studies is thus not necessarily the same across
the different dose groups. This self-selection of
drug exposure by the patients involved in dose-
ranging studies makes it hard to find the opti-
mal dose in the absence of a reliable measure of
external drug exposure.The erroneous assumption of perfect compli-
ance when fitting PK models to observed con-
centration measures leads to a poor fit, often
accompanied by non-convergence and ser-
iously biased results. However, when drug
intake times are electronically recorded over
the course of the study, the appropriate hier-
archical model for nonlinear effects over time
can translate the data into accurate estimates.
Patients vary the dosing interval, and electronic
monitoring of a wide range of over- and under-
dosingpatterns hasbeena neglected opportunity
to observe and model realistic concentration–
time–effect relationships. Underdosing, drug
holidays and undetected early cessation of dos-
ing are common features of clinical trials, and
likely are frequent sources of low response and
high variability in responseto the protocol-
specified dosing regimen. Especially for non-
linear PK/PD estimation, not only bias can be
reduced but also higher precision can be attained
from the same number of data points when
irregular drug intake times occur and are cap-
tured by electronic monitoring, even in well-
controlled studies. Estimators of PK/PD
parameters gain in robustness when they are
based on the actual, rather than assumed, exter-
nal drug exposure.
By a better characterization of the dose-
response characteristics of a product candi-
date, the program will bring two major
advantages to drug development:
& Allow development of the drug at the appro-
priate dose and avoid potential, post-
marketing and post-pricing dose reductions
as are occurring today for more than 22%
of products (Crosset al., 2002; Heerdink
et al., 2002).& Allow identification of compounds that, if
taken at an appropriate level of adherence,
could successfully reach the market but
would have failed a conventional phase III
program if subjected to the usual, wide
range of adherence.The improved quality of the recorded exposure
data is expected to have the net effect of signific-
antly reducing theworkload of the PK/PD group.The number of queries for outlying concentra-
tions will be reduced to a minimum. Typically,
when relying on the assumption of steady
state, more than 20% of observed concentra-
tions have to be queried. Electronic compila-
tion of dosing histories will reduce this
proportion of queries to a<2%.Consequently, the time needed to remove the
observation points based on above queries, as
well as the time to justify that some patient
data were removed, will be reduced.The time needed to arrive at a satisfactory PK/
PD model will be reduced.A sound pharmionic program will reduce by
half the amount of time needed to process a
population PK/PD study.Ultimately, a pharmionic program will hasten
and improve the ‘Go/no-Go’ decision to start
the development phase (phase III).A pharmionic program will allow an earlier
andabettercharacterizationofthedose–time–
response surface in the target population.It will deliver faster and greater insight than
is now possible for dose selection in dose-
ranging studies.
It will facilitate a degree, not now possible, of
‘bullet-proofing’ against post-marketing/post-
pricing reductions of the ultimately recom-
mended dosing regimen. This consideration
grows in importance as pharmaceutical prices
increase.27.6 DURING EARLY PHASES OF DRUG DEVELOPMENT (PHASE I AND II STUDIES) 359