Niclas Jonsson(1) , Janet R Wade(1) , Gun Almqvist(2) and Mats O Karlsson(1)
Dept of Pharmacy, Uppsala University(1)
Sweden Astra Draco AB, Lund, Sweden(2)
Obtaining an adequate dosing history for population PK modelling is associated with considerable problems. Essentially three types of dosing histories can be considered: (i) per protocol, (ii) subjective measures (diary and patient information to monitoring staff), and (iii) ''objective'' measures (electronic monitoring devices and staff supervised intake). All three measures can be erroneous in that at the recorded dosing occasion no (or several) dose(s) may have been taken or that a dose can be taken at an unrecorded time. To document the dosing history in more than one way increases the possibility that at least one dosing history is correct, but one has then to choose between possibly different dosing histories. It is possible to device a strategy for dosing history selection (e.g. between subjective and objective measures) in two steps: (i) A preliminary PK model is used to predict concentrations based on the rival dosing histories. If the difference between rival dosing history predictions is less than a certain cut-off value (e.g. 5%), it is concluded that the dosing histories are in agreement (CDH - consistent dosing histories) and the choice between them is irrelevant. (ii) Using CDH data, a population PK model is developed. This model is then used to select the most likely of the rival dosing histories when these show significant differences (SDH - selected dosing histories).
In a real data example, we have applied the above scheme and
compared it to other ways of constructing the dosing history.
Dosing histories were recorded for 120 outpatients (in total 224
visits) using both subjective (diary, patient information to
monitoring staff) and objective (MEMS) means. For 78
patient visits the two dosing histories were in agreement (CDH
data). Of the SDH data, the MEMS
and subjective
measures were used for 36 and 66 patients visits, respectively.
For 44 visits only the subjective measures were available. The
CDH and SDH data were found to be associated with similar
residual and inter-occasion variability (which are assumed to be
indicative of ''noise'' introduced by errors in dosing history),
whereas basing the analysis on either subjective or objective
measures only, increased these variance terms. The population CL
estimate and its relation to the main covariate was relatively
robust to different dosing history selection criteria, whereas
changes could be observed in volume terms and the estimated
terminal half-life.