Are the current colonoscopy recommendations for interval surveillance in patients with polyps enough? Machine learning-augmented propensity score cohort analysis of 1840 patients

Colonoscopy remains the gold standard for screening and surveillance of colorectal neoplasms, and is associated with a lower risk of colorectal cancer (CRC)-related mortality. The current interval surveillance recommendations in patients with previous adenomas lack sufficient evidence. The prevalence of subsequent adenomas, and especially high-risk adenomas, during surveillance is not well known.

The primary outcome of this study was to determine the prevalence of polyps upon surveillance colonoscopy in patients who have a history of adenomas on initial average-risk-screening colonoscopy, but then have a normal initial surveillance (second) colonoscopy between 2003 and 2017. This is the first known retrospective cohort study of adenoma detection rate (ADR) with sub-group analysis of patients with serial surveillance colonoscopies by abnormal and high-risk surveillance findings separately by prior abnormal colonoscopies and correct surveillance strategies based on the recent March 2020 updated guidelines. After ADR calculation, machine learning-augmented propensity score adjusted multivariable regression with augmented inverse-probability weighting propensity (AIPW) score analysis was used to assess the relationship between guideline adherence, as well as abnormal and high-risk surveillance findings.

A total of 1840 patients with pathologically confirmed adenomas or cancer on an initial average-risk-screening (first) colonoscopy met study criteria. 837 (45.5%) had confirmed adenomas on second colonoscopy, and 1003 (54.5%) had normal findings. Of 837 patients with polyps on both first and second colonoscopy, 423 (50.5%) had adenomas on third colonoscopy. Of the 1003 patients without polyps on second colonoscopy, 406 (40.5%) had confirmed adenomas on third colonoscopy. Guideline adherence was low at 9.18%, though was associated in propensity score adjusted multivariable regression with increased odds of an abnormal third (but not high-risk) colonoscopy, with comparable AIPW results.

This 14-year study demonstrates the ADR to be > 40% on the third colonoscopy for patients with adenomas on initial screening colonoscopy, who then have a normal second colonoscopy. Through advanced machine learning and propensity score analysis, we showed that correct adherence is associated with higher odds of abnormal, but not high-risk abnormal 3rd colonoscopy, with evidence that high-risk surveillance findings are reduced by providers shortening the time between surveillance colonoscopies in contrast to the guidelines for those for whom there is presumed greater clinical suspicion of eventual cancer. Larger prospective trials are needed to guide optimal surveillance for these patients.