Results
A total of 3540 pharmacokinetic records (99.1%) from 219 subjects (72 healthy volunteers and 147 HCV infection patients) were included for model development. 31 (0.9%) samples below LLOQ were excluded from analysis. All CWRES ranged between -6 and 6. No PK data points were identified as outliers in the structural model and were excluded in the process of model development. Baseline demographics and subject characteristics were summarized in Table 2.
A two-compartment model with sequential zero-first order absorption and first-order elimination was investigated to describe the pharmacokinetics of yimitasvir. The structural model was first established using intensive PK data from single ascending dose and multiple ascending dose (CTR20140854, CTR20150048 and CTR20170932) in healthy volunteers. A two-compartment model with first-order elimination was superior to a one-compartment model to describe yimitasvir PK profile. For the tested absorption model including first-order absorption with or without a lag time, sequential zero-first order absorption, transit absorption and saturable Michaelis-Menten absorption, transit absorption model had the smallest AIC value and fitted the data best from the goodness-of-fit plots. Sequential zero-first order absorption model also performed well. When the full PK data were used, transit absorption model converged difficultly and slowly. Finally, we chose sequential zero-first order absorption model. Residual variability was modeled by proportional error model.
After the base structural model was established, the dose effect on PK parameters was tested first. The linear model was simple and adequate to describe the effect of dose on bioavailability. With incorporation of linear model, the objective function value (OFV) decreased by 82 units. Model fitting was improved significantly especially for high concentration data points. The typical value of Alpha was fixed to 0.129 according to the model result using data from SAD and MAD studies, which indicated that the bioavailability of yimitasvir decreased 12.9% for each 100 mg dose increase.
Subsequently, a stepwise forward inclusion and a backward exclusion procedure were performed to screen different covariates. During univariate screen procedure, covariates of ALT and AST, which were highly correlated (r2 = 0.93), were statistically significant on the same PK parameter of apparent clearance. ALT was considered to be more significant (decrease in OFV 18.291) than AST (decrease in OFV 12.675). As a result, the effect of ALT on clearance was reserved for further covariate screening. Following the stepwise forward inclusion process, food status was identified to be a significant covariate on first-order absorption rate (Ka) and bioavailability (F), gender and ALT were significant covariates on apparent clearance and disease status on duration of zero-order absorption (Td). None of these significant covariates was removed during the process of backward elimination. The final model parameter estimiates and their associated precisions (percent confidence of variation, CV%), including the effect of significant covariates (P < .001), were provided in Table 3. The typical values (for a healthy male volunteer with ALT value of 31.6 IU l-1taking yimitasvir under fasted state) for apparent oral clearance (CL/F) and central volume of distribution (V/F) were 13.8 l h-1 and 188 l, respectively. Inter-individual variability for CL/F and V/F and the covariance of them were 48.5%, 73.6% and 56.5%, respectively. The shrinkage values for CL/F and V/F were 3.25% and 8.69%, respectively, which indicated that individual empirical Bayes estimates PK parameters could be used to predict yimitasvir exposures [13]. High-fat meal decreased Ka and F by 90.9% and 38.5%, respectively. Male subjects had a 22.2% higher yimitasvir CL/F than females. Baseline ALT was another significant covariate on apparent clearance. Duration of Zero-order absorption duration was longer in healthy volunteers (2.17 h) than that in patients (1.43 h). All the parameters were estimated with good precisions.
Goodness-of-fit plots for the final pharmacokinetic model are shown in Figure 2. The population and individual predictions agreed with the yimitasvir observations well, but there was a trend of under-prediction at high concentration data points (Figure 2A and 2B). Most of these data were from 200 mg group of phase 2 study. One possible assumption for the bias was the lack of precise dosing history in phase 2 trial although the last dosing time before each PK sampling was recorded. Most CWRES ranged from y = ±4 and were evenly distributed at y = 0 (Figure 2C and 2D) with no obvious bias over the population predictions and time indicating the proper choice of proportional model for the residual variability.
Figure 3 shows the pcVPC plot stratified by study. The observed data were evenly distributed around the median prediction and were mostly within the 90% percentiles of the predicted concentrations, which indicated that the model could adequately describe PK profile of yimitasvir. In the bootstrap for the final model, all the 1000 replications ran successfully. The population parameter estimates were close to the median values from bootstrapping analysis and fell within 95% confidence intervals (Table 3), suggesting that the final model was robust and accurate.
The influence of significant covariates on predicted steady state exposure (AUCss, Ctrough,ss and Cmax,ss) was presented in Figure 4. The result revealed that food status had a great impact on yimtasvir AUCssand Cmax,ss, but not Ctrough,ss. AUCss and Cmax,ss decreased by 38.5% and 58.9% after a high-fat meal, respectively. These results were consistent with the result of the yimitasvir food effect study. The Ctrough,ss of females was 54% higher than that of males, while AUCss and Cmax,ss of females were less than 30% higher than that of males. The magnitude of ALT on the steady-state exposure of yimitasvir was mild (~30%) for healthy volunteers with extreme ALT values (5th and 95th percentiles) relative to the typical healthy volunteers. There was no difference in exposures between patients and healthy volunteers.