Improving pre-bariatric surgery diagnosis of hiatal hernia using machine learning models

Background: Bariatric patients have a high prevalence of hiatal hernia (HH). HH imposes various difficulties in performing laparoscopic bariatric surgery. Preoperative evaluation is generally inaccurate, establishing the need for better preoperative assessment. Objective: To utilize machine learning ability to improve preoperative diagnosis of HH. Methods: Machine learning (ML) prediction models were utilized to predict preoperative HH diagnosis using data from a prospectively maintained database of bariatric procedures performed in a high-volume bariatric surgical center between 2012 and 2015. We utilized three optional ML models to improve preoperative contrast swallow study (SS) prediction, automatic feature selection was performed using patients’ features. The prediction efficacy of the models was compared to SS. Results: During the study period, 2482 patients underwent bariatric surgery. All underwent preoperative SS, considered the baseline diagnostic modality, which identified 236 (9.5%) patients with presumed HH. Achieving 38.5% sensitivity and 92.9% specificity. ML models increased sensitivity up to 60.2%, creating three optional models utilizing data and patient selection process for this purpose. Conclusion: Implementing machine learning derived prediction models enabled an increase of up to 1.5 times of the baseline diagnostic sensitivity. By harnessing this ability, we can improve traditional medical diagnosis, increasing the sensitivity of preoperative diagnostic workout.