An artificial intelligence-derived tool proposal to ease disordered eating screening in people with obesity.

PURPOSE: In people with obesity, food addiction (FA) tends to be associated with poorer outcomes. Its diagnosis can be challenging in primary care. Based on the SCOFF example, we aim to determine whether a quicker and simpler screening tool for FA in people with obesity could be developed, using artificial intelligence (machine learning). METHODS: The first step was to look for the most discriminating items, among 152 different ones, to differentiate between FA-positive and FA-negative populations of patients with obesity. Items were ranked using the Fast Correlation-Based Filter (FCBF). Retained items were used to test the performance of nine different predictive algorithms. Then, the construction of a graphic tool was proposed. RESULTS: Data were available for 176 patients. Only three items had a FCBF score > 0.1: "I eat to forget my problems"; "I eat more when I'm alone"; and "I eat sweets or comfort foods". Naive Bayes classification obtained best predictive performance. Then, we created a 3-item nomogram to predict a positive scoring on the YFAS. CONCLUSION: A simple and fast screening tool for detecting high-disordered eating risk is proposed. The next step will be a validation study of the FAST nomogram to ensure its relevance for emotional eating diagnosis. LEVEL OF EVIDENCE: Level V, cross-sectional descriptive study. CLINICAL TRIAL REGISTRY NUMBER: NCT02857179 at clinicaltrials.gov.