Apologies for any cross-posting
Cluster Randomization Trials Heidelberg
May 22 – 24, 2003
Places still available!
Goals and Contents
A cluster randomization trial (CRT) is a trial in which clusters of
individuals, rather than individuals themselves, are randomized to
different intervention groups. CRTs have come to be an important tool in
the evaluation of non-therapeutic interventions, including lifestyle
modification, educational programmes and innovations in the provision of
health care. The units of randomization in such trials are diverse,
ranging from relatively small clusters such as households or families, to
entire neighbourhoods or communities, but also including places of work,
hospital wards, classrooms and medical practices. The focus of this course
is on primary care trials, which are typically conducted to assess methods
of health care delivery.
This course will consider the design and analysis of such trials. The
primary outcome variable in most cluster randomization trials is
continuous or binary. Therefore, the statistical methods discussed in our
course will present approaches to handle both cases: Models for random
effects and generalized estimation equations (GEE). A short review of
relevant theory as well as hints how to use available software will be
demonstrated on real data sets. The appropriate interpretation of
statistical outcome will play an important role.
With CRTs, there are two sample size issues: How many clusters and how many
patients per cluster? Sample size estimation is usually based on the ICC,
the intra-cluster correlation. There are problems related to the use of
the ICC which depend - like the Pearson correlation coefficient - on the
choice of the independent variable. Issues discussed are the estimation and
range of values of the ICC found in general practice, and the associated
sample size problems.
Practical aspects of CRTs like ethics, handling of drop-outs, formulation of
the intervention, and techniques in maximising compliance will be
discussed. A critical review of the literature and hints how to read a
paper on performed CRTs will be given.
Schedule
Thursday
09.00 - 09.15
Introduction and Welcome
09.15 - 10.30
Introduction in CRT
11.00 - 12.30
Summary measures, Meta-Analysis
13.30 - 15.00
Sample size considerations and ICC
15.30 - 17.00
Classical Hierarchical Models for the analysis of gaussian outcome data
Friday
9.00 - 10.30
Classical Hierarchical Models for the analysis of binary outcome data
11.00 - 12.30
Critique of papers
13.30 - 15.00
Bayesian approach: theory and example
15.30 - 16.30
GEE approach: theory and example
16.30 - 17.00
Interpreting statistical outcome: comparison of marginal and cluster
specific models
Saturday
9.00 - 10.30
Supervised studentwork
11.00 - 12.30
Practical aspects of CRTs: ethics, drop-outs, formulation of the
intervention, maximising compliance
13.30 - 15.00
Discussion of studentwork, Evaluation of the course
Teaching language is English
Lecturers
Rumana Omar, Department of Statistical Science, UCL, 1-19 Torrington Place,
London WC1E 6BT
Michael J. Campbell, Institute of General Practice, Community Sciences
Centre, Northern General Hospital, Sheffield S5 7AU
Ulrich Mansmann, Institut für Medizinische Biometrie und Informatik, INF
305, 69120 Heidelberg (Coordinator)
Organization
Venue
The course will take place in Heidelberg at the university campus “Im
Neuenheimer Feld”. A detailed description of how to get there as well as
other pertinent information will be sent with the confirmation of your
registration.
Course fee
The fee for the course is 450,- Euro; for university employees 310,- Euro.
Information
http://www.biometrie.uni-heidelberg.de/postgraduate_education/postgrad1.html
Concept and Contents
Universität Heidelberg, Abt. Medizinische Biometrie
Coordination
Dr. Christine Wollermann
Tel: +49(0) 62 21/56–47 29, Fax: –41 95
[log in to unmask]
Organization
Academy for Postgraduate Education at the Universities of Heidelberg and
Mannheim
Friedrich-Ebert-Anlage 22-24, D-69117 Heidelberg
Tel: +49(0) 62 21/54–78 10, Fax: –78 19
[log in to unmask]
|