Digging for data on harms in duloxetine trialsBMJ 2014; 348 doi: https://doi.org/10.1136/bmj.g3578 (Published 05 June 2014) Cite this as: BMJ 2014;348:g3578
- Peter Doshi, assistant professor1, associate editor2,
- Julie Zito, professor1,
- Susan dosReis, associate professor1
- 1Center for Drug Safety, Department of Pharmaceutical Health Services Research, University of Maryland School of Pharmacy, MD, USA
- 2The BMJ, BMA House, London, UK
The biomedical research enterprise devotes enormous amounts of time and money to determining how well a medicine works. By contrast, few resources focus on discovering the harms a given drug may cause. Undoubtedly, this is in part because we all—patients, doctors, even drug regulators—crave good news, and ascertaining an accurate profile of drug induced harm only threatens to bring bad news. Even the randomized trial, our so called gold standard, with all its potential for settling debates about drug efficacy, is notoriously poor at detecting serious drug related harms. Some reasons for this are well known: trials are usually too small, too short, or enroll participants who are “too healthy”—which collectively means that relatively uncommon, but none the less real and possibly serious side effects may not occur with sufficient frequency (or at all) during a clinical trial to demonstrate convincing evidence of harm.
Despite their limitations, however, randomized trials remain a critical element in assessing potential harms, especially as randomization assures serious attention by researchers interested in the “causality” game. Although a single study may be underpowered to detect important harms, combining studies through meta-analysis may show important associations. But credible use of trial data requires confidence in the quality of the reporting …