Forest plots: trying to see the wood and the trees
BMJ 2001; 322 doi: https://doi.org/10.1136/bmj.322.7300.1479 (Published 16 June 2001) Cite this as: BMJ 2001;322:1479
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
The word 'logrank' in the headings of Figure 2 is a mistake caused by
us not changing the default settings of the plotting programme. I
apologise for any confusion.
The percentage of deaths for the Wilcox group should be 6.3, rather
than 8.9. There are a couple of errors like this, which were pointed out
at the proofs stage, but unfortunately they have not been corrected. If
anybody wants the corrected totals, then I can supply them.
However, as the paper is about where forest plots came from, rather
than about beta-blockers and mortality, these small errors should not
detract from the overall message.
Competing interests: No competing interests
Dear Sir,
In column 4 of figure 2 of the above article under the
heading of `beta blocker deaths' and above `observed-expected' the word
`logrank' appears.Surely this is inappropriate and confusing,or am I
missing something?
I should also have add that the percentage of deaths for the Wilcox
control group study should be 6.3 rather than 8.9.
Yours sincerely,
Alan C.C.Gibbs
No competing interests.
Competing interests: No competing interests
We agree with Lewis and Clarke (1) that it’s not easy to produce
forest plot graphs. It’s yet more difficult when we already have the
estimate point and confidence interval and just want to represent them in
a graph and wants to produce a forest plot directly. Even in revman, you
do not have how to do it. Microsoft Excel software is widely used, but
there is no way to produce a forest plot graph. Using some tricks in this
software, we were able to create forest plots directly by using stock
charts. This small sheet with instructions can be downloaded at
www.evidencias.com/forest01.xls for free.
1. Lewis S, Clarke M. Forest plots: trying to see the wood and the trees.
BMJ 2001;322(7300):1479-1480.
Competing interests: No competing interests
Esthetics may be in the eye of the beholder. But I wonder if others
who have gazed upon forest plots have shared my aversion to the sizing of
symbols for point estimates in proportion to their meta-analytic weight.
The expressed motivation for this special effect is to counteract a
presumably irresistible urge for the proverbial clinician's eye to be
drawn to the widest confidence intervals and thus to the least precise
estimates. To my admittedly jaded meta-analytic eye, however, the
distended blobs representing the low-variance estimates in these displays
connote not precision but its opposite.
I formulated a hypothesis, which I tested by showing a forest plot
with point estimate symbols of uniform size to my 38-year old sister, who
has trisomy 21. In response to my request to "find the shortest line,"
she instantly and unerringly homed in on the most precise estimate.
I offer this result as potentially generalizable to the proverbial
clinician.
Competing interests: No competing interests
We agree that funnel plots are an important way to investigate the
possibility of bias in a meta-analysis. However, an important advantage
of forest plots as a way to represent the results of a meta-analysis is
that they make it much easier to see the relative contribution of each
study. In addition, if a forest plot is sorted by study weight, it can
produce a visual approximation to a funnel plot without losing any of the
details about the component studies.
Both types of plot are important in meta-analysis. Indeed, we think that
Egger and Davey Smith (1) were suggesting that funnel plots should be used
in addition to forest plots.
(1)Egger M,Davey Smith G, Meta-analysis bias in location and
selection of studies BMJ1998;316:61-66
Competing interests: No competing interests
Reading this article on forest plots gave me a feeling of deja-
vu(1).A quick electronic search of past BMJs was rewarded by a paper on
funnel plots in 1997.(2) Funnel plots appear to be very similar to forest
plots.
Funnel plots are plots of effect estimates against sample size, and may
have an advantage over forest plots both visually and statistically.
Visual interpretation of a funnel plot is immediate, asymmetry of the
chart suggesting bias in the studies selected, and this can be further
explored by statisical analysis.
It has been suggested that funnel plots should be used routinely to
examine for the likely presence or absence of bias.(3)
Funnel plots were not mentioned in this article, but do the authors agree
that this is now the technique of choice in the display of results of meta
-analysis for the detection of possible bias?
(1) Lewis S, Clarke M. Forest plots:trying to see the wood and the
trees BMJ2001;322:1479-80
(2)M, Davey Smith G, Schneider M, Minder CE. Bias in meta-analysis
detected by a simple, graphical test. BMJ1997;315:629-34
(3)Egger M,Davey Smith G, Meta-analysis bias in location and
selection of studies BMJ1998;316:61-66
Competing interests: No competing interests
error in Figure 2
Figure 2 (the second forest plot) has two errors. The % deaths
figures in the first two trials (Wilcox, Norris) are incorrectly reported.
The % deaths figures for Controls in the trials should be 6.3% and
10.5%, respectively.
Competing interests: No competing interests