Machine learning analysis for bowel urgency in ulcerative colitis patients from IBD Tool web application

This thesis focuses on the analysis of bowel urgency, a common and highly problematic symptom associated with inflammatory bowel diseases (IBD). Bowel urgency is characterized by a sudden need to have a bowel movement, which can range in severity from a mild hurry to complete incontinence. This symptom significantly impacts the quality of life and daily activities of individuals suffering from IBD. The IBD Tool, a telemedicine web application platform developed in collaboration between LINKS and the Mauriziano Hospital in Turin, aims to assist in remotely monitoring patients diagnosed with IBD. The core functionality of the IBD Tool is to deliver a set of questionnaires, designed by the international medical community, to gather information about the impact and symptoms of IBD. This thesis specifically focuses on patients with ulcerative colitis (UC), one of the most common types of IBD alongside Crohn's disease. Ulcerative colitis is characterized by inflammation confined to the colon and rectum, affecting only the innermost lining of the intestinal wall. The objective of the thesis is to implement prediction models to estimate the degree of bowel urgency in UC patients. The degree of bowel urgency is obtained from the PSCCAI questionnaire, which tracks disease activity. The questionnaire includes specific questions that address bowel urgency, asking patients about the urgency they feel when they need to go to the toilet. The IBD Tool stores all information in a centralized MongoDB database, a document-oriented database management system. Among the various collections of data stored, two are particularly important for this thesis: one that stores all the information from the filled questionnaires (e.g., scores, answers to each question, and timestamps), and another that stores clinical data (e.g., disease extent, weight, height, pathology duration, ongoing therapies). The thesis employs classifiers such as logistic regression and random forest to predict bowel urgency. These methods are implemented using different strategies, as explained in the Results section, to address issues like class imbalance and similarities between classes. Since the questionnaires are filled out by patients, there is a possibility of improper responses that may not align with medical reality and may include subjective thoughts. During this research, significant relationships between questionnaire responses and medical therapies were identified through statistical methods. Predicting bowel urgency can be a valuable feature for future implementations in the IBD Tool and other telemedicine solutions aimed at providing care for IBD patients, particularly those with ulcerative colitis. Telemedicine is a rapidly growing field that benefits all parties involved. It helps reduce unnecessary medical visits and facilitates communication between patients and physicians, thereby enhancing the overall medical infrastructure.