Decision tree-based prediction of 5-year weight trajectories after bariatric surgery in adolescents and young adults: a retrospective cohort study from France and Sweden.

BACKGROUND: Metabolic-bariatric surgery is an efficient therapy in selected adolescents with severe obesity. However, predicting the postoperative weight loss is challenging. A machine learning calculator predicting 5-year weight loss trajectory has been developed in adults, based on seven preoperative features. The aim of the present study was to adapt and test it in adolescents. METHODS: Retrospective cohort study in patients aged 12-20 years undergoing Roux-en-Y gastric bypass (RYGB), sleeve gastrectomy (SG), or adjustable gastric band (AGB) in France and Sweden between 2001 and 2022. Primary outcome was the accuracy of 5-year BMI prediction, expressed as the median absolute deviation (MAD) between predicted and observed BMI. The model developed in adults, was trained in a subset of 80% of randomly selected adolescents, and secondly tested in the remaining 20%. RESULTS: We enrolled a total of 2255 patients (1705 female [75.6%], 12-20 years [median 19]). Five-year follow-up data were available for 59% of French and 38% of Swedish patients. The median (IQR) 5-year total weight loss was 30.2% (23.9-38.6) for RYGB, 23.4% (13.7-32.8) for SG, and 13.4% (0.0-30.1) for AGB. The adapted model predicted the observed 5-year BMI with a MAD of 3.7 kg/m² (95% CI [3.3-3.9]). The accuracy of the model was maximal for bypass (3.2 kg/m² [3.0-3.7]), good for SG (3.9 kg/m² [3.1-5.0]), and lower for AGB (7.3 kg/m² [5.5-8.4]), and accuracy decreased with time and in adolescents under 19 years. Age, height, weight, and type of intervention influenced 5-year weight loss. Type 2 diabetes influenced weight loss until 2 years after surgery, but not later. CONCLUSION: The model had an acceptable accuracy for adolescents to predict 5-year postoperative weight loss trajectory. Accuracy decreased over time and was influenced by type of intervention and age. This calculator is available online: https://bariatric-weight-trajectory-prediction.univ-lille.fr/ .