Numerical modeling and prediction of late estimated glomerular filtration rate in kidney transplant recipients based on machine learning models and the Monte Carlo simulation method

Overview

This study presents a numerical modeling approach for predicting late estimated glomerular filtration rate (eGFR) in kidney transplant recipients using machine learning classifiers combined with Monte Carlo simulation methodology. The research applies multiple predictive models trained on clinical data from a tacrolimus-based immunosuppressive protocol, incorporating physiological parameters and tacrolimus pharmacokinetics as predictive features.

Methods and approach

A multivariate regression model was constructed based on clinical study data, utilizing independent variables including serum creatinine, six-month post-transplantation eGFR, dose-adjusted trough concentration of tacrolimus (C0/D), and IPV. The eGFR range of 13-36 served as the dependent variable. Monte Carlo simulation methodology was applied to obtain essential data for prediction model optimization. Three machine learning classifiers were trained and evaluated: DecisionTreeClassifier, Random Forest Classifier, and XGBClassifier. Comparative analysis determined the optimal model across accuracy, reliability, and generalizability metrics.

Results

XGBoost classifier demonstrated superior performance relative to the alternative classifiers tested, exhibiting the highest accuracy, reliability, and generalizability. Monte Carlo simulation was identified as a methodological advancement applicable to kidney transplantation modeling. The integration of these approaches produced prediction models capable of analyzing stochastic systems relevant to transplant population outcomes.

Implications

The combination of machine learning classification with Monte Carlo simulation represents a methodological development for quantitative analysis of kidney transplant outcomes. The superior performance of the XGBoost classifier suggests potential utility for modeling complex interactions among physiological and pharmacokinetic parameters in transplant recipients. These numerical methods may contribute to optimization strategies within current immunosuppressive protocols, though further validation in diverse clinical populations would be necessary to establish broader applicability.