Robust fitting of pharmacokinetic models to Phase II/III clinical trial data
Jan Freijer, Inez de Greef - van der Sandt, Teun Post, Bart Ploeger.
LAP&P Consultants BV
Pharmacokinetic analysis of Phase II/III trial data is often hampered by uncertainties in the data. This is not so much due to errors in the observations, but more so due to the non-compliance to the planned doses and the in-correct (unwitnessed) recording of dose events by the subjects enrolled in the trial. Ignoring these errors can result in biased parameter estimates. Various approaches have been proposed to overcome this issue, including the modelling of non-compliance. An alternative way to reduce the influence of the uncertainties in the dosage is by reversing the viewpoint. This considers the observations that are extremely remote from the results of model predictions as being indicators of a mismatch between assumed and applied dosage. By allowing a separate residual error for these observations their influence on the estimated parameters is reduced, thus improving the statistical inference. This approach is comparable to the concept used in robust regression techniques. An iterative procedure is proposed for identifying the remote observations using outlier criteria and refitting the model, until the identified residual variability accounts for all identified remote observations. The technique is explored and illustrated by Monte Carlo simulations under various non-compliance scenarios and sparse sampling schedules, and back fitting the model on the simulated datasets. A comparison is made between the bias of the back-fitted models with and without applying the robust fitting approach.