Colorectal Surveillance Interval Assignment Based on In-Vivo Prediction of Polyp Histology: Impact of Endoscopic Quality Improvement Program

Abstract

In-vivo predictions of polyp histology, without pathological confirmation, (“resect and discard” or “diagnose and leave behind” strategies) has been proposed as a strategy to reduce health care costs associated with CRC prevention. Recent societal guidelines mandate a high level of accuracy for in vivo triage compared to standard pathologic prediction of post-polypectomy surveillance intervals. We hypothesized that an endoscopist training intervention would result in improvement in post-polypectomy surveillance prediction. As part of a larger prospective randomized study on endoscopist training in colonoscopy, endoscopists were asked to predict the histopathology of all resected polyps during routine colonoscopy and to assign colon cancer surveillance intervals using consensus guidelines. Data were prospectively collected for consecutive outpatient colonoscopies between 8/2 and 04/11/2011, including procedure indication, demographics and risk factors, polyp description, endoscopist prediction of histopathology and actual histopathology. Procedures were excluded if incomplete, had poor bowel prep (BBPS <5), or had an indication of post-cancer surveillance, inflammatory bowel disease, hereditary syndrome surveillance, newly found cancer or active gastrointestinal hemorrhage. Polyps ≥10mm were considered “neoplastic” for the purposes of surveillance interval assignment. Comparisons of surveillance interval prediction accuracies were made for each group of study endoscopists (trained and not trained) before (Phase I) and after (Phase II) the training intervention period. Of the 2400 colonoscopies, 1231 patients were found to have polyps. In the group of endoscopists who did not receive training, “optically predicted” surveillance intervals based solely on number of predicted adenomas and size of the largest polyp (independent of risk factors) had an overall accuracy of 81% (236/290) in Phase I and 79% in Phase II (265/337). The group that received training had an overall accuracy of 84% (242/287) in Phase I (prior to training) and 82% (261/317) in Phase II (after training). There was little evidence to suggest that training had a direct, detectable impact on the accuracy of optically predicted surveillance intervals (OR: 1.20, 95% CI: 0.76-1.89, P=0.44). When individual patient risk factors were considered the results were similar. Baseline accuracy for optical predication of surveillance intervals is modest. Endoscopist training did not result in improvement of “optically-predicted” surveillance interval accuracy. These findings highlight the likely need for further studies on methods to train endoscopists to accurately predict polyp histology.