Warren Alpert Medical School of Brown University Providence, RI, United States
Waihong Chung, MD, PhD, Firrah Saeed, MD, Abbas Rupawala, MD Warren Alpert Medical School of Brown University, Providence, RI
Introduction: Current guidelines on the management of diverticulitis recommend the use of colonoscopy to exclude the misdiagnosis of colonic neoplasm in “appropriate candidates”. Despite the low quality of evidence supporting this recommendation, recent cohort studies continue to see colonoscopy being performed in nearly 50% of patients following an episode of acute diverticulitis.
Methods: We conducted a retrospective cohort analysis of 292 patients who underwent colonoscopy for the sole purpose of “follow-up after acute diverticulitis” between 1/1/2014 to 12/31/2019 within our hospital system. Colonoscopic findings were compared against those obtained from a control cohort of 5,840 patients, propensity score matched based on gender and comorbidities, who underwent colonoscopy for “average-risk screening” during the same timeframe.
Results: Colonic polyps of any histological subtypes were less likely to be detected on colonoscopy for the study cohort compared to the control (34.93% vs. 56.11%, p< 0.01). Advanced adenoma (adenoma >10mm in size, villous adenoma, or high grade dysplasia), in particular, was only detected in 1.03% of colonoscopies (compared to 9.30% in the control, p< 0.01), while non-advanced adenoma was detected in 21.92% of cases. No cases of colorectal cancer were identified within the study cohort.
Age >70 years (odds ratio [OR]=2.11, p=0.04) and male gender (OR=1.91, p=0.02) were found on regression analysis to be independent predictors for the detection of advanced or non-advanced adenoma on colonoscopy in the study cohort. In contrast, having a prior colonoscopy for any indication within the previous 18 months was found on time-to-event analysis to be a strong negative predictor (OR=0.14, p< 0.01) for the detection of any colonic polyps.
Finally, we constructed, using the random forest algorithm, a machine learning model for predicting the diagnostic yield for colonic adenoma on colonoscopy in patients with diverticulitis. Our model is able achieve an accuracy of 82% and an F1 score of 0.89.
Discussion: Colonoscopies performed after an episode of acute diverticulitis were associated with a poor diagnostic yield for clinically significant adenoma or colonic neoplasm. It should be deferred until a screening colonoscopy is indicated in average-risk patients, especially younger female patients, and should be avoided in patients who had a prior colonoscopy within the past 18 months. A machine learning model can be implemented to guide clinical decision-making.
Disclosures: Waihong Chung indicated no relevant financial relationships. Firrah Saeed indicated no relevant financial relationships. Abbas Rupawala indicated no relevant financial relationships.
Waihong Chung, MD, PhD, Firrah Saeed, MD, Abbas Rupawala, MD. P1459 - A Machine Learning Model to Predict the Diagnostic Yield of Follow-up Colonoscopy for Colonic Adenoma After an Episode of Acute Diverticulitis, ACG 2021 Annual Scientific Meeting Abstracts. Las Vegas, Nevada: American College of Gastroenterology.