PWE-097 Machine learning can accurately classify chronic constipation patients by symptom burden using pain measures alone

IntroductionThe Patient Assessment of Constipation Symptoms (PAC-sym) is a widely validated questionnaire that numerically quantifies symptom burden in patients with constipation. PAC-sym evaluates various domains, including pain, bloating, tenesmus, rectal bleeding, straining and the quality of the bowel movement by volume or hardness of stool passage. Machine learning presents an opportunity to assist gastroenterology practice1. We investigated if machine learning could accurately classify patients with a high or low total symptom burden, using only self-reported pain severity and frequency.Methods768 patients attending with chronic constipation to a tertiary service underwent quantification of symptom measures in a prospective cohort study design. We used the PAC-sym questionnaire to quantify overall symptom burden, and patients were stratified to high or low symptom burden by a total score of >2.5 or 70% accuracy whether an individual would belong to a ‘high’ or ‘low’ total symptom burden subgroup, using pain severity and frequency alone. This adds weight to the importance of evaluating pain in patients with chronic constipation. Moreover, in a clinical setting when quantifying symptom burden in chronic constipation, the most important questions to assess is how severe and frequent a patient’s pain is.ReferencesRuffle JK, Farmer AD, Aziz Q. Artificial Intelligence Assisted Gastroenterology - Promises and Pitfalls. Am. J. Gastroenterol. 2018; ePub ahead of print.