By Anna Wilde Mathews, The Wall Street Journal

Doctors and patients who rely on articles in prestigious medical journals for information about drugs have a problem: The articles don’t always tell the full story.

Some omit key findings of trials about a drug’s safety and efficacy or inconvenient details about how a trial’s design changed partway through. A study published in the Journal of the American Medical Association last year reviewed 122 medical-journal articles and found that 65 percent of findings on harmful effects weren’t completely reported. It also found gaps in half the findings on how well treatments worked.

Journals – Gatekeepers or Marketing Tools?

The problem calls into question whether journals can play the role of gatekeeper in an era when articles are increasingly used as marketing tools. Editors have “found themselves playing a game of research hide-and-seek,” says Jeffrey Drazen, editor in chief of the New England Journal of Medicine. They have “had experiences where authors tried to pitch it, where they were telling you the good news and not the bad news.”

Now some top journals are cracking down. This year BMJ, a leading British medical journal, started demanding that everyone who submits an article also submit the original study design plans, so that peer reviewers can see whether the authors changed the goalposts when publishing the study. It’s weighing whether to make the plans and the reviewers’ comments public.

“We want to make sure that we’re not misleading the public,” says Kamran Abbasi, deputy editor of BMJ.

JAMA and other top journals are also asking authors more frequently for their original study designs. Editors at JAMA sometimes call in independent statisticians to review the results. Several medical journals are also requiring that sponsors of clinical trials starting this July or later register details such as goals and size in a public database if they wish the results to be published.

The measures reflect a growing body of research about discrepancies between journal articles and the full results of the studies behind them. Journal editors are also responding to the escalating debate in Washington on ensuring drug side effects are properly disclosed. In the wake of the withdrawal of Merck & Co.’s painkiller Vioxx over cardiovascular side effects, some legislators are calling for tougher safety scrutiny of drugs on the market.

The JAMA study last year said articles often cherry-picked strong results to report, even if those results were in a different area than the study was designed to test. Typically scientists set up clinical trials to answer one or two primary questions — for example, whether a drug reduces the risk of a heart attack and stroke. These are called the primary outcomes. The JAMA study found that 62 percent of trials had at least one primary outcome that was changed, added or omitted.

“It was a shock to find that what we thought was the most reliable information wasn’t,” says Douglas Altman, an author of the JAMA study and director of the Center for Statistics in Medicine in Oxford, England.

One well-publicized dispute over data interpretation came in 2000 when a JAMA article said Pfizer Inc.’s painkiller Celebrex minimized damage to the stomach compared with older drugs. It later emerged that the authors used only six months of data even though they had some data extending to 12 months. When all the results were included, Celebrex didn’t look markedly better than its rivals. (Separately, a study last year suggested Celebrex might increase heart risk.) Today, the 2000 article is part of a shareholder suit alleging that Pfizer misled investors about its drug.

“We were burned very badly,” says Catherine DeAngelis, JAMA’s editor in chief. “The journal got tougher as a result. Today, “the single thing we change most often is the conclusion,” she says. “It comes in as, ‘This product is the greatest thing,’ and we say, ‘Under these circumstances, in this population, this medication seems to control a, b and c.'”

Celebrex Study

In an email, a Pfizer spokeswoman said the authors of the Celebrex study “presented the data in the manner that they felt was the most meaningful.” Many patients discontinued the treatment after six months, “making analysis of data beyond six months difficult,” she said.

Journal editors rarely see the complete design and outcome of the studies summarized in articles submitted for publication. A typical article is perhaps six or seven pages long, even when the research behind it took years and involved thousands of patients. Peer reviewers — other scientists who work voluntarily to review articles before they are published — also see only the brief article. They might fail to notice suspicious omissions and changes in focus, or, if they do, lack the time or inclination to follow them up.

The system relies, in essence, on a scholarly honor system. “Science depends on trust,” says Drummond Rennie, a professor at the University of California, San Francisco, and deputy editor of JAMA. “But if you have trust, you’re going to be fooled. You can’t have a policeman in every lab.”

JAMA Has Tightened Scrutiny

JAMA has tightened scrutiny of articles that are co-authored by academic researchers and industry scientists. To prevent the industry authors from slanting the data, JAMA demands that the academic authors, like corporate chief executives under Sarbanes-Oxley requirements, sign statements attesting that they have taken part in the data analysis and stand by their findings.

“I want an academician to put his or her reputation on the line, and that of the institution,” says Dr. DeAngelis, the JAMA editor in chief.

Disputes are rarely clear-cut. Scientists may legitimately disagree whether an article that leaves out a certain figure is deceptive or merely reflects the fact that no several-page summary of thousands of pages of data can be comprehensive.

Approval to Sell Mevacor Without A Prescription

As part of a bid for Food and Drug Administration approval to sell the anticholesterol drug Mevacor without a doctor’s prescription, Merck and partner Johnson & Johnson set up mock drugstores and solicited customers through advertising. The store shelves were lined with products including over-the-counter Mevacor. A label on the drug instructed potential users that they should take it only if they met several conditions, such as having moderately high cholesterol and at least one risk factor for heart disease. The idea was to simulate the real-life circumstances under which the pills would be sold.

An article summarizing the results of the experiment in the November 2004 issue of the American Journal of Cardiology said about two-thirds of the people who decided to try the drug met the conditions or came close. The authors, who worked for Merck and J&J, said the study’s full results made a “compelling case” that Mevacor was suitable to be sold over-the-counter.

In reviewing the case, the FDA highlighted another figure, one that never appeared in the article: Just 10 percent of the people who took the drug fully met the label’s conditions. The others included in the two-thirds figure met many of the conditions but not all. After hearing a presentation by agency officials, an FDA advisory committee in January voted to reject the drug companies’ request.

Edwin Hemwall, a vice president at the Merck-J&J joint venture that wants to sell the drug over-the-counter, says the label was conservatively written and the two-thirds figure accurately captured the percentage of users who were right for the drug. It included people who had been advised by their doctor to take Mevacor and some who were a year younger than the minimum ages on the label (45 for men, 55 for women). “We felt that that really represented, from a global perspective, the type of person who fit the label,” Dr. Hemwall says. The FDA, he says, “went very, very strictly by the label.”

The journal’s editor, William Roberts of Baylor University, said he didn’t remember details of the article and couldn’t comment.

Some editors say it’s impossible to sift through thousands of pages of raw data to check a paper’s fairness, given the cost and demands on the time of reviewers. Ultimately that’s the job of the FDA, says David Pisetsky, a Duke University professor who is the editor of Arthritis & Rheumatism. “People have to be realistic about what a journal can do,” he says.

Alan Goldhammer, an associate vice president at the Pharmaceutical Research and Manufacturers of America, the top drug-industry trade group, says some of the new scrutiny unfairly singles out drug companies — for example, by forcing them to get academic scientists to check off on their work. “When is enough enough?” asks Mr. Goldhammer. “Why are our submitted articles different from all other submitted articles?”

They’re different, journal editors contend, because of their potential use in marketing. In 2003, an article in the Journal of the American Geriatrics Society concluded that the Alzheimer’s drug Aricept helps elderly people live at home longer. It cited “significant delays” in the date when people who took Aricept entered a nursing home. Pfizer and Eisai Co., the drug’s co-marketers, ran ads in medical publications that cited the study and said the drug “helps keep patients in the community for more than five years.”

Critics, in letters to the journal, called the study “seriously flawed” and “misleading.” They suggested that its design tended to weight the Aricept group with the most compliant patients and those with the most social support — making it unclear whether their superior results had anything to do with the drug’s effects. Those in the non-Aricept group included people who refused the companies’ offer for free ongoing treatment with the drug and some who dropped out of an earlier clinical trial of it.

The study acknowledged the possibility of “selection bias” between the groups, but suggested this wasn’t a fatal flaw. David Geldmacher, the article’s lead author and a professor at the University of Virginia, says the study results were “meaningful” and the two groups were “comparable to a reasonable standard.” Thomas Yoshikawa, editor in chief of the journal, says the article was “topical and relevant” and its science “reasonably good.” The Pfizer spokeswoman said in an email that the company “stands by our advertising and the results of this study,” adding that they are consistent with a different study published this January in the same journal.

A 2001 analysis in JAMA found that side effects were adequately reported in only 39 percent of nearly 200 articles surveyed. The median space devoted to safety concerns was roughly a third of a page — about the same as the authors’ names and affiliations.

In 2001 the New England Journal of Medicine published an article about the Eli Lilly & Co. drug Xigris for sepsis, a body-wide response to infection that is often fatal. The article described Xigris as effective in a broad spectrum of patients. But four consultants to an FDA advisory committee later published a commentary in the journal, saying the FDA’s analysis showed the least-sick patients got no benefit and suffered side effects. The FDA approved the drug only for sicker patients.

William Macias, a Lilly official, and Gordon Bernard, a Vanderbilt University professor who was the lead author of the article, say the authors used a different statistical analysis than the FDA, and their method showed no meaningful differences between the subgroups. Dr. Drazen, the New England Journal’s editor in chief, defends the article, saying the main point was to tell readers the overall results.

“One solution to this is to publish the raw data” that emerge from a trial, says Dr. Abbasi of BMJ, the British journal. “The way things are going in terms of openness, you can’t rule it out.”

Moving Goalposts

The main goal of a study may change between the original design and final publication. What some authors do, according to a sampling of articles:

Describe original main goal as secondary: 34 percent
Fail to disclose original goal: 26 percent
Turn original secondary goal into main goal: 19 percent
Create new main goal: 17 percent
Source: “Empirical Evidence for Selective Reporting of Outcomes in Randomized Trials,” JAMA, May 26, 2004