Society: ASGE

LIVE STREAM SESSION

Background: There is a lack of data on training benchmarks to define competence in colorectal EMR (C-EMR) among advanced endoscopy trainees (AETs). Previous pilot data from our group demonstrated a relatively low proportion of AETs achieve competence on key cognitive and technical aspects of C-EMR. We aimed to perform an interim analysis on C-EMR training among AETs and assess their performance using the EMR-STAT during the first trimester of their advanced endoscopy fellowship (AEF).
Methods: Prospective multicenter study evaluating AETs C-EMR training using the EMR-STAT. The tool was previously validated in the pilot study for standardized evaluation of key cognitive and technical C-EMR skills (Figure 1). A 4-point scoring system was used to grade these endpoints. Global rating was provided using a 10-point scoring system. For interim analysis, competence was defined as a score of 3 or 4 for each endpoint and ≥7 for overall assessment. Cumulative sum analysis was used to establish competence for cognitive and technical components of C-EMR and overall performance. Prior to the study, participating AETs completed questionnaire about their GI fellowship training in endoscopic resection.
Results: Twenty-five AETs from 18 institutions are enrolled in this ongoing study. On survey questionnaire, the AETs reported having performed a mean of 41.4 C-EMRs (interquartile range [IQR]: 10-50) before the onset of their AEF and most received cognitive training in C-EMR during their general GI fellowship (n=20; 80%). In the first trimester of their AEF, out of the 25 AETs, 15 have performed a mean of 9.1 C-EMRs (range 1-30). Mean lesion size was 26.7±11.6 and mean EMR time of 26.1±18.1 minutes. En-bloc resection rate for polyp sizes 11-20 mm was 41.3% (19/46). Competence in cognitive skills, such as assessment of polyp morphology and pit/vascular pattern, was achieved by AETs in 90.4% and 83.1%, respectively. AETs were graded as competent in submucosal lift injection and snare resection in 69.9% and 63.2%, respectively. Overall competence based on the global score was attained in 53.7% of the cases. On cumulative sum analysis, only 2 AETs crossed the competence threshold for cognitive skills and 1 AET for technical skills. The minimum threshold to achieve competence was 18 C-EMRs (Figure 2).
Conclusions: Standardized evaluation of competence in C-EMR training is critical for quality assurance in patient care. There was high variability in the number of C-EMRs performed by AETs and low overall en-bloc resection rates for polyps 11-20 mm in size. In aggregate, AETs were graded as competent in only half of the C-EMR cases and only 2 AETs have crossed the minimum threshold of competence. Ongoing data acquisition from this study will provide insight into the current state of C-EMR training during AEF and establish competence thresholds for quality metrics.

Figure 1. EMR Standardized Assessment Tool (EMR-STAT)

Figure 1. EMR Standardized Assessment Tool (EMR-STAT)

Figure 2. Representation of the learning curves among AETs by using cumulative sum (CUSUM) analysis for core cognitive and technical skills during C-EMR training.

Figure 2. Representation of the learning curves among AETs by using cumulative sum (CUSUM) analysis for core cognitive and technical skills during C-EMR training.

Background: The recently developed CAD EYE system (Fujifilm, Tokyo, Japan), which provides artificial intelligence (AI) -aided endoscopic diagnosis, has the potential to improve the detection for colorectal polyps. It is essential that gastroenterology trainees improve the quality of total colonoscopy (CS) operations and accelerate their technical progress. The aim of this study was to determine the utility of CAD EYE for CS by comparing endoscopic observation using CAD EYE with conventional endoscopic observation (i.e., white light imaging) in outpatients undergoing CS performed by gastroenterology trainees (i.e., beginner endoscopists).
Methods: This was a multi-center, randomized controlled trial at Ureshino Medical Center, Karatsu Red Cross Hospital and Saga University Hospital (UMIN000044031). The study received an academic research grant from the Japanese Society of Gastrointestinal Endoscopy in 2021. Patients were divided into group A (observed using CAD EYE) and group B (observed using white light imaging). Six gastroenterologists with limited experience in CS (i.e., trainees in their third or fourth year after graduation) performed CS using a back-to-back method in pairs with a gastroenterology specialist. The primary endpoint was the adenoma detection rate. The secondary endpoints were the adenoma miss rate (AMR) and 14 assessment of competency in endoscopy tool scores. The learning curve of each trainee was evaluated using the cumulative sum control chart.
Results: We analyzed 231 cases (113 in group A, 118 in group B) enrolled from May 2021 to March 2022. There was no difference in the adenoma detection rate of trainees between group A and group B (58.4% versus 61.0%, respectively; p=0.690). There was a significantly lower AMR (26.6% versus 39.7%, respectively; p=0.036) and number of missed adenomas per patient (0.5 versus 0.9, respectively; p=0.004) in group A compared with group B. Group A also scored significantly higher than group B on two items of the assessment of competency in endoscopy tool score—i.e., pathology identification (2.26 versus 2.07, respectively; p=0.030) and interpretation and identifying location of pathology (2.18 versus 2.00, respectively; p=0.038). For the cumulative sum learning curve of trainees, the number of cases in which multiple adenomas were missed by the six trainees who performed CS was lower in group A. Even after accumulating cases, the number of missed adenomas remained consistently lower in group A.
Conclusions: The use of CAD EYE can decrease the AMR and improve the ability to accurately locate and identify colorectal adenomas. Thus, CAD EYE is particularly useful for CS in beginning endoscopists.

Background:
Epidemiologic data highlight the suboptimal impact of screening and surveillance in Barrett’s esophagus (BE). PEEC and PEEN, similar to post-colonoscopy colorectal cancer, undermines the effectiveness of these practices. Using a population-based cohort study, we aimed to conduct a time trend analysis on PEEC and PEEN rates and neoplasia detection rate (NDR) among newly-diagnosed BE patients.

Methods:
This study was conducted in Denmark, Finland and Sweden from 2006-2020. We included data from the national patient, cancer, causes of death and prescribed drug registries. Patients with newly diagnosed BE were included and excluded patients with prior upper GI cancer, or BE endoscopic therapy. Demographics, country, year of diagnosis, smoking, Charlson comorbidity index score, medications, and hospital volume were collected. PEEC and PEEN were defined as EAC or HGD/EAC, respectively, diagnosed 30-365 days from the index endoscopy that diagnosed BE. NDR was defined as HGD/EAC diagnosed from 0-29 days and incident HGD/EAC was diagnosed >365 days. Patients were followed until diagnosis of HGD/EAC, death or end of study period. Incidence rates (IR)/100,000 person-years (pyrs) for the entire study cohort and for 3 calendar periods: 2006-2010, 2011-2015 and 2016-2020 and incidence rate ratios (IRR) with 95% CI with Poisson regression were calculated. A graphical representation of the predicted probability of EAC by month in the first year after entry is reported and for the 3 calendar periods.

Results:
20,588 newly diagnosed BE patients (mean age 64.6 years, 67% men, Sweden 64.6%, Denmark 14.8% and Finland 20.6%) were included. 279 patients were diagnosed with EAC: NDR (41, 14.6%), PEEC (65, 23.2%) and incident EAC (173, 62%). Overall IRs for NDR, PEEC and incident EAC were 2521 (95% CI 1857-3425), 369 (289-470) and 199 (172-231) /100,000 pyrs. The IRs/100,000 pyrs in the 3 calendar periods for NDR and PEEC were: 2006-2010: 3151 and 194, 2011-2015: 2633 and 271 and 2016-2020: 2080 and 590, respectively (Table 1). Diagnoses of HGD were only available in Swedish database, where 279 patients were diagnosed with HGD/EAC: NDR (41, 14.6%), PEEN (48, 17.2%) and incident HGD/EAC (190, 68,1%). Time-trends in IRs for NDR and PEEN in Sweden were: 2006-2010: 4587 and 260; 2011-2015: 3752 and 333 and 2016-2020: 3501 and 669, respectively (Table). Figure highlights the probability of EAC in the first year after index endoscopy overall and based on age, sex and year of endoscopy. Increasing age and male sex were significant predictors for NDR and PEEC.

Conclusions:
Despite improvements in endoscopic technology, this study showed a decline in NDR with a concurrent 2-3 time increase in PEEC and PEEN rates. Future studies should assess the impact of interventions (education, quality metrics and artificial intelligence) in improving NDR and reducing PEEC/PEEN in practice.

<b>Table: Incidence rates and incidence rate ratios for NDR, PEEC/PEEN and incident HGD/EAC – overall and based on the three calendar periods</b>

Table: Incidence rates and incidence rate ratios for NDR, PEEC/PEEN and incident HGD/EAC – overall and based on the three calendar periods

<b>Figure 1: Probability of PEEC in the overall cohort and based on sex, age and year of diagnosis of Barrett’s esophagus</b>

Figure 1: Probability of PEEC in the overall cohort and based on sex, age and year of diagnosis of Barrett’s esophagus

Background:
Single center studies suggest that adenoma detection rates (ADR) vary between gastroenterologists and surgeons. The generalizability of these findings is unclear. We sought to compare ADR between gastroenterologists and surgeons in the US Veterans Health Administration national healthcare system and to examine the association of patient demographic factors with ADR.

Methods:
We identified colonoscopy procedures of all indications using CPT codes from VA national electronic health records between October 2018-September 2022. We used a previously validated text recognition algorithm to determine histology from the associated pathology reports of patients aged 45-75. We classified providers from administrative codes for surgery and gastroenterology. After excluding providers with fewer than 50 colonoscopy procedures, we calculated each provider ADR as the percentage of colonoscopies with at least one adenoma or adenocarcinoma. We compared average ADR in surgeons and gastroenterologists using a 2-tailed t-test and compared the proportion of providers with ADR <30% using a chi-square test. We also calculated provider ADR in patients with different demographics (gender, race, ethnicity, geographic location and FIT+ testing within one year of colonoscopy) and used a generalized linear model to compare the ADRs before and after adjustment for patient demographics.

Results:
We identified 669,434 colonoscopies by 1,095 unique providers. Gastroenterologists (n=906; 82.7%) conducted 88.8% (n=594,710). The patient demographics are shown in the Table. ADR of surgeons (41.4%; 95% CI [39.5, 43.4]) was significantly lower than that of gastroenterologists (53.33%; 95% CI [52.7, 54.0]), p<0.0001 both before and after adjusting for differences in patient demographics. ADRs <30% were observed in 38 of 189 (20.1%) surgeons compared to 19 of 906 (2.1%) of gastroenterologists; p<0.0001. Surgeon ADRs were distributed across proportionally lower values than gastroenterology ADRs (Figure). Within both provider groups, higher ADRs were observed in men vs. women, in whites vs. non-whites, and in patients with a prior vs. no prior FIT test positive, though surgeon ADR remained consistently lower than gastroenterologist ADR. Among the colonoscopies performed after a positive FIT test, the ADR was significantly lower for surgeons (57.3%; 95% CI [53.7, 60.9]) than gastroenterologists (65.9%; 95% CI [64.8, 67.0]), p<0.0001.

Conclusion:
In this large US national healthcare system colonoscopy cohort, surgeons had a significantly lower ADR for colonoscopy of all indications and a higher proportion of endoscopists with ADRs < 30% than gastroenterologists, irrespective of patient demographics. Our findings highlight potential opportunities for targeted quality improvement and further evaluation of colonoscopy training parameters for surgical specialties.

Table. Adenoma Detection Rates of Gastroenterologists and Surgeons by Patient Demographics

Table. Adenoma Detection Rates of Gastroenterologists and Surgeons by Patient Demographics

Figure: Distribution Histograms of Gastroenterologist and Surgeon Adenoma Detection Rates.<br /> Note: Lines represent the 2-period moving averages of number of providers within each ADR range for GIs and surgeons, respectively.

Figure: Distribution Histograms of Gastroenterologist and Surgeon Adenoma Detection Rates.
Note: Lines represent the 2-period moving averages of number of providers within each ADR range for GIs and surgeons, respectively.

INTRODUCTION:
Improvements in colonoscopy quality are associated with reductions in interval colorectal cancer death. However, measurement of colonoscopy quality in practice remains challenging. We aim to describe and validate an interactive tool for artificial intelligence (AI) assessment of colonoscopy quality (AI-CQ) using recorded videos.

METHODS:
Based on quality guidelines, metrics were selected for AI development: insertion time (IT), withdrawal time (WT), retroflexion frequency, polyp detection rate (PDR), polyps per colonoscopy (PPC), and number of right colon evaluations. We also developed two novel metrics: withdrawal time excluding polypectomy time (WT-PT) and high-quality withdrawal time (HQ-WT; withdrawal time with clear colon image). The AI model was pre-trained using a self-supervised vision transformer on unlabeled colonoscopy images (n = 1x10⁷) mutually exclusive from all other datasets. The vision transformer model was finetuned for multi-label classification on another mutually exclusive colonoscopy image dataset (n = 9854) using anatomical, procedural, and pathological labels (label n = 14). During inference, colonoscopy video frame predictions were generated at a resolution of one frame per second and employed a binary threshold of ≥ 0.5 to denote presence; these predictions were subsequently used to calculate all metrics. A timeline of video predictions and metric calculations were presented to clinicians in addition to the raw video using a web-based application. The model was developed using videos at a single hospital and externally validated using 50 screening and surveillance colonoscopy videos from 6 colonoscopists at a second hospital.

RESULTS:
The interactive AI-CQ tool is presented (Figure). The cecum was reached in 48/50 cases; AI-CQ accuracy to identify cecal intubation was 88%. In 6 cases, AI-CQ did not identify the cecum due to inadequate bowel preparation obscuring landmarks (n=4) and failure to recognize landmarks (n=2).

IT (p = 0.26) and WT (p = 0.34) were similar between manual and AI-CQ measurements and significantly (p < 0.001) positively correlated (Table). On average, HQ-WT was 45.9% (IQR: 14) of, and significantly correlated with (ρ = 0.85; p < 0.001), normal WT time. Mean WT-PT was 567s, similar to mean normal colonoscopy WT (558s).

AI-CQ produced similar PDR (p = 0.66) and PPC compared to manual (p = 0.34). Rectal retroflexion was correctly identified in 95.2% of colonoscopies and the number of right colon evaluations in all colonoscopies.

DISCUSSION:
AI-CQ can be utilized to rapidly measure quality and facilitates AI-augmented review of inspection and polypectomy technique to provide actionable feedback. Further, novel inspection metrics such as WT-PT and HQ-WT, which can only be feasibly calculated by AI, may prove beneficial but require further study.

The AI-CQ is an interactive tool that automatically identifies multiple landmark events during a colonoscopy (e.g., the time the cecum is reached, when a polyp is identified, what tool is utilized to remove the polyp, etc.). This facilitates measurement of multiple colonoscopy metrics as well as allows the reviewer to rapidly watch a segment of the colonoscopy video to provide actionable feedback.

The AI-CQ is an interactive tool that automatically identifies multiple landmark events during a colonoscopy (e.g., the time the cecum is reached, when a polyp is identified, what tool is utilized to remove the polyp, etc.). This facilitates measurement of multiple colonoscopy metrics as well as allows the reviewer to rapidly watch a segment of the colonoscopy video to provide actionable feedback.

Manual and Artificial Intelligence Colonoscopy Quality (AI-CQ) timings and polyp detection. Data presented as median (IQR) unless noted. ** p < 0.001

Manual and Artificial Intelligence Colonoscopy Quality (AI-CQ) timings and polyp detection. Data presented as median (IQR) unless noted. ** p < 0.001

Amandeep Shergill

University of California San Francisco

Appearances