Prediction of High-Dose Methotrexate Blood Concentration in Osteosarcoma Patients Using Machine Learning

Guardado en:

Detalles Bibliográficos
Publicado en:	Drug Design, Development and Therapy vol. 19 (2025), p. 3631
Autor principal:	Zhao, J
Otros Autores:	Dai, S, He J, Liu, N, Zhang, B, Li S
Publicado:	Taylor & Francis Ltd.
Materias:	Erythrocytes Datasets Metastasis Bone cancer Regression analysis Telemedicine Machine learning Toxicity Prediction models Methotrexate Decision trees Pharmacokinetics Medical prognosis Creatinine Dosage Hospitals Platelets Missing data Algorithms Chemotherapy Patients Dihydrofolate reductase Regression models Normal distribution Data processing Oncology Blood levels Health services Monitoring Demography L-Lactate dehydrogenase Hydration Customization Learning algorithms Drug dosages Overdose Coefficient of variation Osteosarcoma Lactate dehydrogenase Variables Clinical decision making
Acceso en línea:	Citation/Abstract Full Text + Graphics Full Text - PDF
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Descripción
Resumen:	Introduction: High-dose methotrexate is a typical chemotherapy that is widely used in the treatment of osteosarcoma. However, the unique dose-response relationship of methotrexate makes its treatment window relatively narrow, and its clinical use is in a dilemma: either the drug concentration in the patient’s body cannot reach the effective concentration level, or adverse reactions may occur due to drug overdose. For this circumstance, monitoring and predicting the drug concentration in the patient’s body is well founded and necessary. While pharmacokinetic models exist, they often oversimplify patient-specific covariates. This study addresses the unmet need for early-exposure prediction through interpretable machine learning, enabling data-driven decisions before toxicity manifestation.Methods: In this article, 68 osteosarcoma patients’ information including demography, administration and assay was gathered. We analyzed medical data and selected 10 important features using a random forest, including hydration status, red blood cell distribution width coefficient of variation, platelet distribution width, creatinine, γ-glutamyl transferase, large platelet ratio, serum potassium, lactate dehydrogenase, weight, and prealbumin. Then, cross-validation and SHAP has been conducted to confirm the robust and interpretation of the model.Results: On this basis, 7 machine learning regression models was built to predict the blood concentration of methotrexate. R2, MSE, RMSE, MAE are the evaluation metrics. Finally, LightGBM was selected as the best prediction model with a performance of R2=0.87, MSE=0.020, RMSE=0.141, MAE=0.065.Discussion: This machine learning framework addresses a critical gap in high-dose methotrexate therapeutic monitoring by achieving early and personalized blood drug concentration prediction, allowing for personalized dosing of patients based on predicted concentrations. The interpretability of SHAP-derived feature importance enhances clinical utility, offering a paradigm shift from reactive toxicity management to proactive precision dosing in osteosarcoma therapy.
ISSN:	1177-8881
DOI:	10.2147/DDDT.S515535
Fuente:	Nursing & Allied Health Database