Pancreas Rejection in the Artificial Intelligence Era – The Poor Men’s Tool

Abstract

Introduction: Pancreas transplantation is currently the only treatment able to reestablish normal endocrine pancreatic function. Despite, all efforts pancreas allograft survival and rejection remain a major clinical problem. The purpose of this study was the identification of features, which could signal patients at risk of pancreas allograft rejection. Methods: We collected 76 features from 79 simultaneous kidney-Pancreas transplanted patients (SKP) and used a fast and simple probabilistic classifier based on Bayes’ theorem (Naïve Bayesian Classifier), to build the predictive models. We used the area under the receiver operating characteristic curve (AUROC) to evaluate the predictive performance via leave-one-out cross validation. Results: Rejection events were identified in 13 SKP (16.5%), and significant differences were identified in 11 features (p<0.05) namely, [Type of Dialysis; Warm ischemia (minutes); de novo DSA; SUM pre-transplant DSA MFI ( maximum); vPRA Pre-Tx(%); SUM pre transplant DSA MFI ( latest assay); Donors Age; Internment in the ICU (days); Diabetes mellitus Previous Treatment Long Term Insulin (U/I/day); Patients Height; pancreas donor risk index (pDRI) and Dialysis time (days)]. The results showed that the Naïve Bayes classifier prediction performed very well with a AUROC 0.96% and a classification accuracy (CA) of 0.86% (Table1). Based on this naïve Bayesian model for the prediction with was possible to develop a nomogram (figure1). Conclusion: Our results indicated that it is feasible to develop a successful Bayesian classifier for prediction of graft rejection. The generated nomogram can used for probability prediction thus supporting clinical decision.Tabled 1ModelAUCCAF1PrecisionRecallNaive Bayes0.960.860.870.910.86 Open table in a new tab