Prediction of very early subclinical rejection with machine learning in kidney transplantation

Abstract

Abstract Protocol biopsy is a reliable method for assessing allografts status after kidney transplantation (KT). However, due to the risk of complications, it is necessary to establish indications and selectively perform protocol biopsies by classifying the high-risk group for early subclinical rejection (SCR). Therefore, the purpose of this study is to analyze the incidence and risk factors of early SCR (within 2 weeks) and develop a prediction model using machine learning. Patients who underwent KT at Samsung Medical Center from January 2005 to December 2020 were investigated. The incidence of SCR was investigated and risk factors were analyzed. For the development of prediction model, machine learning methods (random forest, elastic net, extreme gradient boosting [XGB]) and logistic regression were used and the performance between the models was evaluated. The cohorts of 987 patients were reviewed and analyzed. The incidence of SCR was 14.6%. Borderline cellular rejection (BCR) was the most common type of rejection, accounting for 61.8% of cases. In the analysis of risk factors, recipient age (OR 0.98, p = 0.03), donor BMI (OR 1.07, p = 0.02), ABO incompatibility (OR 0.15, p < 0.001), HLA II mismatch (two [OR 6.44, p < 0.001]), and ATG induction (OR 0.41, p < 0.001) were associated with SCR in the multivariate analysis. The logistic regression prediction model (average AUC = 0.717) and the elastic net model (average AUC = 0.712) demonstrated good performance. HLA II mismatch and induction type were consistently identified as important variables in all models. The odds ratio analysis of the logistic prediction model revealed that HLA II mismatch (OR 6.77) was a risk factor for SCR, while ATG induction (OR 0.37) was a favorable factor. Early SCR was associated with HLA II mismatches and induction agent and prediction model using machine learning demonstrates the potential to predict SCR.