Bias in double-blind trials (doctoral thesis)

By Peter C Gøtzsche

In 1990, I defended my doctoral thesis at the University of Copenhagen. It was the first time a whole therapeutic area had been scrutinized with statistical methods, including meta-analyses. I had collected all trials that had compared one non-steroidal, anti-inflammatory drug with another and demonstrated that the design and analysis of the trials were generally flawed, and that the published results were too good for the sponsoring company to be true.

In 82 of the 196 examined reports (42%), bias in the conclusion or abstract consistently favoured one of the drugs, which was the control drug in only one report and the new drug in the remaining 81 (p =3.4 x 10-23). However, the truth was that none of the drugs was clearly better than any other. 

My findings shocked the examiners to such a degree that they proposed that arthritis drugs was a particularly bias-ridden area, but subsequent results in other research areas have shown that my findings can be generalised to trials of other types of drugs.

My thesis became widely known and was the reason that Sir Iain Chalmers contacted me when it came out and invited me to start the Cochrane Collaboration together with him and about 80 other people in 1993. Now, 33 years later, I still feel it is an important document for the history of medicine. I have therefore made the thesis publicly available.