The article by Ray Moynihan (1) cites one of the latest meta-analyses
conducted by an Italian research group which found no increased risk of
heart attacks. This meta-analysis, carried out by Mannucci et al. (2)
states that all-cause mortality during rosiglitazone treatment was
significantly lower than that observed with comparators, while a
nonsignificant trend was observed for cardiovascular mortality.
Conversely, no significant difference was detected in the incidence of
nonfatal coronary events and myocardial infarction during treatment with
either rosiglitazone or comparators. The incidence of serious CHF was more
than two-fold during treatment with rosiglitazone.
According to Moynihan, the article included two authors with ties to GSK
and its competitors, including the first and last authors.
Our review of this meta-analysis highlights also relevant limits
concerning five critical areas:
Firstly, the dataset provided by the authors is incomplete. There are
no raw data on the outcomes on Coronary Events and Acute Myocardial
Infarction; many records contain duplicate data. According to our
verification, the number of trials subjected to meta-analysis amounts to
167 trials and not 164.
Secondly, the authors never provide a formal analysis of
heterogeneity. We analysed heterogeneity through the Chi2 heterogeneity
test 4, and P values were found to be consistently higher than 0.10,
suggesting low levels of heterogeneity in the pool of data.
Thirdly, meta-analysis might not be the appropriate method to apply.
Random effect based techniques produce larger confidence intervals than
the outputs of fixed effect based techniques, which are more suitable in
the absence of heterogeneity (3).
Fourthly, the adjustment of results based on “incidence rates” is not
based on appropriate techniques. The number of patient-years calculated by
the authors by multiplying the number of patients by the duration of the
trial depends exclusively on the number of patients allocated to each arm,
while the observation period is implicitly given as constant (= the
duration of the trial)
Fifthly, in the majority of the trials subjected to meta-analysis the
duration of the follow-up is inadequate to detect the most relevant
outcome. The percentile distribution of follow-up periods in weeks
(median: 24 weeks of observation) demonstrates that as many as 90% of the
records had a follow-up of less than 1 year.
The meta-analysis by Mannucci et al. suffers from serious
methodological problems. The use of rosiglitazone is associated with a
significant increase in the risk of heart failure in very short follow-
ups. Important differences between rosiglitazone and the comparator might
not have been detected because of the excessively short periods of
observation. The work by Mannucci et al, published in a journal with a
low impact factor has been widely publicized in Italy, contributing to
playing down the risks associated with the use of rosiglitazone. In the
light of our conclusions, we therefore deem it important to water down the
message disseminated by the authors.
 Ray Moynihan: Rosiglitazone, marketing, and medical science. BMJ
 Edoardo Mannucci , Matteo Monami, Mauro Di Bari, Caterina
Lamanna, Grancesca Gori, Gian Franco Gensini, Niccolò Marchionni
Cardiac safety profile of rosiglitazone A comprehensive meta-analysis of
randomized clinical trials
Int J Cardiol (2009) doi:10.1016/J.iJcard.2009.01.064
 Systematic Reviews in Health Care – Meta-analysis in context
Egger M Smith GD, Altman DG BMJ Books 2001 ISBN 0-7279-1488-X
Competing interests: No competing interests