Search

Abstract Background Given its narrow treatment window, high toxicity, adverse effects, and individual differences in its use, we collected and sorted data on tacrolimus use by real patients with kidney diseases. We then used machine learning technology to predict tacrolimus blood concentration in order to provide a basis for tacrolimus dose adjustment and ensure patient safety. Methods This study involved 913 hospitalized patients with nephrotic syndrome and membranous nephropathy treated with tacrolimus. We evaluated data related to patient demographics, laboratory tests, and combined medication. After data cleaning and feature engineering, six machine learning models were constructed, and the predictive performance of each model was evaluated via external verification. Results The XGBoost model outperformed other investigated models, with a prediction accuracy of 73.33%, F-beta of 91.24%, and AUC of 0.5531. Conclusions Through this exploratory study, we could determine the ability of machine learning to predict TAC blood concentration. Although the results prove the predictive potential of machine learning to some extent, in-depth research is still needed to resolve the XGBoost model’s bias towards positive class and thereby facilitate its use in real-world settings.