Modeling for prediction of gastric cancer after bariatric surgery: a secondary and multi-method study

Although bariatric surgery is necessary for some obese patients, it is controversial whether it increases the risk of gastric cancer and what the risk factors are. We performed this secondary study to identify the risk factors and compare the possible scenarios. A multi-method study was performed consisted of four phases; (1) meta-analysis of case reports, (2) naïve Bayes modeling, (3) penalized logistic regression, (4) models’ comparison. A total of 61 gastric cancer patients after bariatric surgery were collected from the literature, and 3946 cases of bariatric surgery without cancer from our local cohort were analyzed. In phase one, male sex, age more than 40 years at the time of surgery, and roux-en-Y gastric bypass (RYGB) were identified as potential risk factors. In phase two, naïve Bayes modeling showed the probability of gastric cancer in the best and worst-case scenarios as 0.003 and 0.064, respectively (regarding global base rate of general population). In phase three, Firth logistic regression showed the probability of gastric cancer in the best and worst-case scenarios as 0.003 and 0.095, respectively. In phase four, no significant difference was observed between the two models of naïve Bayes and Firth logistic regression (χ2 = 0.026, DF = 4, P > 0.999). Male sex, age more than 40 years at time of surgery, and RYGB surgery were the risk factors of gastric cancer after bariatric surgery. Planning to use the low-risk scenarios as much as possible may be helpful regarding patients’ medical indications.