Machine Learning-Based Algorithm for Tacrolimus Dose Optimization in Hospitalized Kidney Transplant Patients.

Area of Science:

• Nephrology
• Pharmacology
• Artificial Intelligence in Medicine

Background:

• Tacrolimus is vital for kidney transplant recipients but has a narrow therapeutic index, leading to dose variability.
• Current therapeutic drug monitoring (TDM) is empirical and may not fully capture individual pharmacokinetics.
• Machine learning (ML) offers a precision approach to optimize tacrolimus dosing by analyzing multiple patient variables.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting next-day tacrolimus concentrations in kidney transplant patients.
• To assess the performance of different ML algorithms, including ensemble methods, for tacrolimus dose optimization.
• To identify key clinical and biochemical factors influencing tacrolimus levels using model interpretability techniques.

Main Methods:

• Retrospective analysis of 1351 data points from 87 kidney transplant patients.
• Training of XGBoost, CatBoost, LightGBM, and MLP models to predict tacrolimus levels.
• Evaluation of model performance using R-squared, MAE, and RMSE; interpretability via SHAP analysis.

Main Results:

• An ensemble model achieved the highest performance (R² = 0.6297, MAE = 1.0181, RMSE = 1.2999).
• The MLP model demonstrated superior predictive power among individual models, highlighting nonlinear pharmacokinetic interactions.
• Key predictors included prior tacrolimus levels, cumulative dose, renal function markers (eGFR, creatinine), and albumin.

Conclusions:

• A robust ML-based algorithm can optimize tacrolimus dosing in kidney transplant recipients.
• Improved prediction of tacrolimus concentrations can reduce inter-patient variability and nephrotoxicity risk.
• This AI-driven approach advances precision medicine for safer, more individualized immunosuppressive management in transplantation.