Joint meeting with Manchester University's Biostats Group

Theme: "Developments in Longitudinal Data Analysis"

The validation of surrogate endpoints has been initially studied by Prentice and Freedman.  Noting operational difficulties, Buyse and Molenberghs proposed instead to use jointly the within-treatment partial association of true and on the true outcome.  In a multi-centre setting, these quantities can be generalised to individual-level and trial-level measures of surrogacy.  Buyse and colleagues have proposed a meta-analytic framework to study surrogacy at the trial and individual-patient levels.  Variations for various endpoints have been developed.  Efforts have been made to converge to a common framework.  This includes a so-called variance reduction factor and an information-theoretic approach.  Work has been done regarding sample size assessment, leading to the surrogate threshold effect.

JIANXIN PAN (Univeristy of Manchester)

In tumour xenograft experiments, several treatment regimens are administered and tumour volume for each individual is measured repeatedly over time.  When modelling such data, certain constraints are imposed on regression parameters to account for intrinsic growth of tumour in the absence of treatment.  On the other hand, survival data and cure data are observed due to a portion of individuals who may be curved and so never experience the event.  In this talk, we will show how to jointly model the longitudinal, survival and cure data in order to account for the possible association of those data.  Simulation studies show that the proposed joint modelling approach does improve the separate modelling in terms of mean square errors of parameter estimates.

FRANK WINDMEIJER (University of Bristol)
Estimation of dynamic panel data models by Generalised Method of Moments, the issue of weak instruments

A commonly used estimation procedure to estimate the parameters in dynamic panel data models with constant unobserved individual specific heterogeneity is to transform the model in first differences and use observations on the variables in the past periods to instrument the endogenous differenced explanatory variables, and estimate the parameters by GMM.  Weak instrument problems arise when the series are very persistent.  The presentation will show the effects of this weak instrument problem on the estimation results, discusses tests for the detection of weak instruments and the relative merits of using different moment conditions and/or different estimation techniques, given the moment conditions.