Statistics in Mental Health Research
For all section and study group meetings held at the RSS pre-registration is recommended. You can register by email:
[log in to unmask] or by phone (020) 7638 8998.
Time 2.00-6.00
Venue (Errol Street unless otherwise specified)
Meeting organized by GAS and Medical Section
Meeting organiser: Sabine Landau and Ian White
Contact details: [log in to unmask] <mailto:[log in to unmask]>
and [log in to unmask] <mailto:[log in to unmask]>
2:00-2:30pm Graham Dunn (Biostatistics Group, University of Manchester)
Modelling treatment-effect heterogeneity in RCTs of psychological
treatments
I describe instrumental variable (IV) methods for the estimation of the
'dose'-response effects of psychological interventions in RCTs in which
there is variability in the number of sessions of therapy attended, the
effect of which is modified by the strength of the therapeutic alliance
between patients and their therapists. The IV methods allow for (a)
hidden confounding, (b) measurement errors and (c) that alliance is only
measured in those receiving treatment.
Three two-stage estimation procedures are illustrated, and their
equivalence demonstrated, through Monte Carlo simulation and analysis of
the results of an actual trial (SoCRATES).
2:30-3:00pm Morven Leese (Health Service and Population Research
Department, Institute of Psychiatry, KCL, London)
Aggregate scores as psychiatric outcomes
In the QUATRO trial of therapy to improve adherence to antipsychotic
medication, fifteen outcomes, covering adherence, symptoms and quality
of life, were collected. This large number of multiple measures is
typical of psychiatric research. However, despite the appeal of a wide
perspective on treatment outcome, a smaller set of primary outcomes is
generally required for clinical trials. One solution, the combination of
measures into aggregate scores, is discussed in this talk. The focus is
on assessing the reliability of the resulting scores rather than on the
specific rationale for combination. Various ways of combining the eight
subscales of the SF-36 (one of the quality-of-life measures used in the
QUATRO trial) are compared in an illustrative example.
3:00-3:30pm Tim Croudace (Department of Psychiatry, University of Cambridge)
"Latent Goldberg" - psychometric statistics
and the General Health Questionnaire
Epidemiological studies often use psychometric instrumentation from
Goldberg's GHQ family to assess psychiatric morbidity. New psychometric
models at the interface between latent variable models, mixture models
and multilevel models enable applied researchers to model GHQ
questionnaire responses at the item level in many subtly different ways
that go beyond the somewhat blunt approach of the traditional linear
factor model. Examples of alternative latent structure models for GHQ
data will be introduced. Insights as to how these models may be extended
will also be discussed.
4:00-4:30pm Mick Brammer (Department of Biostatistics and Computing,
Institute of Psychiatry, KCL, London)
Some issues arising in the statistical analysis of functional fMRI data
Functional magnetic resonance imaging (fMRI) is a very widely used
method for imaging human brain function in vivo. This increasing level
of use means that statisticians working in the field of medicine may
encounter such data and be asked to offer advice on analysis. The aim
here is to explain the basic features of fMRI data sets and to introduce
some of the approaches to analysis and inference based both on
adaptations of standard inferential procedures and on data-driven
permutation-based approaches.
4:30-5:00pm Chris Roberts (Biostatistics Group, University of Manchester)
The design and analysis of clinical trials of non-pharmacological
therapies.
The issue of clustering effects due to treatments in individually
randomised trials, whilst long-recognised is generally ignored in both
design and analysis. Such effects may be hypothesised in trials of
psychological treatment involving talking therapies and where treatment
is delivered to patients together as a group rather, for example group
therapies for psychological problems. As with group-randomised trials,
between cluster variation can lead to lack of independence between
patients' outcomes. The implications of this for design, including
sample size estimation, and statistical analysis will be discussed.
|