DiaBar: Predicting type 2 diabetes remission post-metabolic surgery utilizing mRNA expression profiles from subcutaneous adipose tissue

Background: Subcutaneous adipose tissue (SAT) is a metabolic organ, which is involved in the pathogenesis of type 2 diabetes (T2D). Methods to predict diabetes remission after metabolic surgery exist, however their prediction accuracy still needs improvement. We hypothesized, that gene expression profiles in the SAT could predict diabetes remission after metabolic surgery more accurately than any current methods. Methods: . The American Diabetes Association criteria were used to define partial and complete remission. Univariate generalized linear models, tree decision algorithms (Exhausted Chaid, CART and Quinlan's C5 with adaptive boosting) and, multilayer perceptron networks were used to develop classifiers for patients with no, partial or complete remission (DiaBar). Results: In this study 106 patients were included, 66 (62.3%) patients had T2D the remaining 40 (37.7%) were patients with prediabetes. Complete and partial remission were achieved by 69 (65.1%) and 20 (18.9%) patients respectively. Using a multilayer perceptron, we achieved an overall accuracy of 98.0% (remission: no 100%; partial 90.0%; complete 100%). The validated DiaRem Score was used as the comparative score, which had an overall accuracy for classifying patients with complete, partial or no remission of 74.7%. Conclusions: Using gene expression profiles from the SAT, we developed the DiaBar test, which accurately predicts diabetes remission after metabolic surgery and seems to be superior to the DiaRem score.