Advanced imaging (narrow band and acetic acid chromoendoscopy) and guided biopsies in surveillance of Barrett's oesophagus: a systematic review.

Introduction
Barrett’s oesophagus (BO) is the precancerous precursor of oesophageal adenocarcinoma (OAC), with increasing incidence, especially in Europe.1 2 Reflux oesophagitis healing with intestinal metaplasia (IM) can progress to dysplasia (low-grade (LGD) or high-grade (HGD)) risking the development of OAC. Although the overall incidence of this neoplastic transformation is low, the invasive cancer has an extremely poor 5-year survival.14
Early identification of dysplastic BO/early cancer (EC) is the key to cure and improve survival. Hence, regular BO surveillance was recommended at 6 months–5 years interval.1 The surveillance endoscopy needs to be of good quality to ensure higher yield (ie, dysplasia detection rate/neoplasia detection rate (DDR/NDR)), and the guidelines recommended a minimum dataset to be recorded during endoscopy, expressed as key performance indicators.1 Special imaging modalities such as narrow band imaging (NBI) and acetic acid chromoendoscopy (AAC) are used to improve dysplasia detection during surveillance endoscopy. Despite these advances, the current research landscape on special imaging techniques used for surveillance of patients with BO is incomplete.
High-definition (resolution) white light endoscopy (HDWLE) is currently used in most places, replacing standard video endoscopy (SVE) which was graded for the level of quality, according to each outcome of interest using GRADE criteria. Their RoB was assessed for the availability of newer equipment. A biopsy of the oesophagus confirming specialised columnar epithelium (SCE) and the presence of IM is essential to confirm the diagnosis of BO.1 5 Four quadrant random biopsies (4QB) are recommended at every 2 cm distance (Seattle protocol) but are time consuming, labour intensive and prone to sampling error, especially when the BO length is long. It has been proposed that NBI-guided and/or AAC-guided targeted biopsy (TB) may improve the yield and reduce the number of biopsies, time and cost in BO surveillance.
A systematic review and meta-analysis by the American Society of Gastroenterology (ASGE) demonstrated the overall sensitivity, specificity, positive predictive value and negative predictive value (NPV) of NBI and AAC and other advanced imaging methods in BO surveillance.6 This review indicated that AAC and NBI meet the ASGE’s PIVI (prevention and incorporation of valuable endoscopic innovations) threshold for detecting HGD/EC.6 By comparison, this systematic review was designed to systematically identify and examine interventional studies and cohort studies, evaluating NBI and AAC techniques in the surveillance of BO, to synthesise the evidence on their effectiveness.

Objectives

General objective
To identify and systematically map the evidence behind the use of NBI and AAC in BO surveillance.

Specific objectives
To identify and systematically map interventional studies of special imaging (NBI and AAC) used during surveillance endoscopy for BO.

To identify the gaps in research supporting use of NBI and AAC.

To assess the effectiveness of NBI and AAC compared with WLE.

To assess whether NBI-guided and/or AAC-guided TBs can replace Seattle protocol biopsies (4QB).

Methods
This systematic review was conducted using the JBI methodological guidance for systematic reviews7 and reported following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines (online supplemental document 1 (PRISMA checklist)). The preliminary exploration phase followed the conceptual frameworks of Arksey and O’Malley and Levac et al for a scoping review, and the proposal was prospectively registered with Open Science Framework (http://osf.io/kv9hu) on 8 October 2024 (DOI: https://doi.org/10.17605/OSF.IO/MRHQZ).8 The final review fully evolved to a systematic review including structured evidence synthesis and the protocol is available through the following link: https://osf.io/2f6mt/overview?view_only=0125be1c50dd4cfc8db99ad7597ead66.

Search strategy and study selection

Data sources
PubMed, Medline, Scopus, Web of Science, Embase and CINAHL were searched. The grey literature was searched using GreyNet and OpenGrey databases. Additional articles were searched by reading references and an experienced librarian from the University of Manchester was consulted to ensure a thorough search strategy was implemented.

Search strategy
A comprehensive search was carried out using the above databases in a stepwise process. The key search terms were generated through literature review and team discussions to include the terms and their synonyms in searching data sources and are as follows: ‘Barrett’s oesophagus, Barrett’s esophagus, Barrett’s oesophagitis, Barrett’s esophagitis, endoscopic surveillance, surveillance endoscopy, endoscopic screening, NBI, acetic acid, chromoendoscopy’. Medical Subject Headings (MeSH) were used for a PubMed/Medline search. A detailed search strategy is provided in online supplemental document 2. All the search results were exported to Covidence software in Research Information Systems (RIS) files.

Eligibility criteria
Peer-reviewed full publications on interventional studies (randomised and non-randomised) and cohort studies (prospective or retrospective), reporting the effectiveness of NBI or AAC used for BO surveillance, were selected. We included eligible studies, reported in English, up to the dates of the database searches (December 2024–January 2025), with no restrictions based on date of publication. Observational studies including case–control studies, case series and single case studies, review articles, letters and opinion papers were excluded. Abstract-only publications and non-English publications were also excluded.

Data charting and evidence extraction
The screening of articles was carried out by two researchers (CG and TDB) independently using Covidence software and any conflicts were resolved by blinded revision by a senior reviewer (YA). The duplicates were removed by the Covidence software and manually reading through titles and abstracts.
The full texts of selected studies were read in detail, and study characteristics were tabulated including title, year of publication, study aims, design, participants, interventions, comparison and outcomes/results and location in a summary of findings (SoF) table.

Outcome evaluation
Clinically relevant findings were extracted, the research landscape was mapped and existing evidence was evaluated for strength of evidence for the research questions on a per-outcome basis using Grading of Recommendations, Assessment, Development and Evaluation (GRADE) criteria.9 The risk of bias (RoB) was assessed using the RoB2 tool for randomised controlled trials (RCTs) and Risk Of Bias In Non-randomised Studies - of Interventions (ROBINS I) tool for non-randomised studies.10 11 The inferences of the selected articles related to each outcome were developed with narrative synthesis considering the significant heterogeneity observed in the methodology and the available results. The level of quality of evidence is expressed as very low (⊕◯◯◯), low (⊕⊕◯◯), moderate (⊕⊕⊕◯) and high (⊕⊕⊕⊕), where applicable.
There are important research questions nested in these studies: whether NBI and AAC are effective compared with WLE for dysplasia detection and diagnosis of BO? Is it cost effective to use NBI/AAC for BO surveillance? Can NBI-guided or AAC-guided TBs replace the Seattle protocol biopsies? We carried out further analysis to synthesise evidence to potentially answer these questions.

Results
A search of the databases produced a total of 974 articles including 6 articles added through reference reading. A total of 389 articles were removed as duplicates by Covidence software and manual identification. Screening by titles and abstract reading was carried out on the remaining 585 articles applying selection criteria and 526 articles were excluded as ineligible. Abstract-only publications (n=15) were also removed, leaving a total of 44 publications for further evaluation, analysis and evidence synthesis (see PRISMA flow diagram in figure 1).
There were 9 (20.5%)/44 RCTs, 29 (65.9%)/46 non-randomised studies and 6 (13.6%)/44 cohort studies assessing various aspects of effectiveness of NBI and/or AAC in the surveillance of BO. These are enumerated in table 1 and detailed in online supplemental documen 3 (SoF).
The evidence was graded for the level of quality, according to each outcome of interest using GRADE criteria. Their RoB was assessed using the RoB2 tool for RCTs and ROBINS I tool for non-RCTs. How these tools were applied is detailed in online supplemental document 4.

Is the NBI effective for BO surveillance compared with WLE for dysplasia detection?
There were 5 RCTs, 13 non-randomised trials and 5 cohort studies addressing this question, as summarised in online supplemental table 1, with their results and levels of evidence. A narrative of the five RCTs is reported below:
Kara et al conducted an RCT assessing whether the combination of HDWLE with NBI (or indigo, carmine chromoendoscopy (ICC)) improves the detection of HGD/OAC in BO. This showed no statistically significant additional benefit by combining NBI (or ICC) compared with HDWLE when assessed for TBs for detection of HGD/OAC, in per-person analysis.12 However, this study is underpowered and susceptible to precision bias as there were only 28 patients included in the randomisation.
Curvers et al published the results of a multicentre (five centres in the Netherlands, the USA and the UK) crossover RCT comparing the diagnostic accuracy of endoscopic trimodal imaging (ETMI) with SVE. During ETMI, first HDWLE followed by autofluorescence imaging (AFI) was performed and the lesions were mapped. Subsequently, a detailed inspection was done with NBI. Targeted and random biopsies were taken after both procedures, and histological yields were compared. Out of 111 recruited patients 24 were excluded after the first procedure and the analysis was performed per-protocol, including results of 87 patients (the calculated sample size was 84).13
Sharma et al conducted an international multicentre RCT (the USA, France and Germany) with a primary aim of assessing probe-based confocal laser endomicroscopy (pCLE). As a secondary aim, it evaluated the per-location sensitivity and specificity of NBI and compared with HDWLE and pCLE. 122 patients undergoing BO surveillance or those referred for management of BO-related neoplasia were randomised to HDWLE or NBI and had a subsequent crossover of procedures followed by pCLE. Lesions were identified and recorded at each procedure and biopsied after all three procedures. Imaging modalities were then compared with histology results. Both per-location (n=874 locations, as 81 were excluded) and per-patient (n=101 as 21 were excluded after randomisation) analyses were conducted.14
Curvers et al published an RCT comparing ETMI (ie, HDWLE, AFI and NBI) (in 51 patients) versus SVE (in 50 patients) for detection of early Barrett’s neoplasia in patients with LGD. This is a multicentre randomised crossover study conducted in a community-based setting with patients having both procedures 6–16 weeks apart. In both procedures, TBs from visible lesions followed by 4QB were obtained.15
Sharma et al published an international multicentre RCT (in the USA and the Netherlands) comparing NBI with TBs against WLE with 4QB for the detection of IM and neoplasia. Out of 139 random patients, 16 were lost to follow-up leaving 123 who had both HDWLE and NBI. Both per-location (267 with NBI and 321 with HDWLE) and per-person analysis were conducted.16

Is AAC effective for BO surveillance compared with WLE for dysplasia detection?
There were two RCTs and eight non-randomised studies addressing this question, as one of the outcome measures of these studies, as summarised in online supplemental table 2, with their results and level of evidence. The two RCTs are narrated below.
Pohl et al conducted a pilot RCT comparing diagnostic accuracy of computed virtual chromoendoscopy with AAC for detecting HGD/OAC in patients with BO with a suspicion of neoplastic progression. 57 underwent two procedures 4–6 weeks apart and TBs were taken from visible lesions followed by 4QB. The two methods showed no statistical difference although the study was not powered enough to draw definitive conclusions.17
Longcroft-Wheaton et al reported a multicentre RCT comparing Seattle 4QB with Portsmouth protocol (AAC with TBs) with mixed qualitative interviews. Results suggest that the AAC may reduce biopsy burden and costs of surveillance. Here, the eligible patients were those with BO without prior neoplasia in contrast to the above study by Pohl et al, which suggested a need for a large study with more than 2800 participants to demonstrate diagnostic equivalence and safety.18

Are these imaging modalities effective in diagnosing BO by detecting columnar epithelium or IM?
There were three RCTs and eight non-randomised studies addressing this question as an outcome measure and are narrated below, with their results and level of evidence reported in online supplemental table 3.
Hoffman et al published an RCT conducted in two referral hospitals in Germany to assess the effectiveness of AAC with Magnification Endoscopy (ME) for detecting Barrett’s epithelium. 35 patients with BO were randomised to undergo either SVE with 4QB or AAC with ME with a 1:1 crossover. Having 4 dropouts, 31 patients were subjected to analysis with 335 AAC-guided TB and 280 4QB. AAC-guided biopsy detected 188 (78%)/241 SCE and 4QB detected 159 (57%)/280 with p<0.001.19
Hoffman et al conducted an RCT comparing the ability of AAC with TBs (46 patients) against WLE with random biopsies (86 patients) for detecting IM in 95 patients with known BO and demonstrated that AAC has a higher diagnostic yield (57% vs 26%; p=0.012).20
Sharma et al conducted a large international (the USA, the Netherlands, the UK) study, randomising 139 patients with BO, with 123 subjected to per-protocol analysis, which showed that both HDWLE and NBI could correctly identify IM in 104 (92%)/113 patients showing no difference.16 However, per-biopsy analysis showed a significant difference in the number of biopsies needed to detect IM (HD-WLE: mean 7.6/patient vs NBI: mean 3.6/patient, p<0.001).

Can NBI-guided or AAC-guided TB replace the Seattle protocol biopsy (4QB)?
There are six RCTs and eight non-randomised studies assessing whether NBI or AAC TBs could provide adequate yield and whether 4QB can be replaced by them.
Kara et al conducted an RCT (as described above and online supplemental table 1) with 28 participants, which showed that 4QB could only detect 1/14 HGD (sensitivity 7%) compared with 12/14 detected by NBI (sensitivity 86%) with TB.12 However, as this is a small sample size, the level of evidence is low due to RoB for this outcome.
Hoffman et al showed that AAC is significantly better for detecting SCE compared with 4QB.19 20 Sharma et al conducted an RCT (described in online supplemental table 1) that showed NBI had higher sensitivity than HDWLE for detecting HGD/EC, although this difference was not statistically significant, and there was no difference in specificity.14 Furthermore, in per-location analysis, NBI has lower missed cases of HGD/EC (statistical significance not given) for HGD/EC than 4QB (HDWLE: 79 (66%)/120 missed and NBI: 70 (58%)/120), when compared with overall histology results (ie, addition of pCLE 4QB and TB). In per-person analysis, HDWLE has missed 4 (13%)/31 HGD/EC cases, while HDWLE or NBI has missed 1 (3%)/31, which were found in 4QB.
Sharma et al conducted a multicentre RCT comparing HDWLE with 4QB against NBI with TB and showed that NBI with TB detected a similar number of IM cases as 4QB while requiring fewer biopsies per patient (3.6 vs 7.6, p<0.0001).16 NBI with TB detected more dysplasia (81 (30%)/267) than HDWLE and 4QB (67 (21%)/321) for overall dysplasia (p=0.01), although there was a significant difference for the detection of HGD/EC (see online supplemental table 1), considering per-location analysis. In per-person analysis, HDWLE missed five HGD/EC (detected as no IM: 1, IM: 1, LGD: 3), whereas NBI missed only two (detected as IM: 1 and LGD: 1).
The Acetic Acid-targeted Biopsies versus non-targeted quadrantic biopsies during Barrett's surveillance (ABBA) feasibility trial by Longcroft-Wheaton et al, compared AAC-guided TB (Portsmouth protocol) with 4QB (Seattle protocol), including 200 patients and 2365 biopsies (226 in AAC arm and 2139 4QB) across six centres. NDR (HGD/EC) was equal, 4.7% (9/192) in both protocols, with AAC having very low number (nearly one tenth) of biopsies. As reported in online supplemental table 2, this study provides low to moderate evidence for this outcome since the number of dysplasia cases in the study was very low.18
The non-randomised studies addressing this outcome include Wolfsen et al, Verna et al, Pohl et al, Curvers et al, Camus et al, Pascarenco et al, Longcroft-Wheaton et al, Lee et al and Elshealta et al.2129 These are described in onlinesupplemental tables 13. They showed that NBI-guided or AAC-guided TB has not demonstrated adequate DDR to be recommended as an independent alternative to 4QB, providing moderate to high-quality evidence.

Does the use of NBI or AAC provide cost-effective surveillance?
There are no RCTs directly answering this question of cost-effectiveness captured in this study. A non-randomised study by Bhandari et al assessed the cost-effectiveness of AAC compared with protocol-guided biopsy. Accordingly, the cost for histological evaluation for 263 procedures by the Cleveland protocol (standard quadrantic with multiple biopsy pots and extensive histological processing) would be £139 416.30 (£530 per procedure), while AAC TB followed by random biopsies in one pot would be £25 032.50 (£95 per procedure) and if only the AAC TB was performed this would be reduced to £9541.8 (£36 per procedure, with a 4% miss rate for neoplasia).30 This provides low-quality evidence in the absence of any RCTs identified in this subject. No other study has directly assessed the cost- effectiveness of these imaging modalities in BO surveillance.
Finally, based on this evaluation, table 2 summarises the important inferences generated through this systematic review related to the outcomes of interest.

Discussion
Advanced imaging modalities have been incorporated into endoscopic BO surveillance for over two decades. The initial studies have focused on the diagnosis of BO with demonstration of IM and SCE in the BO segment. Later, the focus was more directed to the identification of lesions in the BO, advanced imaging to guide TBs and yield of surveillance with DDR/NDR. Furthermore, studies have looked at the possibility of replacing 4QB with advanced endoscopic technique with TB to reduce the service burden and cost. Thus, our review summarises the current evidence and provides a comprehensive analysis of the levels of evidence as to the clinical utility of NBI and ACC in BO surveillance. Some important observations are elucidated below.
When considering NBI, Kara et al12, the first RCT conducted addressing this topic, showed no significant difference between NBI and WLE for detecting HGD/EC. However, it is underpowered (n=28) and liable for RoB. Later, Sharma et al14 also showed a lack of significance in detecting HGD/EC with NBI compared with HDWLE and rather significantly reduced specificity in HGD/EC detection with NBI than HDWLE, in a larger study.14 However, Sharma et al16 demonstrated a statistically high DDR with NBI when LGD, HGD and OAC were taken together, providing good quality evidence with clinically impactful outcome, proving the effectiveness of use of NBI in BO surveillance.16 Nonetheless, several RCTs and non-randomised studies failed to demonstrate the effectiveness of NBI compared with 4 quadrant Seattle protocol biopsy to safely replace 4QB with image-guided TBs, because of risk of missing lesions. This may need a cautious interpretation because the DDR in these studies through advanced imaging may be partially blunted as they were conducted in expert centres where new detections are limited.
AAC has been shown to be effective in diagnosing BO with detection of SCE or IM with fewer biopsies than with HDWLE and 4QB (Hoffman et al 2006 and 201419 20). However, there is a lack of high-grade evidence to support its superiority in detecting dysplasia more than WLE and 4QB.
A systematic review and a meta-analysis by the ASGE technology committee, in an article published in 2016, states that NBI has a sensitivity of 94.2% (95% CI 82.6% to 98.2%), specificity of 97.5% (95% CI 95.1% to 98.7%) and NPV of 94.4% (95% CI 80.5% to 98.6%) for HGD/EC detection. Similarly, the AAC has a sensitivity of 96.6% (95% CI 95% to 98%), specificity of 84.6% (95% CI 69% to 93%) and NPV of 98.3% (95% CI 95% to 99).6
The cost-effectiveness and survival benefit through these techniques are yet to be confirmed. No RCT has been conducted to assess the cost-effectiveness NBI or AAC, although both procedures could reduce the number of biopsies needed for surveillance and potentially result in cost cutting. The recently published BOSS (Barrett’s Oesophagus Surveillance vs endoscopy at need study) trial (a multicentre RCT with over 3400 participants in the UK) demonstrated no survival benefit in 2 yearly surveillances for BO compared with endoscopy at need.31 However, incorporation of improved imaging modalities with higher yield in surveillance and proactive population screening to detect the majority of patients with unidentified BO in population, particularly through a risk stratification model, will be needed to improve the survival related to OAC. Proactive case finding through population screening and referrals to high-quality endoscopic assessment is being studied currently, together with advanced imaging, training programmes for endoscopists and incorporation of artificial intelligence.32 33
This systematic review comprehensively covered the relevant publications but has several limitations. The available evidence is considerably heterogenous, with different study designs like RCTs, non-randomised studies and cohort studies, which further differ in number of participants, imaging and biopsy protocols and outcome reporting, which prevented direct comparisons or quantitative synthesis. Some studies were conducted a few decades before the endoscopic technologies and operator expertise evolved, limiting generalisability. Furthermore, most of the studies we have captured have been conducted within an enriched population and by expert endoscopists. The interpretation of diagnostic performance parametric in BO must be contextualised by the characteristics of the population studied. Further, this is a key limitation in the existing evidence bases, especially for early studies evaluating image-enhanced endoscopy, which were conducted in enriched or high-prevalence cohorts, in tertiary or referral centres, rather than in routine surveillance populations. Moreover, several studies did not report the prevalence of dysplasia or neoplasia within their cohorts, further constraining the applicability of their results. In recognition of these issues, we have explicitly indicated whether each study population was enriched or non-enriched to enhance careful application. Finally, the lack of RCTs and inconsistent reporting of key surveillance outcomes and exclusion of non-English publications are also among limitations.
Taken together, the current evidence highlights both the promise and the limitations of advanced imaging in BO surveillance. While NBI and AAC evidently offer meaningful enhancements in visual assessment, lesion recognition and dysplasia detection, the heterogeneity of existing studies, the lack of adequate RCTs and non-enriched population studies hinder the application of all the evidence to routine practice. Future research integrating advanced imaging, risk stratified pathways and artificial intelligence will be essential to define their true clinical impact.

Conclusions
Endoscopic surveillance of BO with incorporated NBI and guided TBs has a higher overall dysplasia detection compared with WLE and 4QB. AAC has a higher sensitivity for detection of HGD/EC than standard endoscopy, although not proven to improve the dysplasia detection. Both techniques provide moderate to high evidence for the diagnosis of BO and a promise of a significantly fewer number of biopsies to reduce the patient and service burden and the cost of surveillance. However, there is a lack of high-quality evidence to recommend the sole use of NBI-guided or AAC-guided TBs independent of 4QB for BO surveillance in the current literature. The future research priorities should evolve to assess cost effectiveness and survival benefit with incorporation of these advanced imaging techniques achieving high-quality BO surveillance, bearing in mind that these imaging techniques have improved considerably over the last decade with a potential for higher yield. Training endoscopists in NBI/AAC and in the future integration of artificial intelligence into surveillance endoscopy with non-expert centres has the potential to enhance service performance and ultimately improve disease-specific survival.

Supplementary material
10.1136/bmjgast-2025-002157online supplemental file 110.1136/bmjgast-2025-002157online supplemental file 210.1136/bmjgast-2025-002157online supplemental file 310.1136/bmjgast-2025-002157online supplemental file 410.1136/bmjgast-2025-002157online supplemental table 110.1136/bmjgast-2025-002157online supplemental table 210.1136/bmjgast-2025-002157online supplemental table 3