Application of Machine-Learning Models to Predict Tacrolimus Stable Dose in Renal Transplant Recipients

Li-Jun Zhu,Gui-Wen Feng,Mou-Ze Liu,Jie Tang,Xiang-Guang Meng,Yue-Li Zhang,Yong-Fang Hu,Li-Rong Zhang,Wei Zhang,Wen-Jun Shang,Rong Liu,Hua-Wen Xin,Ming-Jie Shao,Ying-Zi Ming

doi:10.1038/srep42192

Abstract

Tacrolimus has a narrow therapeutic window and considerable variability in clinical use. Our goal was to compare the performance of multiple linear regression (MLR) and eight machine learning techniques in pharmacogenetic algorithm-based prediction of tacrolimus stable dose (TSD) in a large Chinese cohort. A total of 1,045 renal transplant patients were recruited, 80% of which were randomly selected as the “derivation cohort” to develop dose-prediction algorithm, while the remaining 20% constituted the “validation cohort” to test the final selected algorithm. MLR, artificial neural network (ANN), regression tree (RT), multivariate adaptive regression splines (MARS), boosted regression tree (BRT), support vector regression (SVR), random forest regression (RFR), lasso regression (LAR) and Bayesian additive regression trees (BART) were applied and their performances were compared in this work. Among all the machine learning models, RT performed best in both derivation [0.71 (0.67–0.76)] and validation cohorts [0.73 (0.63–0.82)]. In addition, the ideal rate of RT was 4% higher than that of MLR. To our knowledge, this is the first study to use machine learning models to predict TSD, which will further facilitate personalized medicine in tacrolimus administration in the future.

Highlights

Tacrolimus is one of the most widely used immunosuppressive agents to prevent acute rejection following solid organ transplantation
Many factors affecting the pharmacokinetics of tacrolimus have been identified, including clinical factors such as ethnicity, age, gender, concomitant medication, hepatic and renal dysfunction, and genetic factors such as CYP3A5, CYP3A4 and ABCB1 single nucleotide polymorphisms (SNPs)[7,8]
Eight machine learning algorithms were compared with multiple linear regression (MLR) in predicting warfarin dosing, results showing Bayesian additive regression trees (BART), multivariate adaptive regression splines (MARS) and support vector regression (SVR) significantly outperformed other models; machine learning methods performed better than MLR in the low- and high- dose ranges[20]

Summary

Introduction

Tacrolimus is one of the most widely used immunosuppressive agents to prevent acute rejection following solid organ transplantation. Machine learning techniques, compared with traditional statistical models, have many advantages including high power and accuracy, ability to model non-linear effects, interpretation of large genomic data sets, robustness to parameter assumptions and dispense with normal distribution test[21]. It has been widely used in predicting warfarin dose[22]. Eight machine learning algorithms were compared with MLR in predicting warfarin dosing, results showing Bayesian additive regression trees (BART), multivariate adaptive regression splines (MARS) and SVR significantly outperformed other models; machine learning methods performed better than MLR in the low- and high- dose ranges[20]

Objectives

Methods

Results

Conclusion