A machine learning approach to predict types of bariatric surgery using the patients first physical exam information

Background and aim: According to the IFSO worldwide survey report in 2014, 579517 bariatric operations have been performed in a year, of which nearly half the procedures were SG followed by RYGB. This procedure is a proven successful treatment of patients with morbid obesity which induces considerable weight loss and improvement of type 2 diabetes mellitus, insulin resistance, inflammation, and vascular function. In the present study, we aimed to build a machine based on a decision tree to mimics the surgeon s pathway to select the type of bariatric surgery for patients. Methods: We used patient’s data from the National Bariatric Surgery Registry between March 2009 and October 2020. A decision tree was constructed to predict the type of surgery. The validation of the decision tree confirmed using 4-folds cross-validation. Results: We rich a decision tree with a depth of 5 that is able to classify 77% of patients into correct surgery groups. In addition, using this model we are able to predict 99% of bypass cases (Sensitivity) correctly. The waist circumference less than 126 cm and BMI equal to or more than 43 kg/m2, age equal to or greater than 30 years old, and being hypertensive or diabetes are the most important separators. Conclusion: The effects of all nodes have been studied before and the references confirmed the relations of them and surgery type.