Researchers have long worried about “publication bias,” the tendency for negative results of clinical trials never to appear in medical journals. Now a team of researchers has demonstrated the bias in studies of antidepressants: In the research literature, it appears that 94 percent of trials had positive outcomes, but in data submitted to the Food and Drug Administration, just 51 percent did.

In a paper appearing in the new issue of The New England Journal of Medicine, the researchers note that the result of such bias is that antidepressants appear more effective than they actually have proved to be.

Pharmaceutical companies must provide the FDA with data from all studies they perform on a drug they submit for approval, including clinical trials with negative outcomes. The researchers, led by Erick H. Turner, of Oregon Health and Science University, collected reviews from the FDA about 12 antidepressant drugs, which were tested in 74 studies involving more than 12,500 patients from 1987 to 2004.

Dr. Turner’s team found that of the 38 studies that the FDA considered positive, all but one had been published. But of the 36 that the FDA considered either negative or questionable, 22 were not published, and 11 were published with positive conclusions.

The result was that 48 of the 51 published studies appeared to be positive, but only 38 of the 74 reported to the FDA produced positive results. At least one study was unpublished or was published with results contrary to the FDA’s judgment for every one of the 12 antidepressant drugs the researchers considered.

Dr. Turner’s group was unable to determine the underlying cause of the bias: whether study sponsors and researchers were failing to submit articles to journals, or journals were turning down manuscripts that reported negative results. —Lila Guterman