Population pharmacokinetic modelling
The structural model was developed using 374 data of GBP plasma concentration. It consisted of a one-compartment with the inclusion of IIV on absorption constant (Ka), lag time, volume of distribution (Vd) and clearance (CL). The PK profiles are shown in Figure 1A. A correlation between the IIV of Vd and CL also improved the model. A proportional error model was best to estimate the unexplained residual variability rather than combined or additive error models. The estimates of lag time, Ka, Vd and CL were 0.32 h, 1.13 h-1, 140 L and 14.7 L/h, respectively. The estimates of IIV expressed as RSE were 16.4% (lag time), 17.4% (Ka), 14.1% (Vd) and 13.4% (CL), respectively (Table 2).
Covariates selection was based on variables showing parameter-covariates relationship with p-value <0.05 (models No. 2-6, Table 3). Forward inclusion ended with four covariates; the OCTN1 genotype was not included because it resulted in high RSE (Table 3). Backward elimination was performed on the full model obtained and included the following covariates: a) eGFR on CL; b) body height and FGL on Vd. The addition of body height on Vd improved the model. Interestingly, other covariates with a strong correlation with height, such as BMI and weight, or covariates with better clinical explanation, such as weight, did not improve the model. Even though sex was included as a covariate on Ka on the forward inclusion step, it was removed on the backward elimination step.
The pharmacokinetic parameters of the final model and the bootstrap are presented in Table 2. The final model showed good predictive performance since RSE values were below 40%. The precision of the parameter estimates evaluated through bootstrap analysis showed that the zero value was not included in the 95% confidence intervals in any case. The absence of bias in the goodness-of-fit plots presented in Figure 1B-C illustrates the acceptable predictive value of the model.