Table 3.

Overview of papers about postoperative management and complications included in our analysis

Author, years	Study design	Objective	Final cohort	Outcomes	Type of ML	Prediction performance
Sheikhtaheri A 2019	Retrospective, multicenter	Predicting the early complications of one-anastomosis gastric bypass	1509	Complications incidence	ANNs	Accuracy, specificity, sensitivity: 10-day prediction system 98.4%, 98.6%, 98.3%;1-month system 96%, 93%, and 98.4%; 3-month system 89.3%, 86.6%, 91.5%
Cao Y 2019	Retrospective, multicenter	Predicting the risk for severe complication after BS	37811	Complications incidence	LR, LDA, QDA, TR, KNN, SVM, MLP, NN, AdaBoost LR, bagging LDA, bagging QDA, RF, extremely randomized trees, AdaBoost Extra trees, gradient RT, AdaBoost Gradient trees, bagging KNN, AdaBoost SVM, bagging MLP	Best gradient RT and bagging MLP AUC 0.58
Cao Y 2020	Retrospective, multicenter	Exploring whether serious postoperative complications of bariatric surgery recorded in a national quality registry can be predicted preoperatively using deep learning methods	44061	Complications incidence	MLP, CNN, RNN	AUC ≤ 0.6
Nudel J 2021	Retrospective, multicenter	Predicting leak and VTE after BS	436807	Leak and VTE incidence	ANN, XGBs	ANN AUC 0.75; XGBs AUC 0.70
Wise ES 2020	Retrospective, multicenter	To optimize the prediction of the composite endpoint of 30-day readmission, reoperation, reintervention, or mortality, after laparoscopic sleeve gastrectomy	101721	30-day morbidity and mortality prediction after bariatric surgery	LR and ANN	ANN AUROC = 0.581 compared to LR AUROC = 0.572 in the training set
Razzaghi T 2019	Retrospective, multicenter	To identify risks/outcomes associated with BS	11636	Risk-prediction	NB, Radial Basis Function Neural Network, k-NN, SVM, and LR	The combination of a suitable feature selection method with ensemble learning methods equipped with Oversampling (SMOTE) method can achieve higher performance metrics
Cruz MR 2014	Retrospective, single center	To validate a computerized intelligent decision support system that suggests nutritional diagnoses of patients submitted to BS	60	Nutritional monitoring of patients undergoing BS	Bayesian network	The system sensibility and specificity were 95.0%
Liew PL 2007	Retrospective, single center	To compare the predictive accuracy of LR and ANN with respect to the clinicopathologic features of gallbladder disease among obese patients	117	Prediction of gallbladder disease	LR and ANN	The average correct classification rate of ANNs was higher than that of the traditional logistic regression approach (97.14% versus 88.2%). Besides, ANNs also had a lower Type II error when compared with logistic regression

LR logistic regression, TR decision tree, RF random forest, Gbdt gradient boosting decision tree, Xgbc extreme gradient boosting, Gbm light GBM, MV majority vote, NB Naive Bayes, k-NN k-nearest neighbor, Ct classification tree, SVM support vector machine, AdaBoost-SVM adaptive boosting SVM, ML machine learning, RT regression tree, SVR support vector regression, AdaBoost-SVR adaptive boosting SVR, Sr RDI sunrise system-derived respiratory disturbance index, OSA obstructive sleep apnea, GOLD Global Initiative for Chronic Obstructive Lung Disease, NN neural network, NB Naïve Bayes, ROC receiver operating characteristics, PR precision recall, ANNs artificial neural networks, LDA linear discriminant analysis, QDA quadratic discriminant analysis, MLP multilayer perceptron, AdaBoost LR adaptive boosting LR, CNN convolutional nearal network, RNN recurrent neural network, XGBs gradient boosting machines, BS bariatric surgery, VTE venous thromboembolism