Skip Navigation

'Ghost' statisticians exert unseen influence

January 15, 2007 By Erika Check This article courtesy of Nature News.

Pharmaceutical employees are going uncredited on medical papers.

Please log in to rate this page.

View Comments

When a scientific study is paid for by a company with commercial interests, journals take great care to ensure that this information is properly disclosed. But are all the possible sources of bias being reported? A study released today shows that the statisticians in charge of crunching, and sometimes selecting, the data in a study often go uncredited, making their influence impossible to trace.

Peter Gtzsche, director of the Copenhagen-based Nordic Cochrane Centre part of the Cochrane Collaboration, which performs systematic reviews in health care worked with colleagues to analyse 44 trials that had been paid for by industry and published in various peer-reviewed medical journals. They compared the pre-trial plans for the studies, which were submitted to ethical review boards in Copenhagen and Frederiksberg, with the resulting published studies.

In 33 cases, the names of people who made "substantial contributions" to the studies were left off the published papers, they found. In 31 trials, these so-called 'ghost' authors were statisticians experts who analyse trial data.

The study, published today in the journal Public Library of Science Medicine1, is the first to compare a large number of trial plans with their results in order to shed light on the issue of ghost authorship.

"A lot of us have had suspicions that there is a high prevalence of ghost authorship, but until very recently nobody has been able to measure it," says Liz Wager, an independent consultant from Buckinghamshire, UK, who has drafted publication guidelines for the UK-based Committee on Publication Ethics amongst others. Getting the data upon which to conduct such studies has proven very difficult, she notes.

Gtzsche and his colleagues had originally wanted to perform an analysis of UK studies, but lawyers blocked them from obtaining trial plans in that country.

The results of Gtzsche's successful study are worrying, Wager says. "If the overall involvement of the sponsor company is being played down, it's not a true picture that's being given."

Behind the scenes

The practice of leaving a paid writer's name off a list of study authors has recently gained public attention. Professional writers' and editors' associations have, as a result, drawn up guidelines that discourage this practice (see ' Drug trials: Stacking the deck').

But leaving a statistician's name out of an author list is also problematic, says Gtzsche.

"You might say 'no, I don't want to waste my time reading papers written by pharmaceutical employees', but you can't do that when there are authors who are kept secret," Gtzsche says. It is possible that a company may recruit independent, academic researchers to gather data, for example, and then have an un-named employee acting as statistician.

This is of particular concern because statisticians make important decisions about how to analyse information gathered by doctors during a clinical trial such as how to interpret a doctor's written notes on a particular patient, Gøtzsche says.

"There are so many judgements that get made and there are so many statistical methods you can use, so there is quite a lot of subjectivity," Gtzsche says.

For just this reason, in 2005 the influential Journal of the American Medical Association enacted a new requirement for companies wishing to publish the results of clinical trials. If the trial data have only been analysed by a company-employed statistician, the data must be re-analysed by an independent statistician before the journal will publish the results of the trial2.

Visit our newsblog to read and post comments about this story.


  1. Gtzsche P. C., et al. PloS Med., 4. e19 (2007).
  2. Fontanarosa P. B.& Flanagin C. D. J. Am. Med. Assoc., 294 . 110 - 111 (2005).


User Tools [+] Expand

User Tools [-] Collapse

Pinterest button


Please log in to add this page to your favorites list.

Need Assistance?

If you need help or have a question please use the links below to help resolve your problem.