Objective It is unclear whether endoscopic assessment of scars after colorectal endoscopic mucosal resection (EMR) has to include biopsies, even if endoscopy is negative. Vice versa, endoscopic diagnosis of recurrent adenoma may not require biopsy before endoscopic reinterventions. We prospectively analysed various endoscopic modalities in the diagnosis of recurrence following EMR.
Design We conducted a prospective study of patients undergoing colonoscopy after EMR of large (≥20 mm) colorectal neoplasia. Endoscopists predicted recurrence and confidence level with four imaging modes: high-definition white light (WL) and narrow-band imaging (NBI) with and without near focus (NF). Separately, 26 experienced endoscopists assessed offline images.
Results Two hundred and thirty patients with 255 EMR scars were included. The prevalence of recurrent adenoma was 24%. Diagnostic values were high for all modes (negative predictive value (NPV) ≥97%, positive predictive value (PPV) ≥81%, sensitivity ≥90%, specificity ≥93% and accuracy ≥93%). In high-confidence cases, NBI with NF had NPV of 100% (95% CI 98% to 100%) and sensitivity of 100% (95% CI 93% to 100%). Use of clips at initial EMR increased diagnostic inaccuracy (adjusted OR=1.68(95% CI 1.01 to 2.75)). In offline assessment, specificity was high for all imaging modes (mean: ≥93% (range: 55%–100%)), while sensitivity was significantly higher for NBI-NF (82%(72%–93%)%)) compared with WL (69%(38%–86%); p<0.001), WL-NF (68%(55%–83%); p<0.001) and NBI (71%(59%–90%); p<0.001).
Conclusion Our study demonstrates very high sensitivity and accuracy for all four imaging modalities, especially NBI with NF, for diagnosis of recurrent neoplasia after EMR. Our data strongly suggest that in cases of high confidence negative optical diagnosis based on NBI-NF, no biopsy is needed to confirm absence of recurrence during colorectal EMR follow-up. A high confidence positive optical diagnosis can lead to immediate resection of any suspicious area. In all cases of low confidence, biopsy is still required.
Trial registration number NCT02668198.
- endoscopic procedures
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Wide-field endoscopic mucosal resection (EMR) is established as a safe and effective endoscopic alternative to surgery for the removal of large colorectal polyps (>20 mm). Recurrent adenoma is one of the main limitations of EMR as up to 20% occurs at first surveillance colonoscopy and 4% after 12 months.
Post-EMR surveillance is usually performed at 4–6 months after initial procedure and is important for detection of recurrent neoplastic tissue.
The current standard of care is to biopsy, or resects, each post-EMR site. It is essential to establish accuracy of the optical diagnosis of recurrent neoplastic tissue with enhanced imaging techniques.
This approach may alter clinical practice by avoiding unnecessary biopsies when disease is absent and properly directing repeat resection when disease is present. In addition to narrow band imaging an optical zoom function, that is, near focus, might increase the optical diagnostic accuracy through sharper visualisation of the mucosal pit pattern.
Significance of this study
What are the new findings?
This is a large study in the field of post-EMR scar assessment.
This study demonstrated a very high negative predictive value (NPV) and good diagnostic accuracy for all four imaging modalities, with an especially high NPV of 100% using narrow-band imaging (NBI) with near focus for the optical diagnosis of residual neoplasia when assessed with high confidence.
In offline image analyses performed by 26 experienced endoscopists, NBI with near focus was associated with a higher proportion of high-confidence diagnosis and a significantly higher sensitivity.
How might it impact on clinical practice in the foreseeable future?
The results from this study suggest that advanced imaging methods such as NBI with near focus achieve sufficient diagnostic accuracy to exclude recurrence without the need for biopsy.
Colorectal cancer is the fourth leading cause of cancer-related death in England.1 Most colorectal cancers originate from adenomas2 or sessile serrated lesions.3 Screening through colonoscopy enables early detection and removal of colorectal adenomas leading to significantly improved colorectal cancer-related survival.4 Endoscopic mucosal resection (EMR) is well established as management for large laterally spreading colorectal polyps that cannot be managed by conventional polypectomy5–8 and is effective and relatively safe.9 The technique involves submucosal injection of a lifting fluid medium to safely resect the lesion with a snare device. EMR provides curative resection and avoids the higher mortality, morbidity and costs associated with surgical management.10 11 However, recurrent adenomatous tissue is a key limitation. A recent systematic review reported a recurrence rate during follow-up of 20% (95% CI 16% to 25%) for piecemeal EMR, although most of these can be managed by additional EMR.12
Surveillance colonoscopy within 6 months after piecemeal EMR to control for recurrent tissue is recommended.13 During surveillance for potential recurrent neoplastic tissue, several imaging technologies, such as narrow-band imaging (NBI; Olympus, Japan), I-Scan (Pentax Medical) or blue laser/light imaging (Fujifilm, Japan), can aid in the visualisation of surface and vascular patterns that reflect the histology of the colorectal mucosal tissue.14 However, none of these imaging modalities have been validated for surveillance after EMR. Limited optical accuracy leads to extensive reliance on histology with associated excess use of biopsy or resection of benign tissue and occasional non-treatment of residual neoplasia. If physicians could optically establish the absence of recurrent neoplastic tissue with high accuracy, then unnecessary biopsies and resections could be avoided. In addition to NBI, a new optical zoom function (ie, near focus (NF)) that provides sharp focus within 2–6 mm of the mucosa has been developed to better visualise the intestinal mucosa.15 16
We hypothesised that optical assessment of colonic mucosa using NBI with NF magnification would improve the accuracy of real-time diagnosis of EMR scar histology. We specifically aimed to estimate the diagnostic accuracy and level of confidence of four imaging modalities: (1) high-definition white light (WL) imaging alone, (2) WL imaging with near focus (WL-NF), (3) NBI alone and (4) NBI with NF, for the optical diagnosis of recurrent neoplasia at a previous EMR site. Furthermore, we aimed to assess whether use of clips during the initial EMR could influence the real-time diagnostic accuracy and treatment choices.
Materials and methods
This prospective observational study was conducted at the Mayo Clinic, a tertiary care centre in Jacksonville, Florida, from 1 December 2015 to 25 November 2017. The study was conducted and reported according to the Standards for Reporting Diagnostic accuracy studies17 18 guidelines (online supplementary appendix 2 and 3). The study was approved by the Mayo Clinic Institutional Review Board (#16–0 08 771). All patients provided written informed research-specific consent prior to the study.
Patient selection and recruitment
Consecutive patients undergoing follow-up surveillance colonoscopy after a previous EMR of 20 mm or larger non-invasive colorectal neoplasia were included. Exclusion criteria were patients with inflammatory bowel disease, initially incomplete EMR, poor bowel preparation (Boston Bowel Preparation Scale score <5), coagulopathy or thrombocytopenia and patients who did not provide informed consent. Patients were also excluded if the post-EMR site could not be identified during colonoscopy.
All patients received split bowel preparation the evening before and morning of colonoscopy. Colonoscopy was performed using Olympus CF H190L/I EVIS EXERA III colonoscope (Olympus Medical Systems) with NBI and NF capacity. This optical system provides in-focus depth of field of 2–6 mm from the optical lens with 65× zoom (NF at a 2 mm distance), allowing true optical, in-focus zoom, compared with 5–100 mm depth of field in normal focus with 52× zoom (normal focus at a 5 mm distance). NF magnification is achieved with a small mechanical actuator that microscopically moves the lens via an electrical signal, controlled by a single on/off button on the colonoscope handle.16 After completion of optical diagnosis, all lesions were sampled histologically. If there was any suspicion of recurrence (low or high confidence), the tissue was treated by endoscopic resection using standard methods (snare or avulsion) until all visible suspicious tissue was removed. If there was no suspicion of recurrence, the site was sampled by at least 2–4 biopsies of the scar edge. All patients were under close observation by the multidisciplinary medical team to reduce the risk of adverse events during the procedure. Adverse events were not collected after the procedure.
Patient sex and age and initial EMR site baseline characteristics were gathered from patient records charts, endoscopy reports, histology reports and medical correspondence prior to post-EMR site assessment. EMR site characteristics consisted of the date of initial EMR; number of follow-up visits; location of previous EMR (ie, proximal colon from cecum until transverse colon vs distal colon from splenic flexure until rectum); initial size of the lesion EMR type (ie, piecemeal or en bloc); use of adjuvant thermal ablation; use of clips; and initial histology, including tubular, tubulovillous or sessile serrated (with or without high-grade dysplasia) and intramucosal adenocarcinoma.
Real-time assessment of EMR scars
Five EMR endoscopists (EPB, VG, MR, TAW and MBW) participated in real-time assessment of EMR scars. All five received training in scar assessment using NBI International Colorectal Endoscopic (NICE) classification for optical diagnosis of polyp histology.19 Although the NICE criteria were not developed for the assessment of post-EMR sites, the assessment of mucosal pit patterns was used to differentiate recurrent adenomatous tissue from hyperplastic tissue. Endoscopists performed initial examination of EMR scars and predicted, in real-time, the presence of recurrent neoplasia, noting their level of confidence (high or low). Their optical diagnosis was based on sequential assessment of the previous EMR site with the following four imaging high definition modalities: (1) WL colonoscopy alone, (2) WL with NF, (3) NBI alone and (4) NBI with NF. We chose this order, because it most closely resembles clinical practice. An image was captured from every post-EMR site using each imaging modality (figures 1 and 2). Following image capture, each EMR site was biopsied or resected as a reference standard. A pathologist, blinded to the endoscopists’ optical predictions, assessed the histology. Tubular adenomatous, tubulovillous adenomatous, or sessile serrated histology, with or without dysplasia, or intramucosal adenocarcinoma was considered as presence of recurrent neoplastic tissue. Benign colonic mucosa, with or without reactive changes, and lack of pathological changes, granulation tissue and lymphoid aggregates were considered as absence of recurrent neoplastic lesions.
Offline assessment of EMR scars
To reduce the confirmation bias, which was intrinsic to the real-time study (ie, endoscopists who assess a scar to be ‘positive’ with one modality may have been more likely to call it ‘positive’ with subsequent modalities), we conducted an image-based, offline analysis to validate diagnostic accuracy with each imaging modality within a larger, independent group of experts and assess the interobserver variability. We selected 58 high-quality images (29 with and 29 without recurrence) from each of the four imaging modalities made during the real-time assessment of this observational study. Endoscopists were blinded to patient-related factors and final histology. We used an online web application Blackboard (Blackboard) platform for offline assessment of images. Twenty-six EMR experts with more than 10 years of experience, from three continents (North America, Europe and Australia), who did not participate in the real-time part of the study participated in the offline part (online supplementary appendix 1). They first conducted a pretest consisting of 24 EMR scar images (six per imaging modality) in which they received immediate feedback on the presence and absence of recurrent neoplasia. These 24 images were only used during this pretest. Subsequently, a training module was shown consisting of images displaying presence and absence of recurrent neoplasia with different imaging modes, NICE classification19 and an American Society of Gastroenterology (ASGE) expert video module20 demonstrating scar assessment (online supplementary appendix 4). Finally, the 26 endoscopists independently assessed 58 EMR scar sites (29 with and 29 without recurrence) imaged using each of the four imaging modalities (a total of 232 images). To reduce confirmation bias, we randomised the order of images by imaging modality and EMR scar site for each endoscopist. The order of presentation was randomised such that single polyps were not present in sequential order, as was done in contrast with the real-time study. Endoscopists were blinded to patient-related factors and final histology. Because the images came from the real-time part of the study, histology was the reference standard.
Endoscopic haemostatic clipping
Application of haemostatic clips after EMR is commonly used to reduce post-EMR bleeding. Post-EMR scars are typically smooth and flat in the absence of recurrent polyp or clip closure. When clips are placed to mark post-EMR defects, there may be changes on surface pattern of scars with bumps or distortions of tissue on follow-up colonoscopy. This bumpy tissue, often called clip artefact, has normal pit pattern and non-neoplastic histology. Failure to recognise clip artefact as a normal tissue may lead to unnecessary treatment of non-neoplastic tissue. In a secondary analysis of our study, we reported the effect of clip closure in post-EMR scar assessment.
Sample size calculation
In the first phase of this study, we included 230 patients (255 EMR sites) to obtain estimates of the diagnostic utility of NBI with NF in the optical diagnosis of post-EMR sites. The number of patients was determined a priori based on a combination of feasibility and statistical considerations. We specifically aimed for a high negative predictive value (NPV) with minimal loss of specificity. The ASGE used a cut-off for NPV of 95% or higher combined with a specificity of 80% or higher as sufficiently high to safely avoid biopsies in the setting of Barrett oesophagus.21 Among EMR sites with a high confidence diagnosis using NBI with NF, we calculated that a sample size of at least 230 EMR sites was needed to detect a lower bound of at least 95% for NPV assuming an expected NPV of 99% with a probability of 0.80 for the two-sided 90% Wilson score CI and a probability of 0.64 for the two-sided 95% Wilson score CI. Sample size estimates were based on assumptions from preliminary data of 52 EMR sites with optical diagnosis using NBI with NF, where 87% had a high confidence diagnosis and 56% had a negative optical diagnosis.
For the offline assessment, we included 26 endoscopists for purposes of comparing sensitivity and specificity at the endoscopist level between the four modalities. Our initial sample size calculation of 20 endoscopists would give more than 80% power to detect moderate to large differences in sensitivity or specificity between the four modalities (eg, 99% vs 95%, 95% vs 88% or 80% vs 70%) after adjusting for multiple comparisons (six pairwise comparisons, p≤0.0083 considered statistically significant) based on the Wilcoxon signed-rank test assuming only 29 EMR sites were available. Precision of the endoscopist-level estimates of sensitivity and specificity for each of the four modalities was limited by only having 29 EMR sites available for evaluation. For example, the distance from the estimated sensitivity or specificity to the lower bound of the 95% Wilson score CI would be 18% if the actual sensitivity or specificity was 80% and 14% if the actual sensitivity or specificity was 95%.
For descriptive statistics, continuous variables are displayed as median (range) and categorical variables as number and percentages.
For the real-time optical diagnoses, we calculated NPV, positive predictive value (PPV), sensitivity, specificity and overall diagnostic accuracy with 95% Wilson CIs per imaging mode for all EMR sites and for those assessed with high confidence. In a post hoc analysis, we explored the association of clip use during the initial EMR with high confidence diagnosis and accuracy of optical diagnosis using mixed effects logistic regression model with random effects for endoscopist and adjustment for modality and histological recurrence.
For the primary analysis of the offline image-based assessment, we estimated each endoscopist’s sensitivity and specificity for each modality. Given we selected an equal number of images with and without histological recurrence for inclusion in the study, we did not estimate NPV, PPV and diagnostic accuracy as the prevalence of histological recurrence is higher compared with what is observed in a clinical setting. The non-parametric Friedman χ2 test, stratified by endoscopist, was used to evaluate evidence of an association of modality with sensitivity and specificity (p<0.05 was considered statistically significant). If the association of modality with sensitivity or specificity was statistically significant, the Wilcoxon signed-rank test was used to evaluate the six pairwise comparisons between modalities with adjustment for multiple testing using the Holm step-down method as extension of Bonferroni.
Interobserver agreement was expressed as a proportion of pairwise agreements and by Conger’s κ22 with 95% bootstrap CIs per imaging modality for all EMR sites. The proportion of pairwise agreements indicates the probability that a pair of randomly selected observers will agree on the diagnosis for a randomly chosen image. This was estimated based on the 325 endoscopist pairs and 29 images for each modality, separately for those images with and without histological recurrence. Values for κ were interpreted according to the scale proposed by Landis and Koch23 (ie, 0.41–0.60: moderate agreement; 0.61–0.80: substantial agreement; 0.81–1.00: almost perfect agreement). Interobserver agreement was estimated using the Reliability Coefficients package in R.9
We considered a two-sided p value of less than 0.05 as statistically significant except for when the Holm step-down method for adjusting for multiple testing was used. Statistical analyses were performed with SAS statistical software (V.9.4, SAS Institute Inc., Cary, NC) and R language environment for statistical computing version 3.1.3 (R Foundation).
From 268 potentially eligible patients, we enrolled 230 with 255 EMR scar sites who came for surveillance colonoscopy for follow-up after EMR of non-invasive neoplastic lesions of 20 mm or larger (table 1, figure 3, online supplementary appendix 5). The first surveillance colonoscopy was included for 189 (74.1%) lesions. The median follow-up time since EMR was 7 (1–71) months. Two hundred and forty-eight (97.3%) lesions had been removed by piecemeal technique, with the majority of lesions being located in the proximal colon (207 (81.2%)). The most common initial histology was sessile serrated (35.0%), followed by tubular (34.5%) and tubulovillous (29.0%). Clips were used in 111 (43.5%) cases during initial EMR, with clip artefact noted by the endoscopist in 59/111 (53.2%) and clip retention in 8 (7.2%). Recurrent adenoma was histologically confirmed in 62 of 255 (24.3%) post-EMR sites. There were no serious adverse events following the surveillance colonoscopy.
Real-time diagnostic accuracy
Diagnostic values (NPV, PPV, sensitivity, specificity and overall diagnostic accuracy) of optical diagnosis of recurrent adenoma per imaging modality for all cases and for high-confidence diagnoses are shown in table 2. Diagnostic values were high for all four imaging modes (NPV ≥97%, PPV ≥81%, sensitivity ≥90%, specificity ≥93% and accuracy ≥93%). All diagnostic values were even higher when limited to only those with a high-confidence diagnosis (table 2). Furthermore, in the setting of a high-confidence diagnosis, only NBI with NF had an NPV of 100% (95% CI 97.8% to 100%) and sensitivity of 100% (95% CI 93.2% to 100%).
Impact of clip use during initial EMR on subsequent scar assessment
When real-time optical diagnostic accuracy was assessed stratified for clip use during the initial EMR, the PPV was generally lower, regardless of the imaging modality (table 3). Furthermore, no clip use during the initial EMR was associated with optical diagnostic accuracy (adjusted OR=1.68, 95% CI 1.01 to 2.75) and with a high-confidence diagnosis (adjusted OR=2.34, 95% CI 1.62 to 3.36). Overtreatment (ie, snare polypectomy of benign granulation tissue) was significantly more common in EMR sites where clips were used previously compared with cases where no clips were used (24% (95% CI 16% to 33%) versus 8% (95% CI 3% to 15%); p=0.01).
Diagnostic accuracy in image-based offline assessment
Patient and lesion characteristics of the images used in the offline image assessment can be found in online supplementary appendix 6. The sensitivity and specificity per imaging modality per endoscopist are depicted in table 4. The mean (range) endoscopist-level specificity was 95% (83%–100%) for WL, 95% (76%–100%) for WL with NF, 93% (62%–100%) for NBI and 93% (55%–100%) for NBI with NF. Imaging mode was not associated with specificity (p=0.43) but was associated with sensitivity (p<0.001) when stratified by endoscopist using the Friedman test. Pairwise comparisons between imaging modes using Wilcoxon signed-rank tests showed that endoscopist-level sensitivity was higher for NBI with NF (mean 82% (range 72%–93%)) compared with WL (69% (38–86%), p<0.001), WL with NF (68% (55–83%), p<0.001) and NBI (71% (59–90%), p<0.001) after adjustment for multiple testing using the Holm step-down method (p≤0.0167 considered statistically significant). It is worth noting there was some evidence of higher sensitivity with NBI compared with WL with NF (71% vs 68%, p=0.04), but this was not statistically significant after adjustment for multiple testing. There were no other significant differences in endoscopist-level sensitivity between imaging modes (all p≥0.29). Similar findings were observed in a supplemental analysis of only high-confidence offline assessments where imaging mode was associated with sensitivity (p<0.001), but not with specificity (p=0.63) using the Friedman test. The mean (range) endoscopist-level specificity of high-confidence offline assessments was 96% (79%–100%) for WL, 96% (72%–100%) for WL with NF, 95% (62%–100%) for NBI and 96% (52%–100%) for NBI with NF. Endoscopist-level sensitivity of high-confidence offline assessments was higher for NBI with NF (mean 92% (range 78%–100%)) compared with WL (74% (43–94%), p<0.001), WL with NF (76% (63–95%), p<0.001) and NBI (79% (57–95%), p<0.001) after adjustment for multiple testing using the Holm step-down method (p≤0.0167 considered statistically significant). There was some evidence to suggest a possible improvement in sensitivity with NBI compared with both WL (p=0.05) and WL with NF (p=0.092), but these were not statistically significant after adjustment for multiple testing. There was no evidence of a difference in sensitivity between WL and WL with NF when limited to high-confidence offline assessments (p=0.38). NBI with NF was also associated with the highest proportion of high-confidence diagnoses compared with the other image modalities. The proportion of pairwise agreements (WL 84%, WL with NF 86%, NBI 83% and NBI with NF 87%) (table 5) in combination with the κ coefficients (WL: 0.66, 95% CI 0.56 to 0.74, WL with NF: 0.69, 95% CI 0.59 to 0.78, NBI: 0.64, 95% CI 0.54 to 0.72, NBI with NF: 0.73, 95% CI 0.64 to 0.80) indicate substantial interobserver agreement for all four modalities.
Our study suggests that optical diagnosis of recurrent colorectal neoplasia following EMR can be made in real time with high confidence and diagnostic accuracy using currently available endoscopic imaging. NBI with NF achieved a notably higher proportion of high-confidence optical diagnoses, both in real time and offline testing. Sensitivity and NPV of NBI with NF in the setting of a high-confidence diagnosis was 100%, suggesting that this modality may obviate the need for biopsies of benign appearing scars. Use of a clip during the initial EMR was not associated with an important change in the diagnostic accuracy of the imaging modalities in the setting of EMR site follow-up assessment.
Post-EMR scars assessment is usually conducted 4–6 months after initial EMR to ensure adequate detection and treatment of recurrent adenoma. In 2014, Knabe et al24 reported a large trial of 243 EMR cases, with almost 32% having recurrent at first follow-up. Of these, 7% were detected only on biopsy, with macroscopically normal appearing scars. Such data suggested the need to sample all scars for histological (biopsy or snare resection) assessment. The key difference from our study was the overall higher rate of residual disease (and thus pretest probability) and the use of older generation (500 series Fujifilm, 180 series Olympus) endoscopes without NF or spectrally enhanced (blue light/narrow band) imaging.24 To the best of our knowledge, accuracy and utility of NBI with NF colonoscopy for the assessment of post-EMR scars has not been reported. Our findings of high diagnostic accuracy and CIs are in line with earlier findings for small colorectal polyps. ASGE assessment of real-time endoscopic imaging of colorectal polyps suggested that NBI was sufficiently accurate to assess the suspected histology of small (ie, ≤5 mm) polyps.25 More recently, a study demonstrated that NBI with NF may replace pathological diagnosis for the majority of diminutive colorectal polyps.26 Furthermore, the use of NF was associated with an increase in the proportion of polyps assessed with high confidence compared with HD-WL.26 A study by Kuruvilla et al27 demonstrated that the use of a zoom feature in combination with NBI increased the confidence of polyp histology from 68% to 99.3%. Despite these advances, real-time optical diagnosis has been limited by lower accuracy outside of expert centres.28
Other ways to enhance visualisation of the mucosal surface include chromoendoscopy (both dye-based and virtual) and confocal laser endomicroscopy. Although our group has reported a high NPV for the combination of virtual chromoendoscopy and probe-based confocal laser endomicroscopy (pCLE) on assessment of post-EMR scars,29 pCLE is expensive, requires special expertise and exogenous intravenous contrast and is not widely used in practice. A study by Desomer et al30 proposed a standardised imaging protocol to determine the accuracy of conventional imaging modes in post-EMR scar assessment with HD-WL and NBI using Olympus high-definition 180 or 190 series CF or PCF colonoscopes. Unfortunately, the level of confidence of the endoscopists, use of NF and diagnostic accuracy in offline images was not assessed in that particular study. Results showed that sensitivity was higher when the post-EMR scar was assessed with both WL and NBI as opposed to WL alone (93% vs 67%) with a high specificity with and without NBI (94% vs 96%).30 In our study, we noted a substantially higher sensitivity with WL.
Endoscopic clip closure is commonly used following EMR to decrease the risk of bleeding. The closure defect of margins with clips may give rise to difficulty in assessing the post-EMR scar, since hyperplastic or granulation tissue might be mistaken for recurrent neoplastic tissue. The term post-EMR clip artefact, a bumpy scar that has a normal pit pattern and normal histology on biopsy, has been coined for this phenomenon.31 Unnecessary resection of benign mucosal tissue might occur if the presence of recurrent neoplastic tissue is incorrectly predicted. Clip use during the initial EMR was found to be associated with a decrease in the proportion of high-confidence optical diagnoses and a lower PPV compared with cases where no clip was used during the initial EMR, regardless of the imaging modality used.
Our study is strengthened by the blinded assessment of the reference standard, consecutive patient series and assessment by a large number of physicians and institutions. Whether non-EMR expert endoscopists can achieve the same diagnostic accuracy remains to be determined.
Our study also has limitations. Real-time assessment of diagnostic accuracy per imaging modality was sequentially assessed. Thereby, assessment of any subsequent imaging modality was potentially biased by the previous modalities. Nonetheless, we assume the imaging modalities to be used in the same order in clinical practice, and thus, diagnostic values mirror real-life usage. In our offline study, we attempted to minimise this bias by presenting the images in random order (both by modality and by scar site) and found similar diagnostic values for the imaging modalities, although the significant differences in imaging accuracy in offline assessment may suggest a true difference, whereas the lack of difference between modalities in the real-time study is potentially explained by confirmation bias. A disadvantage of the offline mode was that endoscopists could not view the full set of video and still images, wash/suction and view from different angles as is done in real-time cases. As expected, this limitation was associated with lower overall accuracy. We did use the NICE classification within our training tool for optical diagnoses, but this classification has not been validated for optical zoom endoscopy or EMR scar assessment. It was developed to evaluate adenomatous versus hyperplastic tissue, which is a reasonable surrogate for recurrent adenoma versus scar. Although the image-based analyses generally confirmed our findings for real-time assessment, diagnostic accuracy values were lower for the offline assessment. Possible explanations for these lower values are the offline endoscopists had less opportunity to thoroughly evaluate the lesions and some information can be better appreciated by video. Although appropriately powered, testing diagnostic accuracy offline for all EMR scars from video images with all imaging modalities might have mimicked daily practice even better. Finally, only 70% of EMR cases performed at our centre came for a follow-up colonoscopy, and thus, could not be included in the study. Nevertheless, we do not expect the diagnostic accuracy for the post-EMR sites of these patients to be different.
In conclusion, results from our study suggest high NPV and good diagnostic accuracy for all four imaging modalities, with an especially high NPV of 100% using NBI with NF for the optical diagnosis of residual neoplasia when assessed with high confidence. Although it still needs to be confirmed in an independent cohort, our data strongly suggest that in cases of high-confidence negative optical diagnosis based on NBI NF, no biopsy is needed to confirm absence of recurrence during colorectal EMR follow-up. In cases of low-confidence or high-confidence positive optical diagnosis, resection of any suspicious area is recommended.
In offline image analyses, NBI with NF was associated with a higher proportion of high-confidence diagnosis and the highest NPV. Sufficient diagnostic accuracy to diagnose the absence of recurrence can be reached, especially using NBI with NF, which could potentially avoid biopsies of EMR scars in the future. Clip use during the initial EMR did not substantially change the NPV but did lower the proportion of high-confidence diagnoses and PPVs during follow-up.
Patient consent for publication Not required.
Contributors Study concept and design: MBW, PK, W-CC and ECB. Acquisition of data and database preparation: PK and JP. Statistical analysis: PK, ECB and CTB. Interpretation of the results: all authors. Drafting of the manuscript: PK, ECB and MBW Critical revision of the article for important intellectual content: all authors. Study supervision: MBW. Final approval of the article: all authors.
Funding PK: Joyce E Baker Fund for Gastrointestinal Research at Mayo Clinic Florida. ECB received an unrestricted scientific internship abroad grant from the Dutch Digestive Foundation (16-03S).
Disclaimer Mayo Clinic does not endorse specific products or services included in this article.
Competing interests MBW reports consulting income from iLUmen and Interscope and grant support from Boston Scientific, Olympus, Medtronic and Cosmo pharmaceuticals.
Provenance and peer review Not commissioned; externally peer reviewed.
Collaborators Group members and their affiliations are available in supplementary appendix 1.
Author note Portions of this manuscript have been published in abstract form: Kandel P, Brand EC, Chen WC, et al. 690 Diagnostic accuracy of optical detection of colorectal neoplasia after endoscopic mucosal resection: prospective double blind comparison of high definition white light, narrow band imaging and near focus. Gastrointest Endosc 2017;85:AB101–2. doi: 10.1016/j.gie.2017.03.149.
Correction notice This article has been corrected since it published Online First. The abstract and results section have been amended.
Presented at Presented at Digestive Diseases Week, Chicago, Illinois, 6–9 May 2017.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.