.242	0.8–96.8	34.9

k, number of studies; CI, confidence interval; PI, prediction interval; FT, Freeman–Tukey sensitivity estimate. All Q tests had P<.001. Primary models used DerSimonian–Laird random effects on logit-transformed proportions. Patient-level culture yield excludes Gromski because 721 ERCP aspirates arose from 615 individuals. Babekir's detailed results categories were used for culture yield, GNB, and E. coli; sensitivities excluding that internally inconsistent report appear in Table S2. The earlier Miutescu report supplied the overlapping cohort's primary polymicrobial count; substitution of the later count appears in Table S2. Li-excluded analyses are nested sensitivities, not significance tests. Enterococcus was not pooled.

Risk of bias

Fourteen reports were rated low risk for their prevalence estimates and two moderate risk (Table 3). Common limitations were retrospective single-centre sampling and incomplete coverage reporting. This study-level appraisal does not offset review-level bias introduced by the single-database, open-access eligibility criterion.

Table 3.

JBI prevalence-estimate appraisal of the 16 included reports.

Study	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Yes	Overall
Gromski 202212	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Gu 202013	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Zhao 202214	Y	U	Y	Y	Y	Y	Y	Y	Y	8/9	Low
Zhang 202415	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Li 202416	Y	Y	Y	Y	Y	Y	Y	Y	U	8/9	Low
Chen 202417	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Hassan 202518	Y	Y	U	Y	Y	Y	Y	Y	Y	8/9	Low
Kar 202319	Y	U	Y	Y	Y	Y	Y	Y	U	7/9	Moderate
Kruis 202020	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Reuken 201721	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Miutescu-L 202322	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Miutescu-D 202423	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Cozma 202524	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low
Babekir 202325	Y	U	U	Y	Y	Y	Y	Y	U	6/9	Moderate
Xing 202226	Y	Y	U	Y	Y	Y	Y	Y	Y	8/9	Low
Lee 201527	Y	Y	Y	Y	Y	Y	Y	Y	Y	9/9	Low

Q1 sampling frame; Q2 sampling method; Q3 sample size; Q4 participants and setting; Q5 coverage; Q6 valid identification; Q7 reliable measurement; Q8 appropriate analysis; Q9 response/coverage rate. Y, yes; U, unclear. Low, 8–9 yes responses; moderate, 6–7. This appraisal addresses bias in the extracted prevalence/proportion estimates and does not replace a causal-bias assessment of each parent observational design. Review-level open-access and single-database selection bias is discussed separately.

DISCUSSION

This systematic review quantitatively synthesizes bile-culture yield and organism proportions across biliary infection and acute or post-ERCP cholangitis. The analysis found frequent culture yield and a predominance of Gram-negative organisms, especially E. coli and K. pneumoniae, while indicating that case definitions and sampling contexts are not interchangeable. The wide prediction intervals are as important as the pooled point estimates.

The procedure-to-patient correction illustrates why denominator discipline matters. A bile aspirate obtained during ERCP is a sampling event; repeated procedures do not create additional individuals. Similarly, culture yield among patients selected for suspected infection cannot be interpreted as population infection prevalence. The exploratory 82.1% versus 63.3% contrast is clinically plausible, but it was post hoc and the subgroups were too heterogeneous to establish a general rule.

The organism findings support empirical Enterobacterales coverage when cholangitis is suspected, but the evidence does not justify a universal regimen. Enterococcus was frequently reported but could not be pooled after genus/species inconsistencies were identified. Bile-salt tolerance was not measured in the included studies, so organism frequency cannot be converted into a natural bile-salt-resistance ranking. Prior instrumentation, stents, healthcare exposure, and recent antibiotics remain clinically relevant risk markers, but resistance evidence here is descriptive rather than quantitatively causal.⁴

The Li-excluded analyses reduce concern that the organism point estimates simply reproduce one provincial surveillance network, although Li still exposes a structural limitation: 72.5% of source-resolved isolates came from one region and surveillance system. Similar pooled proportions with and without that dataset do not demonstrate regional equivalence, and local biliary antibiograms remain essential.

Clinically, biliary drainage and other source control should not be delayed while microbiology is clarified. When bile is sampled, exact organism and susceptibility results can support early de-escalation. Tokyo Guidelines 2018 emphasize severity, healthcare exposure, and local susceptibility data.²⁸ Short courses may be adequate after successful drainage in selected uncomplicated patients,²⁹ while prophylaxis before ERCP remains context-dependent.^30,31 Device reprocessing and design also contribute to prevention of transmission.³² These applications are stewardship implications of the evidence, not pooled treatment-effect estimates.

The search was limited to PubMed and retrievable open-access reports. Eighteen reports were not retrieved, so their populations and results could not be compared quantitatively with included studies. This creates material selection, publication, language, geographical, and institutional bias. The exact historical PubMed query export was not archived; Supplementary Appendix S1 provides a reconstructed strategy that may not reproduce the historical result count exactly.

Most reports were retrospective and single-centre, with substantial East Asian representation. Clinical infection definitions, community/healthcare exposure, route, timing, and culture technique were inconsistently reported. These limitations produced I² values of 86%–99% and wide prediction intervals. Exploratory subgroups were post hoc and underpowered, and setting/route meta-regression would have overfit sparse, non-mutually-exclusive categories.

Several reports required denominator reconciliation. Five polymicrobial numerators were reconstructed from rounded percentages; one procedure series involved repeated sampling; the two overlapping Miutescu reports gave conflicting polymicrobial counts; Babekir's Methods said all cultures were positive although Results reported 17 with no growth; Kar reported inconsistent Gram-negative class totals; and Cozma reported 143 Romanian patients in Results but 144 in the abstract, while centre-level isolate components totalled 435 against a stated total of 436. These discrepancies are footnoted rather than silently harmonized and are tested where quantitative alternatives were possible. Li et al contributed 72.5% of source-resolved isolates, although Li-excluded point estimates were similar.

Resistance phenotypes could not be pooled because organisms, breakpoints, panels, definitions, and denominators differed; no extractable XDR endpoint was available. The JBI prevalence checklist addresses the pooled proportions but not every causal-bias domain of the parent designs. Only 8–12 studies informed each primary outcome. Funnel asymmetry tests were not used because they were not prespecified and would be difficult to interpret under extreme heterogeneity and a known access-based selection restriction. The protocol was registered retrospectively, which is a further limitation because prospective registration is best practice. Against these limitations, the review has several strengths: eligibility criteria and outcome definitions were applied uniformly, exact numerators and denominators were extracted wherever they were reported, denominator discrepancies were reconciled and tested rather than silently harmonized, and every pooled estimate is accompanied by a prediction interval and a risk-of-bias sensitivity analysis.

Clinical significance

For clinicians and endoscopists, the consistent signal is the need to cover Enterobacterales initially when cholangitis is suspected while accounting for severity, prior instrumentation, healthcare exposure, and the local antibiogram. Drainage remains central. Bile-culture results should be used to narrow therapy promptly; broader anti-ESBL, antipseudomonal, or enterococcal coverage should be reserved for patients whose individual and institutional risk supports it. Figure 3 summarizes this evidence-to-care pathway.

Figure 3.

Evidence-to-care pathway for suspected biliary infection: obstruction or instrumentation, clinically important bacteriobilia, source control and culture, and risk-aware antimicrobial treatment guided by exact susceptibility results and the local biliary antibiogram.

CONCLUSION

Patient-level bile-culture yield was high, and Gram-negative bacteria—particularly E. coli and K. pneumoniae—were common across the included reports. However, the magnitude varied widely by selection and setting; the polymicrobial summary was highly heterogeneous; and Enterococcus and antimicrobial-resistance burdens could not be pooled consistently. These findings support prompt source control, ductal culture when obtained, risk-aware empirical therapy, and deescalation guided by exact susceptibility results and the local biliary antibiogram rather than a universal regimen.

Future prospective multicentre studies should use standard cholangitis definitions; separate community, healthcare-associated, post-ERCP, and long-term-drainage cohorts; report sampling route and repeat procedures; provide exact AMR numerators, denominators, breakpoints, and MDR/XDR definitions; and link antibiotics and drainage to clinical outcomes. Future reviews should search multiple databases, secure full-text access, archive complete strategies and exclusion logs, and register protocols publicly.

SUPPLEMENTS

Appendix S1. Reconstructed PubMed search strategy and reproducibility limitation

Search date: 2 July 2026. Coverage: 1 January 2005 through 2 July 2026. The review log recorded 2,544 PubMed records and 7 records from reference-list searching. The verbatim historical PubMed export and line-by-line query were not retained. The field-tagged strategy below was reconstructed from the retained term blocks and may not reproduce the historical result count exactly.

$$\left(\right(\text{“Bile”[Mesh] OR bile[Title/Abstract] OR biliary[Title/Abstract]}\left)\text{ AND }\right(\text{culture*[Title/Abstract] OR microbiolog*[Title/Abstract] OR bacteriolog*[Title/Abstract]}\left)\text{ AND }\right(\text{cholangitis[Title/Abstract] OR “biliary tract infection”[Title/Abstract] OR “post-ERCP”[Title/Abstract] OR choledocholithiasis[Title/Abstract]}\left)\right)\text{ AND }(\text{“2005/01/01”[Date – Publication]}:\text{“2026/07/02”[Date – Publication]})$$

Resistance-focused screening terms used as supplementary combinations were: (“Drug Resistance, Microbial”[Mesh] OR antimicrobial resistan*[Title/Abstract] OR antibiotic resistan*[Title/Abstract] OR ESBL[Title/Abstract] OR betalactamase*[Title/Abstract] OR multidrug resistan*[Title/Abstract] OR MDR[Title/Abstract]). Reference lists of included reports and relevant reviews were searched manually. No language filter was documented. Because the archived query is incomplete, the search is not claimed as fully reproducible.

Appendix S2. Selection arithmetic

Stage	n
PubMed records	2,544
Hand-search records	7
Duplicates logged	0
Records screened	2,551
Title/abstract exclusions	2,485
Reports sought	66
Reports not retrieved	18
Reports assessed	48
Full-text exclusions	32
Reports included	16

Full-text exclusion reasons (n=32): reviews/guidelines, 8; wrong population, 12; 16S rRNA sequencing without culture data, 2; no extractable counts, 5; intervention trial without organism distribution, 1; other prespecified eligibility reasons, 4. The historical report-level exclusion log did not preserve a more granular description of the final four reports.

Appendix S3. Denominator reconciliation.

No combined patient total was calculated. Study frames included full enrolled cohorts, restricted clinicalinfection strata, positive-culture-only series, repeated procedures, and reports without a person denominator. Gromski reported 721 ERCP aspirates from 615 people; Xing's microbiology analysis used 80 infected patients within 528 enrolled; Lee cultured 138 of 182 patients; and the two Miutescu reports describe the same 488-person cohort. Across 13 reports with source-resolved organism totals, 35 267 source-resolved organism observations were tabulated. This total includes 33 mutually exclusive organism-positive cases reconstructed from Babekir's detailed Results, whose Methods sentence incorrectly said that all 50 cultures were positive. Li supplied 25 573 isolates without a patient denominator, and Reuken's 1764 isolates arose from 531 positive cultures rather than the 83-person clinical subset. These units are reported separately and are not presented as one linked patient-isolate cohort.

Supplementary Table S1.

Structured antimicrobial-resistance reporting.

Study	Phenotype	Numerator	Denominator	Reported value	Definition / limitation
Gromski 2022	ESBL; CRE	NR	Enterobacteriaceae	7.9%; 3.6%	Study laboratory criteria; exact numerator not reported in extractable form
Gu 2020	Ampicillin/ceftriaxone resistance	NR	Organism-specific isolates	Reported as high	Drug-specific data; no harmonized MDR denominator
Zhao 2022	ESBL E. coli	12	85 E. coli isolates	14.1%	Phenotypic ESBL as reported
Zhao 2022	ESBL K. pneumoniae	3	47 K. pneumoniae isolates	6.4%	Phenotypic ESBL as reported
Zhang 2024	ESBL E. coli / K. pneumoniae	163 / 58	370 E. coli / 149 K. pneumoniae isolates	44.05% / 38.93%	Exact species-specific counts reported in Table 3
Li 2024	Carbapenem-resistant E. coli	NR	8,181 E. coli isolates	1.1%	Surveillance breakpoint; numerator NR
Chen 2024	ESBL E. coli / K. pneumoniae	NR	875 / 535 species isolates	37.21% / 17.56%	Study-reported percentages; exact numerators NR
Hassan 2025	MDR; ESBL; carbapenem resistance	29 / 25 / 22	56 bacterial isolates / 47 GNB / 47 GNB	51.8%; 53.2%; 46.8%	Exact counts from Results and Table 4; fungal isolates excluded from MDR denominator
Kar 2023	MDR; resistance to ESBL antibiotics; carbapenem resistance	105 / 142 / 96	209 bacterial isolates for each	50.24%; 67.94%; 45.93%	Source wording does not establish that 142 isolates were phenotypic ESBL producers
Reuken 2017	MDR pathogen	24	83 patients	28.9%	Patient-level presence of MDR pathogen
Miutescu-L 2023	MDR; ESBL; CRE	97 / 60 / 22	488 patients for each	19.9%; 12.3%; 4.5%	Exposure-stratum counts are also reported; the overlapping report uses the same cohort
Miutescu-D 2024	MDR; ESBL; CRE	97 / 60 / 22	488 patients for each	19.9%; 12.3%; 4.5%	Published exposure strata do not fully sum to the stated overall MDR and CRE totals
Cozma 2025	MDR	51 / 15	241 Romanian / 194 French isolates	21.16% / 7.73%	Results and fractions used; the French abstract percentage of 7.21% is inconsistent
Babekir 2023	Drug-specific susceptibility; graphical resistance profile	NR	Organism-specific tested isolates not extractable	Meropenem susceptible 54%; ciprofloxacin susceptible 46%	Exact tested numerators and denominators cannot be assigned safely
All reports	XDR	NR	NR	Not extractable	No reproducible XDR numerator/ denominator was available

NR, not reported or not extractable; CRE, carbapenem-resistant Enterobacterales; ESBL, extended-spectrum β-lactamase; GNB, Gram-negative bacteria; MDR, multidrug resistant; XDR, extensively drug resistant. We did not convert reported percentages into exact numerators unless the source supplied both quantities. Study definitions and breakpoints were retained; no across-study AMR meta-analysis was performed.

Supplementary Table S2.

Sensitivity and exploratory subgroup analyses.

Analysis	k	Events/N	Pooled % (95% CI)	I2 (%)	τ2	95% PI / interaction
Li excluded: GNB	9	4,298/6,429	70.2 (64.2–75.6)	93.8	0.137	48.0–85.8
Li excluded: E. coli	11	2,194/8,628	26.9 (22.1–32.3)	95.2	0.163	12.4–49.1
Li excluded: K. pneumoniae	7	877/6,175	14.8 (12.1–18.0)	84.1	0.072	7.6–27.0
Babekir excluded: culture yield	9	3,666/5,097	78.3 (68.5–85.7)	98.0	0.561	35.6–95.9
Babekir excluded: GNB	9	22,400/31,969	69.3 (65.1–73.2)	95.3	0.068	53.9–81.4
Babekir excluded: E. coli	11	10,357/34,168	26.3 (22.4–30.5)	96.7	0.108	14.0–43.7
Moderate-risk studies excluded: culture yield	8	3,503/4,863	79.3 (68.6–87.1)	98.2	0.616	34.7–96.5
Moderate-risk studies excluded: GNB	8	22,232/31,760	67.9 (63.5–72.1)	95.6	0.066	52.5–80.3
Moderate-risk studies excluded: E. coli	10	10,274/33,959	25.1 (21.2–29.5)	97.0	0.108	13.2–42.4
Moderate-risk studies excluded: K. pneumoniae	7	4,096/31,539	14.2 (12.3–16.5)	88.0	0.036	9.1–21.6
Moderate-risk studies excluded: polymicrobial	11	3,013/5,318	45.4 (27.0–65.2)	99.3	1.856	3.4–95.2
Polymicrobial: exact numerators only	7	1,739/3,685	32.7 (13.7–59.8)	99.4	2.242	0.8–96.8
Polymicrobial: later Miutescu count substituted	12	3,069/5,481	44.0 (26.7–62.9)	99.2	1.813	3.3–94.7
Polymicrobial: Miutescu cohort omitted	11	2,919/5,141	44.0 (25.2–64.7)	99.3	2.018	2.7–95.8
Culture definition: clinical infection	7	2,765/3,524	82.1 (71.4–89.3)	97.2	0.617	34.3–97.6; p=0.014
Culture definition: broad sampling	3	934/1,623	63.3 (51.4–73.7)	89.1	0.156	0.5–99.8
Culture region: East Asia	6	3,113/4,270	84.4 (72.3–91.9)	98.7	0.784	Interaction p=0.009
Culture region: other	4	586/877	63.9 (54.1–72.7)	84.4	0.139	-
GNB region: East Asia	7	22,185/31,699	67.2 (62.5–71.6)	96.2	0.066	Interaction p=0.016
GNB region: other	3	247/303 81.7	(71.1–89.0)	55.8	0.142	-
E. coli region: East Asia	7	9,885/31,699	28.5 (25.0–32.3)	94.2	0.046	Interaction p=0.875
E. coli region: other	5	490/2,502	27.6 (17.5–40.5)	95.3	0.392	-
K. pneumoniae region: East Asia	6	4,075/31,478	13.2 (11.7–14.9)	82.3	0.019	Interaction p=0.317
K. pneumoniae region: other	2	49/270	22.0 (7.8–48.2)	92.4	0.690	-
Polymicrobial region: East Asia	5	1,524/3,055	34.4 (9.2–73.0)	99.6	3.482	Interaction p=0.459
Polymicrobial region: other	7	1,533/2,426	50.9 (32.8–68.7)	98.4	0.998	-

PI, prediction interval. Babekir-excluded analyses address the article's contradictory culture statements. The Miutescu substitutions address two overlapping reports from the same 488-person cohort. Region and infection-definition analyses were exploratory and unadjusted; nominal interactions for culture and GNB were not evidence of causality. Freeman-Tukey interaction results were directionally similar. Setting and sampling-route models were infeasible because reporting was incomplete and key cells were sparse. The risk-of-bias sensitivity excluded Kar and Babekir simultaneously; for endpoints to which Babekir did not contribute, only Kar was removed.

Funding Statement

None.

AUTHOR CONTRIBUTIONS

A.A.S. conceived and designed the review, developed the search strategy, performed study selection and data extraction, and conducted the statistical analysis. M.A.S., A.A., A.E., and A.M. contributed to data verification, interpretation, and critical revision. All authors approved the final version and accept accountability for the work.

ARTIFICIAL INTELLIGENCE DISCLOSURE

OpenAI Codex was used only to assist with language editing, document preparation, reference organization, and consistency checks. It was not used to determine study eligibility or draw scientific conclusions. The authors critically reviewed the evidence, data, interpretations, references, and final text and take full responsibility for the manuscript.

DATA AVAILABILITY

Extracted aggregate data are provided in the tables and Supplementary Appendix; additional audit details are available from the corresponding author on reasonable request.