Related Articles
Press One for Safety
Assessing Safety in Clinical Trials
Data Analytics Takes on the Unexpected
Confusion over how to interpret the data that come out of many clinical trials is not a new phenomenon and sometimes seems
endemic to the process. Most recently, concern over the safety of FDA approved medications has been fueled by a seemingly
steady stream of examples, with GSK's diabetes drug Avandia among the most recently profiled. This confusion emanates from
multiple sources—manufacturers, academic journals, mass media reporting—and has far reaching negative consequences for much
needed informed public discourse about new and existing medications.
A recent and widely publicized meta-analysis published in the June 14, 2007, issue of The New England Journal of Medicine by Nissen and Wolski1 raised concerns about GSK's diabetes drug Avandia (rosiglitazone). The finding that received the most attention was a suggestion
that it is linked to an increased risk of myocardial infarction (MI). Headlines and conclusions to emerge from this article
almost universally ran along the lines of "43% more heart attacks with Avandia"—and this reporting was widely echoed across
mass media publications, from The New York Times2 and Wall Street Journal3 to online outlets such as http://MedicineNet.com/,4 as well as medical journals such as The Lancet.5 Even the FDA entered the fray, calling for hearings and committees to examine Avandia's safety. While we don't seek to minimize the concern expressed over this issue, what is largely missing from ongoing discussions about
Avandia's safety is consideration of absolute risk levels—a topic perhaps less headline worthy, but nonetheless important
as part of the overall clinical picture.
To understand the source of the concern and confusion about Avandia, one must first consider the primary statistical metric
employed in the Nissen and Wolski meta-analysis, something called an odds ratio. Widely used in epidemiology, odds ratios
are generally regarded as difficult to interpret and apply in medical practice. So, what exactly does an odds ratio tell us?
A literal definition is "the odds of an event in the active treatment group (P1) divided by the odds of an event in the control
group (P2)"6 as expressed in the following formula:
Table 1. Odds ratio calculations for hypothetical product efficacy.
|
Although often incorrectly taken as such, an odds ratio does not correspond to the probability of an event occurring or typically
to the relative risk between groups (i.e., the ratio of two probabilities). A brief and simple illustrative example helps
to clarify how an odds ratio can create confusion around research findings. As shown in Table 1, treated patients are 4.5
times more likely to show improvement than control patients (90% ÷ 20%). However, these same data can also be expressed as
an odds ratio of 36 (by dividing 9 by .25).
Although the calculation of this odds ratio is clear enough, its interpretation is certainly less so on any intuitive level.
Treated patients in this example are clearly not 36 times more likely to show improvement than control patients. Instead,
the proper—albeit puzzling—interpretation is that the odds of treated patients showing improvement are 36 times greater compared
to the odds of control patients. It seems quite fitting that the odds ratio has been called "a stranger to both physicians
and gamblers, but a friend of many biostatisticians and epidemiologists."7
