
Hoda Javadikasgari
Tehran University of Medical Sciences, Iran
Title: Determining the best mathematical model for prediction of comorbidity outcome in morbid obese patients who undergo bariatric surgery
Biography
Biography: Hoda Javadikasgari
Abstract
Obesity is a chronic disease which became a critical pandemic issue in past decades. Related comorbidities are the leading cause of death in morbid obese patients. Bariatric surgery is one of the most reputed treatments for morbid obese patients but the comorbidity outcome after surgery remained unclear. In this study, we aimed to use four mathematical models to predict comorbidity outcome after bariatric surgery In this study, 224 morbid obese patients who underwent bariatric surgery were enrolled. Four mathematical models were implemented with preoperative laboratory tests, comorbidity variables, and types of surgeries. Ten fold cross validation was done and their area under receiving characteristic curve (AUC) were reported. The comorbidity outcomes were satisfactory (≥ 50% of individual comorbidities improved or resolved after 6 months) and not satisfactory. The mean age and BMI of participants were 38 ± 9.4 and 44.83 ± 6.3, respectively. Two thousands four patients (91.1%) had satisfactory outcome while 20 patients (8.9%) had not satisfactory one. After 10 fold cross validation, naïve bayes classifier, artificial neural network (ANN), logistic regression (with seven variables), and decision tree (with 33 nodes and 14 variables) had AUC of 0.58, 0.49, 0.48, and 0.38, respectively. It has been shown that naïve bayes had the highest accuracy (P < 0.05) and decision tree had the lowest accuracy (P < 0.05) while the accuracy of ANN and logistic regression were not statistically different (P = 0.48). In conclusion, naïve bayes classifier showed the best performance for predicting comorbidity outcome after bariatric surgery.