The next Royal Statistical Society Medical Section meeting will
be from 2pm on Tuesday 30th November, and will be held at the
Royal Statistical Society, 12 Errol Street, LONDON, EC1Y 8LX.
The meeting will consist of five talks on various aspects of
PUBLICATION BIAS in META-ANALYSIS. The programme appears below,
and abstracts can either be found below the programme or at
http://www.prw.le.ac.uk/epidemio/personal/kra1/pubias.html
Directions to the Royal Statistical Society can be found at
http://www.rss.org.uk/index2.html and any further details can be
obtained from Keith Abrams, Tel 0116 252 3217 or
e-mail [log in to unmask]
All are very welcome
Royal Statistical Society Medical Section
Meeting on PUBLICATION BIAS in META-ANALYSIS
2pm - 5.30pm, Tuesday 30th November 1999, at
Royal Statistical Society,
12 Errol Street, LONDON, EC1Y 8LX
2.00 - 2.40 Introduction to Publication Bias and Related Problems
FUJIAN SONG (University of York)
2.40 - 3.15 Tests for publication bias. Do they work, and how do we
interpret their results?
JONATHAN STERNE (University of Bristol)
3.15 - 3.40 Detecting bias and correcting for bias: The case of
placebo-controlled trials of homoeopathy
MATTHIAS EGGER (University of Bristol)
3.40 - 4.15 Tea
4.15 - 4.50 Adjusting the results of a meta-analysis for publication
bias
ALEX SUTTON (University of Leicester)
4.50 - 5.30 Modelling Publication Bias
JOHN COPAS (University of Warwick)
************************* ABSTRACTS ****************************
Introduction to Publication Bias and Related Problems
FUJIAN SONG (University of York)
Abstract: A review was conducted to identify and appraise studies that have
examined methodological issues and provided empirical evidence about
publication bias. The available evidence demonstrates that research with
significant results or favourable results is more likely to be published
than that with non-significant or unfavourable results. The extent and
direction of such selective publication is uncertain, and may vary greatly
depending on the circumstances. Methods available to detect or adjust
publication bias in systematic reviews are by nature indirect and
exploratory, and the risk of publication bias should be assessed in all
systematic reviews.
Tests for publication bias. Do they work, and how do we interpret
their results?
JONATHAN STERNE (University of Bristol)
Abstract: Funnel plot asymmetry has long been used to diagnose publication
bias. We examine the power of statistical analogues of the funnel plot which
have been proposed as tests for publication bias, and the frequency with
which they detect bias in the medical literature. Publication bias is one of
a number of possible causes of funnel plot asymmetry, and multivariable
meta-regression analyses may be needed to understand different causes of
between-trial heterogeneity.
Detecting bias and correcting for bias: The case of placebo-controlled
trials of homoeopathy
MATTHIAS EGGER (University of Bristol)
Abstract: A recent meta-analysis of 89 placebo-controlled trials of
homoeopathy was interpreted as showing that the effects of homeopathy are
unlikely to be due to placebo (Linde et al, Lancet 1997). The authors
corrected for publication bias using a selection model and performed a
number of sensitivity analyses. In this short presentation I will discuss
the approach of Linde et al. and present a re-analysis using random-effects
meta-regression which shows that a combination of publication bias and low
methodological quality of component studies may explain the apparent
beneficial effect of homoeopathy.
Adjusting the results of a meta-analysis for publication bias
ALEX SUTTON (University of Leicester)
Abstract: Publication bias can distort and possibly invalidate the
conclusions of a systematic review. The newly proposed method of "Trim and
Fill" is described, which, in addition to offering a significance test for
the presence of publication bias, also produces a pooled estimate adjusted
for the estimated level of publication bias in a meta-analysis. This method
adjusts for publication bias by imputing studies suspected missing using
symmetry assumptions, and can be considered a form of sensitivity analysis.
This results of applying this method to meta-analyses in the Cochrane
Database of Systematic Reviews are presented, together with a comparison
between the various significance tests for publication bias.
Modelling Publication Bias
JOHN COPAS (University of Warwick)
Abstract: Correcting for publication bias is impossible without making
unverifiable assumptions. Equivalently, any model which explains publication
bias must include unidentifiable parameters. However, a sensitivity analysis
is possible, which can identify the range of estimates which give a good fit
to the observed funnel plot. Two recent systematic reviews about the health
risks of passive smoking will be discussed.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
