Joint meeting with Manchester University's Biostats Group

Theme: "Screening for identification, efficiency and prediction"

For survival data we discuss two approaches for accounting for model selection uncertainty with the main emphasis on variable selection in a proportional hazard Cox model. The first approach is Bayesian model averaging. The second approach carries out model averaging with weights estimated from bootstrap resampling.

FIONA MATTHEWS
Estimating measures of disease incidence from longitudinal two-phase designs in the presence of missing data, do we lose more than we gain?

Two phase designs are common to estimate disease prevalence, but less is known about their use for longitudinal estimates. This talk will discuss the issues surrounding estimation of disease from such longitudinal designs. The modelling of the missing data due to study design, drop-out and mortality will be discussed.

ROSEANNE MCNAMEE
Efficient designs for evaluation of screening tests

Evaluation of screening tests from 2-phase cohort designs seems attractive as the gold standard need only be applied to random samples of subjects and formulae for the optimal design of such studies exist. The statistical efficiency of these and other 2-phase designs, eg for prevalence, regression coefficients, will be questioned.