Model Development
Population pharmacokinetic analysis of yimitasvir was performed by Phoenix NLME software (Version 8.1, Certara) using the first-order conditional estimation-extended least squares (FOCE-ELS) method.
The structural model was first tested by fitting a one-compartment or two-compartment model to the log-transformed PK data. Different absorption models including first-order absorption with or without a lag time, sequential zero-first order absorption, transit absorption and saturable Michaelis-Menten absorption model, were also tested. The optimal structural model was selected based on the Akaike Information Criteria (AIC), minimization success, visual inspection of goodness-of-fit plots and individual fit.
Inter-subject variability was estimated by exponential model for PK parameters as follows (Eq. 1):
θi = θTV × eηi (1)
where θi is the parameter estimation for the i th individual, θTV is the typical value of the parameter estimation in the population, ηi is a random variable which assumed to be normally distributed with a mean of 0 and a variance of ω2. Proportional error model and proportional plus additive error model were tested as residual error models.
Following the development of structural model, the dose effect on bioavailability was evaluated first due to the less than dose-proportional profile of yimitasvir. Sigmoidal maximum effect (Emax) (Eq.2) and linear models (Eq.3) were tested to quantify the relationship between bioavailability and dose:
F = θF – Fmax × (Dose - 100)/(F50 + (Dose - 100)) (2)
where θF is the bioavailability in individuals who received 100 mg yimitasvir, which was fixed to 1. Fmaxis the maximal reduction in bioavailability and F50 is the dose associated with a half-maximal reduction in bioavailability.
F = θF – Alpha × (Dose -100)/100 (3)
where Alpha is a slope term determining the relative change in bioavailability for each 100 mg increase in yimitasvir dose.
Subsequently, different covariates were tested using a stepwise forward inclusion (a decrease in objective function value [OFV] of > 6.63, P < 0.01) and a stricter backward exclusion procedure (an increase in OFV of > 10.83,P < 0.001). The covariates included age, gender, body weight (BW), body mass index (BMI), baseline haemoglobin (HGB), baseline aspartate aminotransferase (AST), baseline alanine aminotransferase (ALT), baseline albumin (ALB), baseline total bilirubin (TBIL), baseline creatinine clearance (CLcr) calculated by Cockcroft-Gault formula12, co-medication of sofosbuvir, disease status (healthy volunteers vs patients) and food effect.
The effect of continuous covariates was modeled using a power function after normalization by the population median (Eq. 4):
θi = θTV × (cov i/cov median)θx (4)
The effect of categorical covariates was modeled using exponential format as follows (Eq. 5):
θi = θTV × eθx cov = k(5)
where cov i and cov medianrepresent covariate values for the i th individual and the population median, respectively. k is a categorical variable, and θx is a coefficient used to describe the strength of the covariate effect.
When two covariates were highly correlated (r2> 0.7) such as ALT versus AST, only the most significant one was reserved in the model if both covariates were considered to be significant for the same PK parameter during univariate screen process.