Remission of type 2 diabetes after gastrectomy for gastric cancer: diabetes prediction score.

Abstract

Although type 2 diabetes (T2D) remission after gastric cancer surgery has been reported, little is known about the predictors of postoperative T2D remission. This study used data from a nationwide cohort provided by the National Health Insurance Service in Korea. We developed a diabetes prediction (DP) score, which predicted postoperative T2D remissions using a logistic regression model based on preoperative variables. We applied machine-learning algorithms [random forest, XGboost, and least absolute shrinkage and selection operator (LASSO) regression] and compared their predictive performances with those of the DP score. The DP score comprised five parameters: baseline body mass index (< 25 or ≥ 25kg/m2), surgical procedures (subtotal or total gastrectomy), age (< 65 or ≥ 65years), fasting plasma glucose levels (≤ 130 or > 130mg/dL), and antidiabetic medications (combination therapy including sulfonylureas, combination therapy not including sulfonylureas, single sulfonylurea, or single non-sulfonylurea]). The DP score showed a clinically useful predictive performance for T2D remission at 3years after surgery [training cohort: area under the receiver operating characteristics (AUROC) 0.73, 95% confidence interval (CI), 0.71-0.75; validation cohort: AUROC 0.72, 95% CI 0.69-0.75], which was comparable to that of the machine-learning models (random forest: AUROC 0.71, 95% CI 0.68-0.74; XGboost: AUROC 0.70, 95% CI 0.67-0.73; LASSO regression: AUROC 0.75, 95% CI 0.73-0.78 in the validation cohort). It also predicted the T2D remission at 6 and 9years after surgery. The DP score is a useful scoring system for predicting T2D remission after gastric cancer surgery.