Page 395 - Contributed Paper Session (CPS) - Volume 6

CPS1995 Daniel B. et al.
and from one biologic product to a biosimilar product (or one formulation to
a bioequivalent formulation).
The regulatory guideline for extrapolation calls for the extension of
information and conclusions available from studies in one or more subgroups
of the patient population (source population(s)), or in related conditions or
with related medicinal products, to make inferences for another subgroup of
the population (target population), or condition or product. This definition
was proposed in a European Medicines Agency (EMA) concept paper on
extrapolation of efficacy and safety in medicine development. This then
reduces the need to generate additional information (types of studies, design
modifications, number of patients required) to reach conclusions for the target
population, or condition or medicinal product.
The procedure for extrapolation is facilitated by the notion of estimand as
introduced in the revision of International Conference on Harmonization (ICH)
E9(R1). The proposed framework states that an estimand reflects what is to
be estimated to address the scientific question of interest posed by a clinical
trial. The choice involves the population of interest, endpoint of interest, and
measure of intervention effect. Ultimately one can think about the problem
of extrapolation as weighting of available data based on defined estimands
[Akacha, M., et. al. (2017)] for the scientific question of interest.
We assume that observations of interest arise from a parametric
distribution function whose parameters can be estimated based on an
estimating equation such as the derivative of the log-likelihood function.
Allowing the possibility of censored/grouped data transforms the likelihood
expression into a likelihood involving counts of interval data by utilizing the
latent variable concept [Lazarsfeld, P. F. & Henry, N. W. (1968)]. We also
assume that the observable outcomes are intervals of the form (a, b], a < b,
where b is the maximum measurement level at which the subject experiences
a defined response, e.g. treatment success, and a is the lower bound for such
a response. In this construction the measurement level defining the response
is latent.
The paper is organized as follows. Section 2 presents the maximum
likelihood procedure for estimating distribution parameters under a
generalized linear models setting. Section 3 provides the extrapolation
construction for a normal latent measurement variable and a homogeneous
Poisson latent count of events. Then using large sample approximation, we
show an approach for goodness-of fit testing that can be extended to
accommodate an extrapolation setting. Utility of the estimation and
goodness-of-fit testing approach is shown in a setting where we evaluate
efficacy in a subgroup of a clinical trial population using a marker of efficacy.
Finally, Section 4 gives some concluding remarks.

384 | I S I W S C 2 0 1 9