Exponential model was used to describe
interindividual variability of CL/F (ω2
CL/F), while an additive, proportional,
and combination error models were tested for residual
variability of VPA concentration (2), which includes model
misspecification, intraindividual variability, interoccasional variability,
and errors in dosing, sample collection, and assay. The
model appropriateness was evaluated by standard diagnostic plots.
Additional criteria were convergence of minimization, number of
significant digits more than 3, successful covariance step and gradients
in the final iteration in the range between 10−3 and 102.
Alternative models were compared by the likelihood ratio test. Criterion
for selection of a model was a change in minimum value of
objective function (OBJ) of at least 3.84 per one additional parameter,
corresponding to p < 0.05