Intended for healthcare professionals

CCBYNC Open access

Guillain-Barré syndrome and adjuvanted pandemic influenza A (H1N1) 2009 vaccine: multinational case-control study in Europe

BMJ 2011; 343 doi: (Published 12 July 2011) Cite this as: BMJ 2011;343:d3908

Rapid response to: Guillain-Barre syndrome and adjuvanted pandemic influenza A (H1N1) 2009 vaccine: multinational case-control study in Europe

Corresponding Author : Felix Arellano (

In their paper on Guillain-Barre Syndrome (GBS), Dieleman et al.1
concluded that "the risk of occurrence of Guillain-Barre syndrome is not
increased after pandemic influenza vaccine, although the upper limit does
not exclude a potential increase in risk up to 2.7-fold". This
conclusion, based on the upper limit of the 95% confidence interval (CI),
is misleading. Using the same reasoning as the authors, one could also
conclude that it can "not exclude" a potential decreased risk of 0.3-fold
either, which is not stated in the authors conclusion. In fact, 95% CI do
"not exclude" any value in the reference population but estimate the range
of values where the population value is likely to be included but not all
values within the CI are equally likely to be found in the reference
population. The values at the upper and lower limits are much less
probable than those close to the point estimate. The very small likelihood
that the upper or lower limits of the CI represent the population value
was not quantified by the authors in their conclusion. Therefore,
focusing on the upper (or lower) limit gives a distorted view.

When computing a CI, we estimate the range of values within which the
population parameter (i.e., odds ratio [OR] in this case) under study is
most likely included. The estimated OR of GBS in this study is 1 with an
estimated 95% CI ranging from 0.3 to 2.7. This interval suggests that we
are 95% confident that the true population OR will be between 0.3 and 2.7;
however, the data are more compatible with a true OR close to the point
estimate (1 in this case) rather than a true OR close to one of the CI
boundary estimates (0.3 or 2.7). To illustrate this, the "p-value
function" (or the confidence curve) provides graphical representation of
the p-values for every alternative to the null hypothesis or, in other
words, the entire set of confidence intervals for different levels of
confidence 2 3. As an example, considering a data set similar to the
published results, plotted as a p-value function or confidence curve
(Figure 1), it can easily be seen that a true value of 1 is more likely to
reflect the observed data than one close to the boundaries of the 95% CI.

The potential for misinterpretation of confidence interval boundary
values that reflect low probability events should not be underestimated.


1.Dieleman J, Romio S, Johansen K, Bonhofer J, Sturkeboon M. Guillain
-Barr? syndrome and adjuvanted pandemic influenza A (H1N1) 2009 vaccine:
multinational case-control study in Europe. BMJ 2011;343:d3908 doi:

2.Rothman KJ, Greenland S. Approaches to Statistical Analysis. In, Rothman
KJ, Greenland S. eds. Modern Epidemiology. Philadelphia: Lippincott-Raven,
Philadelphia, 1998:183-199.

3.Ahlbom A. The p-value, the p-value function and the confidence interval.
In: Biostatistics for Epidemiologists. Boca Raton, Florida: Lewis
Publishers, 1993:35-53.

Competing interests: The authors are emloyees of GlaxoSmith Kline Biologicals and may own stock of the company

12 September 2011
Felix M Arellano
Francois Beckers
GSK Biologicals