Population PK modelling
The identification of structural differences in the PK properties of s.c. and i.d. administration, while accounting for covariates such as the presence of anti-adalimumab antibodies, was investigated using a population non-linear mixed effects modelling approach in NONMEM (ICON plc, V7.3). Based on literature information, a one compartment structural model with linear absorption and linear elimination was used during model development (23). For this structural model, the effect of anti-drug antibodies on the CL of adalimumab was tested as a time-varying covariate, increasing the CL of adalimumab at higher titre levels with the following equation: CL = ΘTVCL * (1+ ΘTVTitre-slope * TITRE), Where individual TITRE levels proportionally increase the CL of an individual over time.
When a structural misspecification was identified in the absorption phase, modifications to the absorption part of the model were explored, in which transit models, different absorption compartments, and a MTIME function in which the ka changes after an estimated time point, were investigated, modelled separately for each administration route.
After identification of the best structural absorption models for each route of administration, log-transformed inter-individual variability (IIV) was included following a forward inclusion procedure (p<0.01) and covariates (age, weight, body mass index, sex, serum creatinine, and albumin) were explored following a forward-inclusion (p<0.01) with backward-elimination (p≤0.001) procedure. Continuous covariates were tested following a power relationship centered around the median. Models were evaluated on basis of the objective function value (OFV), the parameter uncertainty (judged by the relative standard error [RSE]), goodness-of-fit figures, individual model predictions versus observations over time, and confidence interval visual predictive checks (ciVPC) based on 500 Monte Carlo simulations. Bootstrapping was not considered of added value as additional model evaluation tool. Data transformation was performed in R (V3.6.1(22)) and models were executed in conjunction with Perl-speaks-NONMEM (V4.8.1) (24).