A latent-variable model for Sorafenib-induced Hand-Foot Syndrome (HFS) in non-selected patients to predict toxicity kinetics according to sorafenib administrations
Emilie Henin (1), Benoit Blanchet (2), Pascaline Boudou-Rouquette (2), Gilles Freyer (1), François Goldwasser (2), Michel Tod (1)
1: EMR HCL/UCBL 3738 CTO, Faculté de Médecine Lyon-Sud, Université de Lyon; 2: Laboratoire de pharmacologie-toxicologie, Service de Pharmacie, Hôpital Cochin, Paris
Introduction & Objectives: Sorafenib is a multi-kinase inhibitor, targeting especially Ras-ERK and VEGFR pathways, for the treatment of advanced renal cell carcinoma (RCC) and unresectable hepatocellular carcinoma (HCC). Sorafenib was found to induce cutaneous, hematological and metabolic toxicities. Hand-Foot Syndrome (HFS) is characterized by an inflammation of the skin on palms and soles. Its physiopathological mechanism has not been fully understood yet, but several hypotheses suggested the accumulation of a toxic compound in skin cells. The objective of our work was to propose a physiologically coherent model for the sorafenib-induced HFS on a long-term basis in non-selected patients and to quantify the risk dynamics, linked to sorafenib doses.
Patients & Methods: 89 patients treated with sorafenib were unselectively considered for the modeling work. Treatment duration, sorafenib regimen and number and frequency of HFS observations were highly variable.
A non-linear mixed effect model was built to link sorafenib administrations to the risk of each HFS score. The drug, whose PK was described by a saturable absorption one-compartment model as proposed by Hornecker et al, was considered as impacting the kinetics of a latent variable (LV). The probability of each HFS score is computed from a probit function of the LV level and corresponding threshold parameters. Parameters were estimated in NONMEM7.1.2. Model evaluation was driven by goodness-of-fit and simulation-based diagnostics, using Xpose package in R.
Results: A physiologically coherent model, relating sorafenib administrations and, per se its exposure, to HFS dynamics has been built. The latent variable can be interpreted as an unobserved quantity (e.g. a non-identified biomarker), which would be directly related to the resulting HFS score. The latent variable has a half-life of 7 days, whereas sorafenib has a plasma half-life of 35 hours. Our model allows taking into account the differences between the kinetics of the drug concentration, and the kinetics of the toxicity. HFS toxicity is of great inertia: HFS development and resolution are generally slow processes compared to sorafenib kinetics.
Understanding the dynamic relationship between drug administrations and an induced adverse event is essential to control toxicities and adequately adjust treatment modalities. A pharmacokinetic-pharmacodynamic model for sorafenib-induced HFS can be used as an early predictor of severe toxicity risk in patients.
