Joint modelling of the disease process and the treatment
allocation process in long term follow-up studie
Juni Palmgren
Rolf Nevanlinna Institute, Section of Biometry
00014 University of Helsinki
Finland
Abstract
Assessment of the effect of 'treatment actually
received' on a disease endpoint is obscured by the fact that
(i) those who 'adhere to treatment', i.e. the compliers,
may differ from
the noncompliers on important risk factors, and (ii) the disease
endpoint itself may influence the treatment process. Using the
notion of potential outcomes we specify a model for the treatment
process and for the disease process and for the association
between these.
We relate observed data to the potential
outcomes and discuss estimation strategies for retrieving
'causal' parameters of interest. We illustrate the argument on
a randomized placebo controlled study
by relating the total active dose received to a survival endpoint.
We discuss implications of the randomization.