A Predictive Model for the Diagnostic and Therapeutic Yield of Colonoscopy Performed for Lower Gastrointestinal Bleeding.

Abstract

Although colonoscopy is routinely performed for the management of lower gastrointestinal bleeding (LGIB), the quality of evidence supporting its use is poor and its yield for active bleeding or malignancy is low in practice. We conducted a retrospective analysis of all adult patients who underwent colonoscopies for LGIB at our hospital system between January 1, 2015, and December 31, 2019. A statistical model was built on a cohort of 5195 cases using multiple logistic regressions to predict the detection of various colonoscopy findings and the use of different colonoscopy maneuvers. The model was converted into a risk scoring system, named the TYPICAL Index, and was validated against a separate cohort of 914 cases. Active bleeding was only seen in 3.8% of colonoscopies performed for LGIB and endoscopic hemostasis was applied in 43.7% of actively bleeding lesions. Malignant-appearing lesions were detected in 2.5% of LGIB procedures and all cases of proximal colonic mass were identified in subjects age above 60. The TYPICAL Index, derived from age, gender, hemoglobin, creatinine, international normalized ratio, and prior colonoscopy for LGIB has a concordance statistic of 0.71 and a negative predictive value of 93.8% for potential bleeding sources requiring hemostasis, malignant-appearing lesions, or active bleeding on colonoscopy when applied using a threshold of 6. Active bleeding and malignancies are rarely encountered and endoscopic hemostasis is seldom required during colonoscopy performed for LGIB. We developed and validated a risk scoring system to identify cases of low predicted diagnostic and therapeutic yield and to guide clinical decision-making.