Barrett's esophagus: How can we miss high-grade dysplasia/cancer?

Detection of neoplasia in Barrett’s esophagus (BE) remains a persistent challenge despite major advances in endoscopic imaging and training. A substantial proportion of lesions are still overlooked during initial examinations, highlighting the multifactorial nature of diagnostic performance. Kako et al. should be congratulated for their thoughtful retrospective study of 91 patients with Barrett’s neoplasia (BN), of whom almost one-third were identified as having post-endoscopy Barrett’s neoplasia (PEBN). This multicenter work, covering two decades and six institutions, provides valuable insight into long-term surveillance quality and the determinants of early detection in a low-prevalence setting.

BE remains far less prevalent in Japan than in Western countries, which inevitably limits clinical exposure and may influence diagnostic performance. A lower background prevalence also means that endoscopists encounter fewer Barrett’s cases in routine practice, reducing opportunities for pattern recognition and systematic audit. Volume, in itself, does not define expertise, but sustained exposure and repeated feedback are essential to maintain detection skills. Within this context, the modest overall number of patients identified in the study is understandable, although it underscores the challenge of achieving and measuring proficiency in low-prevalence settings.

Given the 20-year inclusion period, the evolution of endoscopic technology, from standard-definition to high-definition (HD) systems and virtual chromoendoscopy, cannot be ignored. It would be valuable to know whether the retrospectively identified lesions were equally distributed across the HD and non-HD eras, and whether exposure errors were more frequent in examinations performed with earlier, lower-resolution systems.

Kako et al. reported that in nearly two-thirds of cases, a visible lesion had, in fact, been present at the index endoscopy, suggesting that perceptual errors, rather than purely exposure failures, accounted for most missed neoplasia. This finding challenges the common assumption that undetected lesions are mainly the result of technical limitations.

However, it should also be recognized that the review was based on still images rather than full video sequences, which inherently limits interpretation. Even though systematic photodocumentation appeared to have been performed, static images may not fully capture mucosal exposure or dynamic inspection quality. Recent European Society of Gastrointestinal Endoscopy (ESGE) quality recommendations emphasize the importance of structured photodocumentation during Barrett’s surveillance, including images of all landmarks and one picture per centimeter of Barrett’s length, as well as of any visible lesions ¹ . Such systematic imaging not only facilitates external audit and teaching but also provides an objective basis for later review and performance assessment.

A notable feature of the study is absence of systematic use of the Seattle biopsy protocol in Japan, where targeted biopsies remain the rule. In Western practice, this protocol remains the cornerstone of Barrett’s surveillance because it continues to provide most histologic diagnoses of dysplasia or early adenocarcinoma. Robust evidence supports its value: After targeted sampling of visible abnormalities, random four-quadrant biopsies every 2 cm along the Barrett’s segment detect dysplasia in 19.1%, compared with 2.6% when non-protocolized sampling is used (relative risk 6.27, 95% confidence interval [CI] 2.75–14.33) ² . Despite this clear benefit, adherence in daily practice remains variable across studies and centers. In U.S. and European registries, compliance rarely exceeds 50%, whereas recent UK data show that dedicated Barrett’s lists achieve markedly higher guideline adherence and, consequently, a significantly higher dysplasia detection rate (6.3% vs 2.7%) than standard mixed lists ^{1 3 4} . This variability illustrates how organization and training directly influence quality. Systematic implementation of the Seattle protocol could represent one of the most effective and measurable approaches to reducing interval neoplasia, although this remains to be formally demonstrated. In the series by Kako et al., several of the “missed” prevalent lesions might have been histologically detected if random biopsies had been performed according to this standard.

This study highlights how the quality of the index endoscopy largely determines subsequent management and outcomes in patients with BE. Both the ESGE and American Gastroenterological Association have recently updated their quality criteria for Barrett’s surveillance, emphasizing technical performance, mucosal visualization, documentation, and histologic sampling strategy ^{1 3} .

HD systems, now the accepted minimum standard, enhance detection of subtle mucosal abnormalities that standard imaging may miss ⁵ . Because the inclusion period of the current study extends over two decades, such equipment was not consistently available, and comparisons between early and late eras, therefore, should be interpreted with caution.

Mucosal visualization is another critical determinant of diagnostic yield. The ESGE recommends use of simethicone and adequate washing to achieve a clear, bubble-free field, followed by formal assessment of mucosal visibility using a validated scoring system ¹ . As in colonoscopy, one can only detect what is adequately exposed.

Inspection time is equally important. Independent studies have demonstrated a direct correlation between the time spent inspecting the Barrett’s segment and the dysplasia detection rate ^{6 7} . A practical benchmark is to dedicate at least 1 minute per centimeter of circumferential Barrett’s before starting systematic biopsies ^{1 3} . Sequential inspection under white light with maximal luminal distension, followed by virtual chromoendoscopy, is now recommended as the standard technique for Barrett’s examination. The pull-through technique involves slow, continuous withdrawal of the endoscope through an insufflated esophagus to maintain mucosal distension and optimize exposure of all mucosal surfaces. The inspection sequence during Barrett’s surveillance typically combines complementary modalities. Careful withdrawal inspection under HD white light, followed by virtual chromoendoscopy, remains the recommended standard approach. Acetic acid may be used as an adjunct. Early prospective work demonstrated that acetic acid chromoendoscopy increased dysplasia detection (9.4% vs 3.6%) while reducing the number of random biopsies compared with the Seattle protocol ⁸ . The multicenter ABBA trial later confirmed the feasibility and safety of acetic acid-targeted biopsies but found no significant difference in dysplasia detection compared with the standard four-quadrant approach, despite a substantial reduction in biopsy burden ^{9 10} . More recently, the large multicenter ACID trial conducted across Dutch community hospitals reported no additional benefit of acetic acid over meticulous high-quality white-light inspection in a standard surveillance setting (7.4% vs 7.1%, odds ratio 0.97, 95% CI 0.65–1.44) ¹¹ . The high detection rate observed in the standard surveillance arm suggests that examinations were already performed to a high technical standard, which may explain the absence of incremental gain and in addition, low-grade dysplasia (LGD) was included as an outcome, which is often flat and only detected on random biopsies.

Together, these studies indicate that the value of acetic acid lies primarily in optimizing lesion detection in less experienced hands, whereas in expert settings where careful pull-back inspection with white light and virtual chromoendoscopy is systematically performed, its incremental diagnostic benefit appears limited.

Even when the Seattle protocol is meticulously applied during high-quality endoscopy, referring patients with apparently “invisible” dysplasia to expert centers, as recommended by the ESGE, often leads to discovery of previously missed visible lesions. In the Dutch series among patients referred for high-grade dysplasia or early adenocarcinoma without a visible lesion, repeat endoscopy at an expert center revealed a visible lesion in approximately 75% of cases ¹² . Similarly, in another Dutch study, patients referred with apparently flat LGD were found to harbor a visible lesion in about 23% of cases after expert reassessment ¹³ .

This improvement is not merely the result of superior equipment but reflects a diagnostic mindset specific to expert centers, where frequent exposure to dysplastic patterns trains visual perception and cognitive recognition: One detects best what one already knows. Subtle mucosal changes may thus remain “invisible” to non-specialized eyes even when technically visible, as underscored in the current study by Kako et al., where two-thirds of interval neoplasia corresponded to lesions retrospectively visible on the index endoscopy.

Artificial intelligence (AI) may help narrow this gap. The recent BONS-AI study demonstrated that a deep-learning computer-aided detection (CADe) system increased the sensitivity of non-expert endoscopists for BN from 74% to 88% on still images and from 67% to 79% on video sequences, achieving non-inferior performance to international experts without loss of specificity ¹⁴ . Such findings suggest that AI could transfer expert-level visual awareness to community practice, if integration is accompanied by appropriate training, validation, and ethical oversight. But at the dawn of the introduction of AI for detection of BE lesions, mucosal exposure remains important. Recently the Amsterdam group has suggested that a stationary pull through with 10- to 15-second videos and still image every 2 cm can be useful ¹⁵ . This corresponds to a so-called optical Seattle protocol. However, it does not replace random biopsies, which can still catch LGD that may not be endoscopically visible. Another way to prevent PEBN may lie in more accurate assessment of individual risk at the time of the index endoscopy. New adjunctive techniques may help refine such risk stratification. Automated tissue system assessment, using multiplexed immunofluorescence and quantitative image analysis applied to biopsy samples, have shown the ability to predict 5-year progression and to reclassify roughly one-third of indefinite or low-grade cases ¹⁶ . Likewise, wide-area transepithelial sampling with 3-dimensional computer-assisted analysis (WATS-3D) now incorporates deep-learning algorithms that automatically highlight atypical cellular clusters, improving sensitivity across Barrett segment lengths ^{17 18} . Although still considered adjunctive, these tools may help focus surveillance or even early treatment on truly high-risk patients while reducing unnecessary procedures in low-risk individuals.

Ultimately, the goal is not to do more endoscopies but to perform better ones. Too many procedures are still performed for marginal indications, while patients at genuine risk have to wait too long. Focusing endoscopic resources on well-triaged, high-risk individuals, examined with adequate preparation, inspection time, documentation, and biopsy protocol, will yield more benefit than expanding the number of poorly executed examinations.

Defining and measuring PEBN has emerged as a potential quality metric in Barrett’s surveillance ¹ . However, the literature remains inconsistent regarding terminology and time windows. In the NordBEST Study, the term post-endoscopy esophageal neoplasia, conceptually equivalent to PEBN, was used to describe neoplasia detected between 30 and 365 days after the index endoscopy, with cases diagnosed within 30 days classified as prevalent disease (i.e., detected neoplasia) ¹⁹ . In contrast, other series, in a recent meta-analysis, have applied broader intervals, typically ranging from 7 months to 3 years ²⁰ . Kako et al., for instance, defined PEBN as lesions diagnosed between 7 months and 3 years after the initial endoscopy. This heterogeneity reflects the lack of a universally accepted definition, complicating cross-study comparisons and benchmarking. A major disadvantage of retrospective community-based registry studies is that there may, in fact, be an overestimation of “missed” cancers, due to delayed treatment, for instance. It is well known that pathology is often upgraded after resection, with a more benign pathology in the biopsy prior to resection ²¹ . In the study by Kako et al, for instance, it is unclear why 65% of all PEBNs were diagnosed within 7 to 12 months after index endoscopy. Nevertheless, all reports converge on the view that post-endoscopy neoplasia, whatever the time frame adopted, represents a meaningful indicator of diagnostic quality in Barrett’s surveillance and endotherapy programs, meriting formal inclusion among performance metrics in future ESGE quality frameworks. Reported rates across studies vary between 3% and 6%, underscoring both the clinical relevance and potential benchmarking value of this metric ¹ .

Incorporating PEBN assessment within endoscopy units would allow systematic feedback on detection performance and facilitate longitudinal quality monitoring. Tracking the proportion of neoplasia diagnosed within defined intervals after an index endoscopy could serve as a pragmatic outcome-based key performance indicator, much like post-colonoscopy colorectal cancer rates in lower gastrointestinal practice. The challenge now lies in harmonizing definitions and time frames across registries to enable international comparability and guide quality improvement initiatives.

The work of Kako et al. reminds us that Barrett’s endoscopy is not merely a technical act but an entire system of care. Most dysplasia is not invisible, it is unseen. Preventing interval neoplasia requires continuous attention to every link in this chain, from patient selection and procedure preparation to inspection quality, biopsy adherence, pathology review, and referral to experienced centers. Only by embedding these principles into routine practice will we move from missing high-grade dysplasia to managing it effectively.