Easy-to-use AI Calculator Predicting 5 y-Weight Trajectories After Bariatric Surgery: A Sophia Study

This Medfyle was published more than two years ago. More recent Medfyle on this topic may now be available.

Key messages

• These authors used artificial intelligence to develop, and validate in multiple international cohorts, an easy-to-use, interpretable, and generalizable model that provides personalized prediction of 5 year-weight loss trajectories (WLT) after bariatric-metabolic surgery.
• The study enrolled 9,807 participants with up to 5-year follow-up after Roux-en-Y gastric bypass (RYGB), sleeve gastrectomy (SG) or gastric band (GB), from six prospective cohort studies in France (A / n=1,498 / RYGB-SG-GB; B / n=348 / SG; C / n=237 / RYGB-SG), Netherlands (D / n=5,888 / RYGB-SG-GB), Sweden (E / n=642 / RYGB-GB), Singapore (F / n=977 / SG-RYGB), and one randomized controlled trial in Switzerland (G / n=217 / RYGB-SG); a machine learning model was derived on A, trained in B, externally validated on CDEF, and tested in G.
• Among >1000 attributes available at baseline in cohort A, 7 variables (height, weight, intervention type, age diabetes status, diabetes duration, and smoking status) were selected based on LASSO algorithm, and used to build interpretable and time dependent regression trees predicting WLT during 5 years after bariatric-metabolic surgery.
• In cohorts CDEF, the resulting model predicted postoperative body mass index, across all interventions, with an average RMSE of 3.6, 4.2, and 4.6 kg/m², at 1, 2, and 5 years respectively, and the model also outperformed previously published models in the same cohorts; when applied to each participant in group G, the model confirmed the difference in WLT observed between RYFB and SG during the study.
• These authors developed and validated an easy to use and interpretable model predicting individual 5 year-WLT after the three most common bariatric-metabolic interventions; the companion tool based on this model can be used at the point of care by patients and health care providers to inform clinical decisions (https://bariatric-weight-trajectory-prediction.univ-lille.fr).