Nonmem 7

Nonlinear mixed‐effects population modeling was used to build a joint PK model using both the DVT data and the AF data and to test the potential difference between the 2 populations. Similar to the DVT and AF models,12, 13 the joint‐based model was a 1‐compartment model, parameterized in terms of apparent plasma clearance after oral administration (CL/F), apparent volume of distribution after oral administration (V/F), and a first‐order absorption rate constant (k_a). Age and serum creatinine concentration were included as covariates on CL/F and, likewise, age and lean body mass on V/F. Interindividual variability (IIV) terms were incorporated in exponential format on CL/F and V/F. A proportional error model was used to describe the random residual error. The significance of study‐specific terms on the relative bioavailability (F1) and CL/F were examined using the likelihood ratio test with a critical level of P < .05. Model evaluations included graphical inspection of basic goodness‐of‐fit plots, reduction in objective function values, visual predictive checks, and the precision of parameter estimates. The analyses were performed using NONMEM 7.2 software (ICON Development Solutions, Hanover, Maryland) with the first‐order conditional estimation method with interaction. Additional statistics and graphs were generated using R 3.2.2.

Included in the PK/PD model development were 609 data points for serum etrolizumab concentrations and 448 data points for free circulating β7 receptors from 38 etrolizumab‐treated UC patients in the phase 1 study.
This PK/PD modeling was performed using the nonlinear mixed‐effects model implemented in NONMEM 7.2 (ICON Development Solutions, Ellicott City, Maryland). The first‐order conditional estimation method with interaction and ADVAN13 PREDPP subroutine was employed for all model runs. The software R‐3.2.2 (R Foundation, Vienna, Austria) was used for data‐set assembly, statistical computations, and graphics.

The data were analysed in a sequential manner using NONMEM VII (ICON Development Solutions, Ellicott City, MD, USA). A population approach was used where longitudinal data were available. Model selection was based on goodness of fit and diagnostic plots, parameter estimate precision, the Akaike Information Criterion value and the Objective Function Value. Internal validation of the models was performed by graphical comparison of the raw data used to develop the model with the 5th, 50th and 95th percentiles of 1000 model simulations (visual predictive check (VPC)). In addition, external validation was performed by VPC of the model vs experimental data not included in the fitting process, where available.