Evaluating the role of capsule endoscopy for non-responsive, refractory and complicated celiac disease: a systematic review and meta-analysis

Ian Io Leia,b,c*, Rinda Nareshc*, Anastasios Koulaouzidisd,e, Ramesh P. Arasaradnama,b,f

University Hospital of Coventry and Warwickshire, Coventry UK; University of Warwick, Coventry, UK; Royal Berkshire Hospital, Reading, UK; University of Southern Denmark, Odense, Denmark; Odense University Hospital, Denmark; University of Leicester, UK

aInstitute of Precision Diagnostics and Translational Medicine, University Hospital of Coventry and Warwickshire, Coventry, UK (Ian Io Lei, Ramesh P. Arasaradnam); bWarwick Medical School, University of Warwick, Coventry, UK (Ian Io Lei, Ramesh P. Arasaradnam); cDepartment of Gastroenterology, Royal Berkshire Hospital, Reading, UK (Ian Io Lei, Rinda Naresh); dDepartment of Clinical Research, University of Southern Denmark, Odense, Denmark (Anastasios Koulaouzidis); eDepartment of Surgery, Odense University Hospital, Denmark (Anastasios Koulaouzidis); fLeicester Cancer Center, University of Leicester, UK (Ramesh P. Arasaradnam)

Correspondence to: Ian Io Lei, Clifford Bridge Rd, Binley, Coventry CV2 2DX, UK, e-mail: brian.lei@uhcw.nhs.uk

* These authors contributed equally to this work as joint first authors

Received 11 January 2026; accepted 23 May 2026; published online 9 July 2026
DOI: https://doi.org/10.20524/aog.2026.1092
© 2026 Hellenic Society of Gastroenterology

Abstract

Background Small-bowel capsule endoscopy (CE) is used in selected adults with nonresponsive, refractory, or complicated celiac disease (CD), but its diagnostic yield (DY), clinical impact, and safety remain uncertain.

Methods We searched Medline, EMBASE, CENTRAL, and PubMed from January 2000 to November 30, 2025, for studies of CE in adults with established CD and persistent symptoms, suspected refractory disease, or suspected complications. Two reviewers independently selected studies, extracted data, and assessed risk of bias using ROBINS-E; certainty was assessed using GRADE. Random-effects meta-analyses using generalized linear mixed models estimated pooled DY and secondary outcomes. Meta-regression and subgroup analyses were considered exploratory.

Results Seventeen studies, including 994 adults, were eligible. The pooled DY of CE in complex CD was 82% (95% confidence interval [CI] 67-91%), with substantial heterogeneity (I2=82.5%). DY was highest in refractory CD cohorts. Pooled DY was 3% for ulcerative jejunitis and 3% for small-bowel malignancy. Capsule retention was uncommon. CE findings were associated with management changes in several studies; however, this outcome was reported in only 9 studies, and definitions varied. Egger’s test and trim-and-fill analysis suggested possible small-study effects, reducing the adjusted DY estimate to 79%. The pooled conversion to device-assisted enteroscopy was estimated at 18% (95%CI 9-30%).

Conclusions CE may be a valuable second-line, noninvasive triage tool in selected high-risk patients with complex CD, particularly refractory CD. However, the available evidence remains of very low certainty and should be interpreted cautiously, given the substantial methodological and clinical heterogeneity across studies.

Keywords Capsule endoscopy, small bowel, diagnostic yield, refractory celiac disease, complicated celiac disease

Ann Gastroenterol 2026; 39 (5): 552-562

Introduction

Celiac disease (CD) is a chronic, immune-mediated enteropathy triggered by gluten exposure in genetically susceptible individuals [1,2]. Although adherence to a strict gluten-free diet (GFD) remains the cornerstone of treatment, up to 30% of adults continue to experience ongoing symptoms despite reported adherence [3,4]. These cases may represent non-responsive CD (NRCD) or refractory CD (RCD), which are associated with nutritional deficiencies, impaired quality of life, and an elevated risk of lymphoproliferative malignancy [5-7].

NRCD is characterized by persistent symptoms, abnormal serology or ongoing mucosal abnormalities after 6-12 months of a strict GFD. In most cases, NRCD reflects inadvertent gluten exposure or alternative gastrointestinal (GI) pathology (e.g., irritable bowel syndrome), rather than true treatment failure [3,4]. In contrast, RCD is defined by persistent or recurrent malabsorptive symptoms with ongoing villous atrophy, despite strict GFD for ≥12 months, after exclusion of alternative causes of enteropathy [8,9]. RCD is further subdivided into type I, characterized by a normal intraepithelial lymphocyte phenotype, and type II, defined by an aberrant clonal intraepithelial lymphocyte population, which carries a substantially poorer prognosis [8,9]. Complicated CD (CCD) encompasses structural and malignant sequelae, including ulcerative jejunitis, strictures and enteropathy-associated T-cell lymphoma (EATL); these typically arise in longstanding NRCD or RCD and carry high morbidity and mortality [8,9]. Early recognition and precise differentiation of these disease states is critical for guiding appropriate further management.

For clinicians managing patients who have persistent symptoms despite confirmed dietary adherence, determining the optimal diagnostic pathway remains challenging. Traditional evaluation relies on repeated esophagogastroduodenoscopy with duodenal biopsies; however, this is invasive, prone to sampling error, and limited to the proximal small bowel (SB) [10]. SB capsule endoscopy (CE) provides a minimally invasive assessment of the entire SB, and is more sensitive than conventional endoscopy for detecting villous atrophy and complications in established CD [10,11]. According to the European Society of Gastrointestinal Endoscopy (ESGE) guidelines, CE is recommended in suspected NRCD or RCD as a second-line test, after serology, biopsies and dietary review, to assess villous atrophy and identify complications [12]. Cross-sectional imaging and device-assisted enteroscopy (DAE) are generally used for the targeted characterization or biopsy of lesions identified on CE [10,11,13].

Despite these recommendations, the evidence supporting CE in NRCD or CCD remains fragmented and of low certainty. Existing meta-analyses have primarily examined CE at initial diagnosis, rather than its diagnostic yield (DY) in high-risk NRCD, RCD or CCD cohorts [14,15]. Evidence regarding the clinical impact of CE, specifically how findings influence subsequent investigations or management, remains limited [10]. We therefore conducted a systematic review and meta-analysis to evaluate the DY of CE in NRCD, RCD and CCD, identify factors associated with diagnostic performance, and assess procedure-related complications.

Materials and methods

This study protocol was designed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines (see supplemental table 1) [16]. The protocol was registered with the PROSPERO International Register of Systematic Reviews (Registration ID: CRD420251243243). No amendments were made to the protocol following registration. This study aimed to systematically evaluate the DY, clinical impact and safety of CE in adult patients with NRCD, RCD, or CCD.

Eligibility criteria

We included full-text observational studies (retrospective or prospective) that evaluated the use of CE in adults with established CD presenting with NRCD, RCD or CCD. All types of CE were included, and all referral indications were considered, including symptomatic patients, asymptomatic patients undergoing evaluation, positive serology or previously confirmed RCD. Conference abstracts were excluded because of insufficient methodological detail and/or limited extractable data. Review articles, systematic reviews, editorials, study protocols, case reports and studies with <10 participants were excluded.

Information sources

We conducted a comprehensive literature search across the following electronic databases: Medline (Ovid), EMBASE (Ovid), Cochrane Central Register of Controlled Trials (CENTRAL), and PubMed. The search strategy incorporated controlled vocabulary terms (e.g., MeSH terms) and free-text keywords relevant to CD and CE. The complete search strategy for each database is available in Supplementary Table 2, in accordance with the PRISMA-S (PRISMA-Search) guidelines. The search was limited to studies published between 1 January 2000 and 30 November 2025, and only English language publications were included. To ensure comprehensiveness, we also conducted a manual review of reference lists of full-text articles and relevant systematic reviews.

Study selection

All records were imported into Rayyan (Rayyan Systems Inc.), a web-based platform designed for systematic review screening. Duplicate entries were identified and removed. Title and abstract screening, followed by full-text eligibility assessment, were independently performed by 2 reviewers (IIL and RN). Disagreements were resolved through discussion and consensus.

Inclusion criteria

  • Participants had a confirmed diagnosis of CD established by serology and/or duodenal histology

  • The study included adults with established CD undergoing CE for any clinical indication

  • The sample size was ≥10 participants

  • Any CE system was permitted, including gastric, SB, colon, or panenteric capsules, irrespective of manufacturer

  • The study did not need CD as a primary or secondary endpoint, provided that relevant CE outcomes were extractable

Exclusion criteria

  • Pediatric studies (<18 years of age) or non-human studies

  • Studies with suspected but unconfirmed CD

  • Studies focusing exclusively on obscure GI bleeding or non-CD indications

  • Studies using CE primarily for the initial diagnosis of CD, rather than for evaluation, surveillance or assessment of complications

Data compilation

The final set of studies was reviewed, and data extraction was independently performed by IL and RN. Extracted variables included patient demographics, disease duration and indications for CE. Clinical data included the number of patients with NRCD, RCD or CCD, subsequent DAE, elevated biomarkers, and symptomatic presentations. NRCD refers to persistent symptoms, abnormal serology or mucosal abnormalities, despite 6-12 months of a GFD [3,4]. RCD was defined by ongoing malabsorptive symptoms and villous atrophy, despite ≥12 months of GFD, and both Type 1 and 2 were included [8,9]. CCD encompassed structural or malignant complications arising from longstanding NRCD or RCD, including ulcerative jejunitis, strictures and EATL. For the purposes of this meta-analysis, complex CD (CxCD) was used as an operational umbrella term encompassing NRCD, RCD, and CCD; however, these entities are clinically distinct and should not be considered equivalent in terms of disease severity, risk profile or DY. Villous atrophy on CE was identified by macroscopic surrogate markers, including scalloping of mucosal folds, mosaic pattern, fissuring, nodularity, and absence of villi. DY was defined as the proportion of CE examinations identifying 1 or more clinically significant small-bowel abnormalities in patients with CxCD. The histological reference standard varied across studies, and the Marsh–Oberhuber classification based on duodenal biopsies was most commonly used, while flow cytometry for aberrant intraepithelial lymphocytes was used to subtype RCD in 3 studies [17-19]. Technical details included CE system type, completion rate, bowel preparation quality, use of patency capsule and adverse events. Diagnostic outcomes included overall and lesion-specific DY (ulcers, ulcerative jejunitis, Crohn’s disease, malignancy), and whether CE changed diagnosis or altered patient management.

Risk of bias

The risk of bias (RoB) for all non-randomized studies was evaluated using the ROBINS-E tool for non-randomized studies of exposures. We acknowledge that QUADAS-2 is the standard tool for diagnostic accuracy studies; however, given that our review focused on DY and downstream clinical impact, rather than sensitivity and specificity, and included observational studies without a uniform reference standard, we selected ROBINS-E as a more appropriate framework to assess bias relevant to an exposure–outcome paradigm. ROBINS-E assesses potential bias arising from confounding, participant selection, exposure classification, deviations from intended exposure, missing data, outcome measurement and selective reporting [20]. Two reviewers independently assessed each domain and assigned ratings of low, moderate, serious or critical risk, with disagreements resolved by consensus or third-reviewer adjudication; overall study ratings reflected the highest-risk domain. The certainty of evidence for each outcome was assessed using the GRADE framework, considering RoB, inconsistency, indirectness, imprecision and publication bias [21].

Statistical analysis

We performed random-effects meta-analyses using a generalized linear mixed model to estimate pooled DYs for active RCD, as well as secondary outcomes, including ulcerative jejunitis, impact on management, conversion to DAE, lymphoma, and non-lymphomatous small-bowel tumors. Heterogeneity was assessed using the I2 statistic and the Cochran’s Q test. To explore potential sources of heterogeneity, we performed univariate meta-regression using the “rma” function in the “metafor” package [22]. Effect sizes were calculated as logit-transformed proportions using the “escalc” function. Prespecified moderators included mean age, ulcer and tumor DY, CE completion rate, proportion with positive celiac serology (tTG), proportion with established RCD, and symptom burden. Subgroup and sensitivity analyses were performed based on study-level characteristics and RoB. All analyses were conducted using R (version 4.3.2) [23] with the “meta” [24], “metafor” and “dplyr” [25] packages.

Results

Literature search and study selection

A systematic search conducted on 30 November 2025 identified 574 records across 4 databases. Following duplicate removal and title/abstract screening, 27 articles underwent full-text review, from which 17 studies met the inclusion criteria for the systematic review and meta-analysis. The reasons for the exclusion of the remaining 11 studies are detailed in Fig. 1, with the most common being inclusion of suspected rather than confirmed CD, and conference-only abstracts. One additional study focusing on chronic diarrhea in patients with established CD was identified through reference checking. Of the included studies, 6 were retrospective and the remainder prospective, contributing a total of 994 patients to the meta-analysis (Supplementary Table 3). The mean age across cohorts ranged from 45.9-55.8 years (Table 1).

thumblarge

Figure 1 PRISMA flowchart

Table 1 Characteristics of included studies

thumblarge

Across the 17 included studies, a wide variety of positive findings encompassed: (i) macroscopic markers of villous atrophy, including scalloping of mucosal folds, mosaic pattern, fissuring, nodularity, and flattening or absence of villi; (ii) mucosal inflammation, including erosions, erythema and edema; (iii) ulcerative lesions, including aphthous ulcers and ulcerative jejunitis or jejunoileitis; (iv) stricturing disease; and (v) suspected small-bowel malignancy, including mass lesions and circumferential thickening. These features are consistent with the established CE criteria for CD described in the ESGE CE guidelines [26] and the European Society for the Study of Celiac Disease’s 2025 updated diagnostic guidelines [12]. In addition, the definitions and granularity of “management change” varied considerably across studies, ranging from broad categories to specific therapeutic interventions (Supplementary Table 4).

The RoB assessment was performed using the ROBINS-E tool, summarized in Supplementary Table 5. The majority of studies were observational and predominantly single center, with small sample sizes and substantial heterogeneity in study design, patient selection and outcome definitions. RoB was rated as serious in 5 studies, particularly for selection bias, unclear indications, and non-consecutive or highly preselected cohorts. Reader blinding was inconsistently applied, and reference standards varied substantially across studies, with many lacking systematic histological confirmations or relying only on biopsies of high-risk lesions.

As summarized in Supplementary Table 6, the certainty of the evidence, as assessed using GRADE, across the included studies was predominantly very low. This was driven by small observational designs, indirectness and substantial imprecision. Only 2 studies achieved a low-certainty rating, with secondary outcomes similarly limited by inconsistent reference standards, unblinded outcome assessment, and sparse reporting of clinical impact or adverse events.

Overall results

The pooled DY of CE in CxCD was 82% (95%CI 67-91%), with substantial heterogeneity (I2=82.5%). This estimate should be interpreted in the context of heterogeneous definitions of positive CE findings across studies. Subgroup analysis identified disease phenotype as the primary determinant of yield, with the highest DY observed in RCD (99%, 95%CI 82-100%; I2=0) (Fig. 2). The pooled DY for CCD further highlighted this pattern, with a markedly higher yield in the RCD subgroup (18%, 95%CI 4-56%; I2=86.3%) compared with NRCD (4%, 95%CI 2-10%; I2=0%), indicating that complications are concentrated in RCD populations (Fig. 3). Accordingly, pooled estimates should not be interpreted uniformly across disease phenotypes. CE findings were associated with changes in clinical management in a substantial proportion of patients (Supplementary Fig. 1), although this outcome was highly heterogeneous and is presented descriptively. The pooled conversion rate to DAE was 18% (95%CI 9-30%; I2=88.8%), supporting a role for CE as a triage investigation, with only a minority requiring further invasive assessment. For specific findings, the pooled DY was 9% (95%CI 5-16%; I2=70.2%) for ulcers, 3% (95%CI 1-7%; I2=64.3%) for confirmed ulcerative jejunitis, and 3% (95%CI 1-6%) for malignancy (Supplementary Fig. 2,3). Capsule retention was uncommon (0.05%, 95%CI 0.02-0.14%). Completion rates ranged from 62-100%, and bowel preparation adequacy from 53-100%. PillCam (Medtronic, USA) was the most frequently used system (8 studies), followed by M2A (Given Imaging, Israel) (5 studies), panenteric PillCam Crohn’s capsule (2 studies), and MiroCam (IntroMedic, South Korea) (1 study).

thumblarge

Figure 2 Pooled diagnostic yield of capsule endoscopy in CxCD, stratified by disease phenotype. Forest plot demonstrating the overall pooled diagnostic yield of 82% (95%CI 67-91%) with substantial heterogeneity (I2=82.5%). Subgroup analysis shows marked variation by phenotype, with the highest yield observed in RCD (99%, 95%CI 82-100%) compared with NRCD (79%, 95%CI 44-95%) and mixed (69%, 95%CI 56-80%) cohorts. The prediction interval (15-99%) highlights considerable between-study variability, indicating that diagnostic yield should be interpreted in the context of the underlying disease phenotype CxCD, complex celiac disease; NRCD, non-responsive celiac disease; RCD, refractory celiac disease; CI, confidence interval; I2, Higgins heterogeneity statistic; τ2, between-study variance; df, degrees of freedom; χ2, chi-squared statistic

thumblarge

Figure 3 Pooled diagnostic yield of CCD, stratified by disease phenotype within the CxCD cohort. Forest plot showing an overall pooled CCD diagnostic yield of 6% (95%CI 3-11%) with substantial heterogeneity (I2=82.7%). Subgroup analysis demonstrates higher yield in RCD (18%, 95%CI 4-56%) compared with NRCD (4%, 95%CI 2-10%) and mixed (6%, 95%CI 3-11%) cohorts, highlighting the concentration of complications in RCD populations. The wide prediction interval (0-48%) reflects considerable between-study variability CCD, complicated celiac disease; CxCD, complex celiac disease; NRCD, non-responsive celiac disease; RCD, refractory celiac disease; CI, confidence interval; I2, Higgins heterogeneity statistic; τ2, between-study variance; df, degrees of freedom; χ2, chi-squared statistic

Publication bias was evaluated using funnel plot inspection, Egger’s regression test, and the trimandfill method. The funnel plot visualization suggested mild asymmetry, which was confirmed statistically by Egger’s regression test (t=2.19, df=15, P=0.0446), suggesting potential small-study effects or publication bias. Application of the trim-and-fill method imputed 4 potentially missing studies on the left side of the funnel plot, shifting the pooled estimate downward to 79% (standard error 34%; P=0.0186) (Supplementary Fig. 4). Although the adjusted effect remained statistically significant, the attenuation suggests that the observed DY may be modestly inflated in the published literature.

Meta-regression

Meta-regression analyses were undertaken to explore covariates that might influence the DY of CE in CxCD. The exploratory univariate model was examined, with the results summarized in Supplementary Table 7. Two factors demonstrated significant associations with higher DY for positive CE findings: the proportion of patients with known RCD and the proportion of patients whose clinical management was altered following CE. Studies enrolling a greater proportion of patients with established RCD consistently reported higher DYs.

A dedicated meta-regression evaluating the DY of malignancy was also performed. Known RCD again emerged as a significant predictor of higher malignancy detection (P=0.006), as shown in the bubble plot (Supplementary Fig. 5). Similarly, a higher reported impact of CE on clinical decision-making was strongly associated with more malignant DY (P=0.004). Sensitivity analyses using leave-one-out procedures showed stable effect estimates across all studies. No individual study exerted disproportionate influence on the pooled results (Supplementary Fig. 6).

Subgroup analysis

A subgroup analysis was undertaken using covariates identified as significant in the meta-regression. The relationship between DY and the impact of CE on subsequent management was first examined. Studies were stratified by the proportion of patients whose management changed following CE (high >55%, moderate 30-55%, low <30%). Only 9 eligible studies reported this outcome. The high-impact group demonstrated a nearly maximal pooled DY of 99% (95%CI 20-100%; I2=0%), with progressively lower yields in the moderate- and low-impact groups; the between-group difference was significant (P<0.001) (Fig. 4). This gradient suggests a direct association between DY and downstream changes in management.

thumblarge

Figure 4 Subgroup analysis of diagnostic yield stratified by reported impact of capsule endoscopy on clinical management. Studies were grouped by the proportion of patients experiencing a change in management following capsule endoscopy (high >55%, moderate 30-55%, low <30%). Diagnostic yield increased across groups, with the highest yield observed in studies reporting high clinical impact (99%, 95%CI 20-100%), compared with moderate (86%, 95%CI 81-90%) and low (51%, 95%CI 44-59%) impact groups. A statistically significant between-group difference was observed (P<0.001). However, only 9 studies reported management change and definitions varied across studies; therefore, these findings should be considered exploratory CI, confidence interval; I2, Higgins heterogeneity statistic; τ2, between-study variance; df, degrees of freedom; χ2, chi-squared statistic

A further subgroup analysis assessed whether the proportion of patients with known RCD influenced tumor-specific DY. Studies without established RCD reported a pooled malignant DY of 2% (95%CI 1-4%; I2=0%), whereas studies including patients with known RCD showed a higher yield of 4% (95%CI 2-10%; I2=69.5%) (Supplementary Fig. 3). This pattern reflects the recognized increased risk of EATL and ulcerative jejunoileitis in established RCD, particularly type 2. Subgroup analyses by age, disease duration, study design and CE type revealed no significant differences between strata.

Discussion

CD, particularly when associated with persistent villous atrophy, carries an increased risk of complications and mortality. Current ESGE guidelines emphasize clinical review, serology, and repeat duodenal biopsies for surveillance, reserving CE and DAE for the uncommon scenario of suspected NRCD or RCD. These targeted modalities are therefore used selectively in a high-risk subgroup. However, no previous systematic review and meta-analysis has specifically evaluated the DY of CE in this context, its influence on subsequent management, or the covariates that may affect DY. Furthermore, the certainty of evidence supporting the use of CE in patients with persistent symptoms despite strict dietary adherence, seronegative CD, or suspected RCD or CCD remains limited [12].

This work presents the first systematic review and meta-analysis to characterize the DY of CE in NRCD, RCD and CCD. Across studies, the pooled DY was 82% (95%CI 67-91%), a figure comparable to the DY of 80% (95%CI 74-87%) reported for CE within 48 h of overt GI bleeding [27], and broadly consistent with the DY range of 55-68% described in Crohn’s disease [28,29]. These findings demonstrate that CE has a high DY for detecting clinically important abnormalities in this population, including villous atrophy, erosions, ulcerative jejunoileitis and EATL. Diagnostic performance in this cohort mirrored that of DAE, which achieved an 83% yield (19/23 cases) in the sole published series [30]. The RCD subgroup exhibited a DY of 99% (I2=0). In contrast, the NRCD subgroup had the lowest CCD DY, at 4% (95%CI 2-10%; I2=0%), compared with 18% in the RCD group. These findings demonstrate the critical role of CE in the high-risk RCD cohort. Despite a high DY of 82% with CE, DAE was performed in only 18% of cases (95%CI 9-30%). These findings are consistent with a role for CE as a noninvasive triage investigation before DAE in patients with suspected CCD, although prospective comparative studies are required to define this sequential approach. CE may therefore be considered a first-line, noninvasive investigation in high-risk patients, with DAE reserved for targeted biopsy, evaluation of ulcerative jejunoileitis, or assessment of stricturing disease, in line with ESGE guidance [26].

Given the high DY, a central question is whether CE meaningfully influences clinical management in this population. In our study, CE findings were associated with changes in clinical management in a substantial proportion of patients across studies; however, this outcome was highly heterogeneous and should be interpreted descriptively rather than as a precise pooled estimate (Supplementary Fig. 1). The proportion of patients experiencing a management change was lower than that reported for CE in suspected general small-bowel disease, in which management changes occur in approximately 70% of patients [31]. An exploratory subgroup analysis suggested higher DY in studies reporting greater management change. However, only 9 studies contributed data, while outcome definitions were heterogeneous, limiting the robustness of this observation. Within the subset of patients with confirmed RCD only, studies showed higher malignant DYs (6%; 95%CI 1-37%; I2=0) compared with studies without RCD (2%; 95%CI 0-6%; I2=0) (Supplementary Fig. 3). These findings reinforce the use of CE as a surveillance tool in this high-risk cohort [12]. However, few studies distinguished between RCD type 1 and type 2, precluding subtype-specific analyses. Although the need for surveillance in RCD is clear, the optimal interval and duration remain uncertain and represent important areas for future research. Importantly, these findings should be interpreted in the context of disease phenotype. The DY and clinical impact of CE were highest in patients with established RCD, whereas evidence supporting its use in NRCD alone remains more limited and of lower certainty.

Across a limited number of studies, patients with negative CE findings did not subsequently develop new CD-related complications during follow-up, suggesting a high negative predictive value for clinically significant small-bowel pathology. In a large multicenter European cohort, no patient with a negative examination was later found to have EATL, ulcerative jejunoileitis, or other major complications, with alternative non-celiac diagnoses identified in a substantial proportion. Therefore, a negative small-bowel CE appears clinically reassuring, with available evidence suggesting a low risk of missed celiac-related complications and frequent identification of alternative diagnoses. It may therefore support a strategy of conservative management and avoidance of further invasive investigation, although this inference is based on limited and heterogeneous data. Assessment of publication bias suggested the presence of small-study effects. Funnel plot asymmetry, supported by Egger’s regression test, and subsequent trim-and-fill analysis indicated that the pooled DY may be modestly overestimated, with adjustment reducing the estimate from 82% to 79%. These findings are consistent with the recognized tendency for single-arm diagnostic studies with lower yields to be under-reported, although the limited number of included studies constrains the reliability of formal bias assessment.

Limitations of this study include the predominance of retrospective studies, which resulted in a substantial proportion being rated as having a serious RoB, and contributed to an overall very low certainty of evidence. Missing data were frequent, with many studies omitting both negative CE examinations and information on how findings influenced subsequent management. Rare outcomes, such as malignancy, further contributed to imprecision, because multicenter studies varied in their definitions of GFD adherence and indications for CE, thereby exacerbating the already substantial heterogeneity. A potential overestimation of CCD DY cannot be excluded, as some studies lacked clarity on whether structural and malignant findings, such as strictures and malignancy, were reported as distinct entities, introducing a risk of double-counting individual patients. These methodological constraints reduce confidence in the pooled estimates and underscore the need for larger, prospective, standardized studies to better define the DY and clinical impact of CE in NRCD and RCD. Advances in artificial intelligence (AI), particularly in lesion characterization, may eventually enable the prediction of the RCD subtype directly from CE imaging, without the need for DAE and biopsy [32]. A notable methodological limitation across the included studies is the absence of a uniformly applied, validated scoring system for CE in CD, and the variation in histological reference standards. The threshold defining a “positive” examination differed substantially between study cohorts (Supplementary Table 8). This substantial heterogeneity in the definition of a positive CE examination and clinical impact constitutes a key study-level limitation, undermining direct cross-study comparability and limiting the interpretation of pooled DY estimates. Consequently, the pooled estimates should be regarded as summary measures across heterogeneous definitions of clinically significant findings, rather than precise estimates applicable to a single standardized diagnostic threshold. A further limitation is the absence of a uniformly applied reference standard across studies, with CE findings not consistently validated against histology or DAE in all participants. This limits the ability to interpret DY in relation to underlying disease status, and may have contributed to heterogeneity in reported yields. In addition, most studies were conducted in tertiary referral settings, and included highly selected populations enriched for refractory disease, raising the possibility of spectrum bias and overestimation of DY. Standardization of CE reporting criteria and reporting outcome in CD remains a priority for future prospective studies.

In conclusion, this study adds to the growing evidence on the DY of CE across NRCD, RCD and CCD cohorts, and suggests that it can meaningfully inform subsequent clinical management. Its noninvasive nature and low conversion to subsequent DAE suggest it may have a role as a second-line investigation and triage tool in patients with CxCD. However, these findings should be interpreted cautiously, given the substantial heterogeneity and predominantly very-low-certainty evidence. Future studies should evaluate the role of AI in image interpretation, although robust prospective evidence is needed before AI-assisted approaches can be integrated into routine diagnostic pathways for NRCD, RCD and CCD.

Summary Box

What is already known:

  • Capsule endoscopy (CE) is recommended as a second-line investigation in suspected refractory or complicated celiac disease (CD), but no prior systematic review has evaluated its diagnostic yield specifically in this population

  • Up to 30% of adults with CD have persistent symptoms, despite a gluten-free diet, and these cases carry an elevated risk of complications including ulcerative jejunitis and lymphoma

What the new findings are:

  • The pooled diagnostic yield of CE in complex CD was 82% (95% confidence interval 67-91%), rising to 99% in refractory CD, with low pooled rates of ulcerative jejunitis (3%) and small-bowel malignancy (3%)

  • The proportion of patients with known refractory CD and the proportion experiencing management change were significantly associated with a higher diagnostic yield, and only 18% required conversion to device-assisted enteroscopy

References

1. Al-Toma A, Volta U, Auricchio R, et al. European Society for the Study of Coeliac Disease (ESsCD) guideline for coeliac disease and other gluten-related disorders. United European Gastroenterol J 2019;7:583-613.

2. Lebwohl B, Sanders DS, Green PHR. Coeliac disease. Lancet 2018;391:70-81.

3. Penny HA, Baggus EMR, Rej A, Snowden JA, Sanders DS. Non-responsive celiac disease: a comprehensive review from the NHS England National Centre for Refractory Celiac Disease. Nutrients 2020;12:216.

4. Chetcuti Zammit S, Kurien M, Sanders DS, Sidhu R. What is the role of small bowel capsule endoscopy in established celiac disease? Clin Res Hepatol Gastroenterol 2020;44:753-761.

5. Blom JJ, Gidrewicz D, Turner J, Duerksen DR, Pinto-Sánchez MI. Diagnosis and management of celiac disease. CMAJ 2025;197:E1258-E1265.

6. Malamut G, Soderquist CR, Bhagat G, Cerf-Bensussan N. Advances in nonresponsive and refractory celiac disease. Gastroenterology 2024;167:132-147.

7. Elfström P, Granath F, Ekström Smedby K, et al. Risk of lymphoproliferative malignancy in relation to small intestinal histopathology among patients with celiac disease. J Natl Cancer Inst 2011;103:436-444.

8. Malamut G, Cellier C. Refractory coeliac disease. Curr Opin Oncol 2013;25:445-451.

9. Nijeboer P, van Wanrooij RL, Tack GJ, Mulder CJ, Bouma G. Update on the diagnosis and management of refractory coeliac disease. Gastroenterol Res Pract 2013;2013:518483.

10. Lewis SK, Semrad CE. Capsule endoscopy and enteroscopy in celiac disease. Gastroenterol Clin North Am 2019;48:73-84.

11. Branchi F, Ferretti F, Orlando S, et al. Small-bowel capsule endoscopy in patients with celiac disease, axial versus lateral/panoramic view: Results from a prospective randomized trial. Dig Endosc 2020;32:778-784.

12. Al-Toma A, Zingone F, Branchi F, et al. European Society for the Study of Celiac Disease 2025 updated guidelines on the diagnosis and management of celiac disease in adults. Part 1: Diagnostic approach. United European Gastroenterol J 2025;13:1855-1886.

13. Elli L, Casazza G, Locatelli M, et al. Use of enteroscopy for the detection of malignant and premalignant lesions of the small bowel in complicated celiac disease: a meta-analysis. Gastrointest Endosc 2017;86:264-273.

14. Rokkas T, Niv Y. The role of video capsule endoscopy in the diagnosis of celiac disease: a meta-analysis. Eur J Gastroenterol Hepatol 2012;24:303-308.

15. Shapiro M, Niv Y. Diagnostic yield of video capsule endoscopy (VCE) in celiac disease (CD): a systematic review and meta-analysis. J Clin Gastroenterol 2025;59:598-606.

16. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. J Clin Epidemiol 2021;134:178-189.

17. Shiha MG, Oka P, Nandi N, et al. Role of capsule endoscopy and double-balloon enteroscopy in the management of adult patients with coeliac disease and persisting symptoms. Dig Liver Dis 2025;57:206-212.

18. Elli L, Soru P, Roncoroni L, et al. Clinical features of type 1 and 2 refractory celiac disease: Results from a large cohort over a decade. Dig Liver Dis 2023;55:235-242.

19. Chetcuti Zammit S, Schiepatti A, Aziz I, Kurien M, Sanders DS, Sidhu R. Use of small-bowel capsule endoscopy in cases of equivocal celiac disease. Gastrointest Endosc 2020;91:1312-1321.

20. Higgins JPT, Morgan RL, Rooney AA, et al. A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E). Environ Int 2024;186:10↚.

21. Brozek JL, Canelo-Aybar C, Akl EA, et al; GRADE Working Group. GRADE Guidelines 30: the GRADE approach to assessing the certainty of modeled evidence-An overview in the context of health decision-making. J Clin Epidemiol 2021;129:138-150.

22. Viechtbauer W. Conducting meta-analyses in R with the metafor package. In Journal of Statistical Software 2010;36:1-48.

23. R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2024. Available from: https://www.R-project.org/[Accessed 25 June 2026].

24. Balduzzi S, Rücker G, Schwarzer G. How to perform a meta-analysis with R: a practical tutorial. Evid Based Ment Health 2019;22:153-160.

25. Wickham H, François R, Henry L, Müller K. dplyr: A Grammar of Data Manipulation. R package version 1.1.2; 2023. Available from: https://CRAN.R-project.org/package=dplyr [Accessed 25 June 2026].

26. Pennazio M, Rondonotti E, Despott EJ, et al. Small-bowel capsule endoscopy and device-assisted enteroscopy for diagnosis and treatment of small-bowel disorders: European Society of Gastrointestinal Endoscopy (ESGE) Guideline - Update 2022. Endoscopy 2023;55:58-95.

27. Estevinho MM, Pinho R, Fernandes C, et al. Diagnostic and therapeutic yields of early capsule endoscopy and device-assisted enteroscopy in the setting of overt GI bleeding: a systematic review with meta-analysis. Gastrointest Endosc 2022;95:610-625.

28. Dionisio PM, Gurudu SR, Leighton JA, et al. Capsule endoscopy has a significantly higher diagnostic yield in patients with suspected and established small-bowel Crohn's disease: a meta-analysis. Am J Gastroenterol 2010;105:1240-1248.

29. Liao Z, Gao R, Xu C, Li ZS. Indications and detection, completion, and retention rates of small-bowel capsule endoscopy: a systematic review. Gastrointest Endosc 2010;71:280-286.

30. Ferretti F, Branchi F, Orlando S, et al. Effectiveness of capsule endoscopy and double-balloon enteroscopy in suspected complicated celiac disease. Clin Gastroenterol Hepatol 2022;20:941-949.

31. Alali AA, Alrashidi R, Allahow F, Dangi A, Alfadhli A. Predictive factors of significant findings on capsule endoscopy in patients with suspected small bowel bleeding. Diagnostics (Basel) 2024;14:2352.

32. Messmann H, Bisschops R, Antonelli G, et al. Expected value of artificial intelligence in gastrointestinal endoscopy: European Society of Gastrointestinal Endoscopy (ESGE) Position Statement. Endoscopy 2022;54:1211-1231.

33. Daum S, Wahnschaffe U, Glasenapp R, et al. Capsule endoscopy in refractory celiac disease. Endoscopy 2007;39:455-458.

34. Maiden L, Elliott T, McLaughlin SD, Ciclitira P. A blinded pilot comparison of capsule endoscopy and small bowel histology in unresponsive celiac disease. Dig Dis Sci 2009;54:1280-1283.

35. Barret M, Malamut G, Rahmi G, et al. Diagnostic yield of capsule endoscopy in refractory celiac disease. Am J Gastroenterol 2012;107:1546-1553.

36. Tomba C, Elli L, Bardella MT, et al. Enteroscopy for the early detection of small bowel tumours in at-risk celiac patients. Dig Liver Dis 2014;46:400-404.

37. Perez-Cuadrado-Robles E, Lujan-Sanchis M, Elli L, et al; Enteroscopy and Capsule Endoscopy Spanish Society Group of the Spanish Society of Digestive Endoscopy (SEED). Role of capsule endoscopy in alarm features and non-responsive celiac disease: a European multicenter study. Dig Endosc 2018;30:461-466.

38. Chetcuti Zammit S, Sanders DS, Cross SS, Sidhu R. Capsule endoscopy in the management of refractory coeliac disease. J Gastrointestin Liver Dis 2019;28:15-22.

39. Katsinelos P, Fasoulas K, Beltsis A, et al. Diagnostic yield and clinical impact of wireless capsule endoscopy in patients with chronic abdominal pain with or without diarrhea: a Greek multicenter study. Eur J Intern Med 2011;22:e63-e66.

40. Kurien M, Evans KE, Aziz I, et al. Capsule endoscopy in adult celiac disease: a potential role in equivocal cases of celiac disease? Gastrointest Endosc 2013;77:227-232.

41. Atlas DS, Rubio-Tapia A, Van Dyke CT, Lahr BD, Murray JA. Capsule endoscopy in nonresponsive celiac disease. Gastrointest Endosc 2011;74:1315-1322.

42. Foerster F, Neumann H. Panintestinal capsule endoscopy in patients with celiac disease. Eur J Gastroenterol Hepatol 2021;33:e1022-e1026.

Notes

Conflict of Interest: IIL, RN and RPA report no competing interests. AK received research support from Given Imaging Ltd. (via the ESGE) and IntroMedic; has served as a consultant for Jinshan Ltd.; and has received consultancy fees from Jinshan, lecture honoraria from Covidien/Medtronic, and educational travel support or honoraria from Jinshan, Dr Falk Pharma UK, Ferring, Aquilant, and Almirall. He has also served on advisory boards for Tillotts, Ankon, Jinshan, and Dr Falk Pharma UK, and is co-founder and director of iCERV Ltd