Influence of the skeletal muscle index on pharmacokinetics and toxicity of fluorouracil

Abstract

Background The body composition of patients has been associated with tolerability and effectiveness of anticancer therapy. This study aimed to assess the influence of the skeletal muscle index (SMI) on the pharmacokinetics and toxicity of fluorouracil. Methods Patients treated in an oncological practice with fluorouracil‐based chemotherapy and undergoing therapeutic drug monitoring were retrospectively investigated. Computed tomography images were analyzed to measure abdominal skeletal muscle areas in Hounsfield units for the psoas major muscle, back and total skeletal muscle to determine the SMI. For the latter, an automated segmentation method was used additionally. SMI measures were tested as covariates on fluorouracil clearance in a population pharmacokinetic model. Furthermore, regression analyses were performed to analyze the influence of SMI measures on the probability of clinically relevant adverse events (CTCAE grades ≥ 2). Results Fluorouracil plasma concentrations of 111 patients were available. Covariate analyses showed significant improvements of the model fit by all SMI measures. However, interindividual variability of fluorouracil clearance was only slightly reduced, whereas the SMI of the back muscle showed the largest reduction (−1.1 percentage points). Lower SMI values of the back muscle increased the probability for polyneuropathy and lower SMI of the psoas increased the probability for fatigue. Conclusions Our results suggest that pharmacokinetics and toxicity of fluorouracil may be associated with specific SMI measures which deserve further investigation.

Comparing Scientific Machine Learning with Population Pharmacokinetic and Classical Machine Learning Approaches for Prediction of Drug Concentrations

“…In the Pop-PK analyses, all PK parameters and their IIVs as defined in the original publications [28,29] could be estimated for all data splits. The mean estimated parameter values were in a similar range to the original estimated values for the whole datasets as depicted in Table 2 and the ߟ values appeared to be normally distributed for all tested methods.…”

“…Our proposed MMPK-SciML model was able to predict the population volume for 5FU, which was impossible to obtain with FOCE-I due to convergence problems [29]. We observed that the predicted volume was often overpredicted, which could be due to outliers in the training data.…”

“…The residual variability was modelled as proportional and the body surface area, centered on the population median, was included as a linear covariate on clearance. Differently from the original model [29], the skeletal muscle index was not included as a covariate, because it was not available for all included patients. While Schmulenson et al used the FOCE-I method to estimate the parameters [29], we compared this method to stochastic approximation expectation maximisation (SAEM) using NON-MEM®.…”

“…Differently from the original model [29], the skeletal muscle index was not included as a covariate, because it was not available for all included patients. While Schmulenson et al used the FOCE-I method to estimate the parameters [29], we compared this method to stochastic approximation expectation maximisation (SAEM) using NON-MEM®. First, we estimated the pharmacokinetic parameters for the training patients of each split using the different methods and initial estimates as in [29].…”

“…While Schmulenson et al used the FOCE-I method to estimate the parameters [29], we compared this method to stochastic approximation expectation maximisation (SAEM) using NON-MEM®. First, we estimated the pharmacokinetic parameters for the training patients of each split using the different methods and initial estimates as in [29]. In the next step, these values were used to simulate the expected concentrations for the test data.…”