Deep learning algorithms in predicting severe complications after bariatric surgery

Accurate models to predict severe postoperative complications could be of value in the preoperative assessment of potential candidates for bariatric surgery. Traditional statistical methods and machine learning (ML) methods have so far failed to produce high accuracy. To find a useful algorithm to predict the risk for severe complication after bariatric surgery, we will train and compare 3 deep learning (DL) algorithms, i.e. multilayer perceptron, convolutional neural network, and recurrent neural network, using the information from 37,811 patients operated with a bariatric who underwent bariatric surgery in Sweden between 2010 and 2014. The algorithms will be then tested on 6,250 patients operated in 2015. In the context that only 3% of patients experienced severe complications, most ML algorithms may show high accuracy (>90%) and specificity (>0.9) in both the training and test data. However, it is difficult to achieve an acceptable sensitivity in the test data. Therefore, we will use the synthetic minority oversampling technique to augment the positive outcomes. We will also try to tune the hyperparemeters of the DL algorithms to maximize sensitivity. We aim to find perceptible improvement in the DL methods better than what we found in a previous study.