Population PK Analysis
PopPK was performed to provide model predicted individual
pharmacokinetic (PK) parameter estimates for exposure-response (ER)
analyses. Data from seven Phase 1 studies, four Phase 2 studies, and
three Phase 3 studies were analyzed using nonlinear mixed-effects
modeling [NONMEM (Version 7.3.0 or later; GloboMax, Hanover, MD)]
and Perl‑Speaks‑NONMEM (PsN; Uppsala University, Sweden). Summary of the
data used in the PopPK modeling is shown in supplementary Table 1.
Filgotinib and GS-829845 models were developed separately. Various
structural models and random effect models were evaluated to reach the
base models. Stepwise forward addition and backward deletion was
implemented in the covariate model building process, with evaluated
covariates including demographics (age, sex, body weight, race),
pathophysiological factors [baseline estimated creatinine clearance
(CLcr), baseline bilirubin, baseline alanine
aminotransferase, baseline aspartate aminotransferase, RA disease status
(subjects with RA vs healthy subjects), baseline C-reactive protein
(CRP), and RA duration], and fed status (always fed vs mixed
fasted/fed vs always fasting; evaluated on absorption-related
parameters). Model selection was done based on a log-likelihood ratio
test at an acceptance p-value of 0.01 (forward addition) or 0.001
(backward elimination). The difference in -2 times the log of the
likelihood (-2LL) between a full and reduced model was assumed to have a
χ2 asymptotic distribution with degrees of freedom equal to the
difference in number of parameters between the 2 models. Model
performance evaluation was based on Goodness-of-Fit evaluation,
Prediction Corrected Visual Predictive Check (pcVPC), and Bootstrap
Resampling Techniques. Individual PK parameter estimates were predicted
from the final models for ER analyses.