Population pharmacokinetics of high-dose methotrexate using deep learning analysis of CT scan radiomics and biometric and laboratory data: the PREDICT-MTX study

Abstract

BACKGROUND: High-dose methotrexate (HDMTX) is widely used as an established treatment for various haematological malignancies. However, prolonged exposure can lead to significant toxicity. Recent research showed the possibility to accurately estimate the creatinine clearance with the CT-based estimate of RenAl FuncTion (CRAFT) equation using automated body composition analysis of clinically obtained CT scans and automated deep learning algorithms.

Based on existing pharmacokinetic models for HDMTX, this study investigated the added value of incorporating the CRAFT, radiomics parameters and biometric and laboratory values for better predicting the pharmacokinetics of HDMTX.

METHODS: The PREDICT-MTX study is a retrospective single centre pharmacokinetics study that included patients treated with HDMTX (≥ 500 mg/m2) with a clinical acquired CT-scan covering the L3 segment. A population pharmacokinetics model was constructed using non-linear mixed effect modelling (NONMEM) to estimate population and individual pharmacokinetic parameters. Radiomics parameters, biometric and laboratory values were evaluated as covariates.

RESULTS: The MTX concentration–time course was best described by a three-compartment model. Significant covariates that retained in the final model were serum creatinine concentration and CRAFT on methotrexate clearance (CLmtx) and white blood count and 90th percentile radiation attenuation of long spine muscles on volume of distribution in the central compartment (V1).

CONCLUSION: This is the first proof-of-concept study that uses deep learning body-composition analysis of clinically acquired CT-scans to better describe the pharmacokinetics of HDMTX. We constructed a three-compartment population pharmacokinetic model that characterised the CLmtx and V1 of MTX in adult patients with various malignancies treated with HDMTX.