Among the reasons Sen. Charles Grassley, R-Iowa, has targeted Professor of Psychiatry and Human Behavior Martin Keller in his scrutiny of conflicts of interest in clinical research is Keller’s authorship of a controversial study of the antidepressant Paxil in children.
The study and its authors have been fiercely criticized in medical journals for allegedly misrepresenting data, suppressing information linking the drug to suicidal tendencies and reaching a conclusion unsupported by the relevant data.
The Herald contacted Keller several times starting Sept. 10, but he said he was unavailable for comment before press time.
Keller and his colleagues began looking into paroxetine – marketed as Paxil in the United States – in 1992, when they successfully sought funding for a study of the drug from its maker, SmithKline Beecham – now known as GlaxoSmithKline after a 2001 merger with GlaxoWellcome.
The study, which compared Paxil with another class of antidepressant and a placebo, ran from 1994 to 1997.
Referred to by GSK as Study 329, the study investigated whether Paxil was safe and effective for use in adolescent children suffering from depression.
Paxil is a popular antidepressant widely prescribed by physicians to adult patients, but few large-scale studies of its use in children were conducted before Keller’s.
The results of the study were compiled in 1998 and published in 2001 in the Journal of the American Academy of Child and Adolescent Psychiatry, which listed Keller as the lead author in front of 21 others.
It found that “paroxetine is generally well tolerated and effective for major depression in adolescents.”
But even as questions about Keller’s financial relationships with drug companies remain open, several members of the medical and legal communities have independently criticized his research on scientific grounds.
Articles in medical journals and letters to journal editors have called into question the scientific practices that led to Study 329′s conclusion that Paxil is a safe and effective way to treat depression in children.
Misrepresentation, profs say
In 2003, University of Adelaide physicians Jon Jureidini and Ann Tonkin wrote a strongly worded letter to the editor of JAACAP expressing their concern over what they felt was a conclusion unsupported by the data.
The letter argues that the criteria originally decided upon to determine whether Paxil was effective failed to show that Paxil was better than no treatment at all.
Specifically, improvements on the Hamilton rating scale for depression – a questionnaire that rates the severity of symptoms like mood, insomnia and anxiety – were not significantly greater for patients on Paxil than for patients on a placebo or a different drug, imipramine. Paxil patients did improve more than the other two groups, but the improvements in the “primary outcome measures” were too small to support the study’s conclusion that Paxil was effective.
In their response to the letter, Keller and seven of Study 329′s co-authors conceded that the primary outcome measures decided upon “did not reach significance,” but pointed to a number of “secondary outcome measures” – such as scores on other depression assessment tests – that did show a significant improvement.
Keller’s response added that in the time between 1993, when the relevant outcome measures were chosen, and 1998, when the study was written, the psychiatric community had moved away from the Hamilton scale and toward the measures that had turned out to be significant.
Though much of the medical community appeared to side with Jureidini and Tonkin – including the editorial boards of leading medical journals the Lancet and the Canadian Medical Association Journal – the disagreement appeared to have reached an impasse, which wasn’t bridged until this year.
In May, Jureidini and two other colleagues published a review of Study 329 in the International Journal of Risk and Safety in Medicine.
In the intervening time between the 2003 letter and the 2008 article, lawsuits against GSK by families of patients who claimed to have been harmed by Paxil gave Jureidini access to previously unreleased documents.
Those documents were subpoenaed by the law firm Baum, Hedlund, Aristei and Goldman, which compensated Jureidini for his work.
Among the documents Jureidini could access were drafts of the paper that had been submitted for peer review.
Before medical journals accept papers for publication, drafts are circulated among medical professionals to carefully scan the paper for miscalculations, careless errors, misleading figures and occasionally larger, more critical inaccuracies.
Jureidini learned that the paper had been rejected by the Journal of the American Medical Association, the world’s most widely circulated medical journal.
But, more crucially, Jureidini could now sketch out a timeline of the evolution of the paper, from the first draft submitted to JAMA in 1999 to the final draft published in JAACAP in 2001.
The drafts showed that none of the eight outcome measures outlined in the original 1993 protocols showed a significant positive effect for Paxil over placebo.
Instead, Jureidini argued that the authors had used “selective reporting” to produce their positive conclusion, and introduced new outcome measures in the 2001 paper, some of which were only decided upon after the blind was broken. That means researchers knew which data belonged to the Paxil groups before they chose a method for determining whether the drug was effective.
‘Commercially unacceptable’ results
Study 329 may have obscured critical findings on Paxil users’ suicidal tendencies using creative coding of results.
In the study, “suicidal ideation/gestures” were reported under the label of “emotional lability,” which Jureidini sharply called a “euphemism.”
In total, Study 329′s data showed that patients on Paxil were six times more likely to exhibit “emotional lability” than patients on placebo, and twice as likely than patients on imipramine, the other antidepressant tested in the study. Jureidini pointed out that the study failed to report the significance of those data, which were presented in a table but were not discussed in the text of the article.
Additionally, a GSK report on the study from 1998 identified at least three other patients who threatened suicide or had suicidal thoughts but were coded under “hostility,” “worsening depression” or even “euphoria” and so were not grouped with emotionally labile patients. Juredini’s calculations showed the serious adverse effects on Paxil to be significant over placebo.
“The misrepresentation about the effectiveness of the drug is something that can’t have been accidental,” Jureidini said.
The Herald has also obtained a copy of a 1998 GSK position piece marked “for internal use only,” in which the company reacts to the findings of Study 329.
The memo says that Study 329′s original standards used to measure drug efficacy were not met, but points to “trends in favour of paroxetine.”
The document goes on to say that competitors Eli Lilly and Pfizer were making significant progress in testing their respective antidepressants, and concludes with the overall goal to “effectively manage dissemination of these data in order to minimise any potential negative commercial impact … it would be commercially unacceptable to include a statement that efficacy had not been demonstrated.”
By contrast, the company’s promotional literature from 1999 trumpeted the efficacy of the product to physicians.
And a 2001 internal memo from Paxil’s product manager to sales representatives selling the drug asserted that Keller’s “cutting-edge” study showed that “Paxil demonstrates REMARKABLE Efficacy and Safety in the treatment of adolescent depression.”
‘Potentially confusing’ coding
There have also been more controversial allegations raised against Keller and his Study 329 co-authors, many of which appear in former Boston Globe reporter Alison Bass’s book, “Side Effects: A Prosecutor, a Whistleblower and a Bestselling Antidepressant on Trial,” which was released this summer.
In it, Bass claims that Keller coded some suicidal patients as “non-compliant,” which means that their adverse effects would not have been included in the article.
University officials would not release memos about the coding that Keller sent to the University’s Institutional Review Board, which could serve to confirm or refute Bass’s claims.
But Donna Howard, an assistant administrator under Keller in the mid-1990s, said she had personally faxed those memos to the IRB.
“There was two or three (instances) where there was severe adverse reaction that brought the child into contact either with the police or a hospital, and the children that were involved were eventually dropped from the study and coded as non-compliant,” Howard told The Herald.
She said the mood in her office at Butler Hospital would turn sour anytime a patient experienced a serious adverse effect.
“You would come in on Monday and it’s, ‘Oh no, patient so-and-so attempted suicide or hit a cop or overdosed,’” Howard said.
GSK’s Director of U.S. Media Relations Sarah Alspach denied the claim that Study 329 underreported serious adverse effects, citing a 2002 clinical review of the study by the Food and Drug Administration, which “found no statistically significant signal for suicidality,” Alspach wrote in an e-mail to The Herald.
Though the FDA review did not find a link in the study between Paxil and increased suicidal tendencies, it noted that researchers’ coding of adverse reactions was “potentially confusing.”
The report also concluded that Study 329 was a “failed trial,” because Paxil was not significantly more effective than placebo.
Two years later, the FDA issued a suicide warning for all selective serotonin reuptake inhibitors – the class of antidepressant to which Paxil belongs – after the British Medicines and Healthcare Products Regulatory Agency issued a similar warning earlier that year for paroxetine in the United Kingdom.
Then-New York Attorney General Eliot Spitzer sued GSK in June 2004, charging the company with consumer fraud and alleging that it had deliberately withheld data that showed that Paxil increased suicidal tendencies in adolescents. Spitzer’s case targeted only the pharmaceutical company and did not investigate Brown’s or Keller’s involvement. GSK settled the suit for $2.5 million two months later but admitted no wrongdoing.
As part of the settlement, GSK agreed to release all results from future clinical trials, whether negative or positive.
Jureidini said he is confident that data in Study 329 were deliberately misrepresented but added that he can’t definitively attribute the perceived misrepresentation of data to Keller.
“Whether Keller sat down with somebody and said ‘Let’s try to make this drug look good when really it’s not,’ is not something I have evidence for,” he said.
Charges of ghostwriting
The issue of blame is also complicated by allegations made by Bass and others that Study 329 was ghostwritten by a GSK-affiliated employee.
In any given clinical study with multiple authors, the actual article is usually written by one author and looked over by the others. In this case, there appears to be evidence that the article was drafted by Sally Laden, an editorial director for Scientific Therapeutics Information, the company that prepared the manuscript on GSK’s behalf.
Alspach denied the ghostwriting charges but acknowledged Laden’s contribution to the study, which was recognized with the line “editorial assistance was provided by Sally K. Laden” in the fine print of Study 329′s first page.
But a leaked memo appears to paint Laden as the one who did the vast majority of the writing. Referring to the first draft of the manuscript, a letter from Keller to Laden dated Feb. 11, 1999, begins “Dear Sally, You did a superb job with this. … It is excellent. Enclosed are rather minor changes from me.”
Clinical Associate Professor of Community Health David Egilman ’74 MD’78 said such work is unethical unless proper credit is given.
That credit is missing in Study 329, he said.
“In the third grade you’re told not to copy,” he said. “Now that we’re all grown up we’ve got a really long word for it, but it’s the same thing and it’s just as wrong.”
Egilman noted that ghostwriting is not an entirely uncommon practice in writing up clinical studies.
But it’s still considered unethical, he added, especially when the ghostwriter has a stake in the findings.
Lack of U. discussion
Each allegation of scientific misconduct in Study 329 – the claimed ghostwriting, outcome measures switch and miscoding of suicidal patients – have led scientists to call for Brown to lead a public investigation into the issues.
“What I find most distressing is that there has been absolutely no attempt at Brown to even discuss these issues, much less have a forum or an investigation,” said Clinical Associate Professor of Medicine Roy Poses ’73 MD’78.
“If Keller’s actions have been exemplary and that’s what everyone at Brown thinks, then they should go public and refute the charges.
If they can’t do that, the accusations are serious enough and they should address them.”
“When they find out that there’s been potentially ethically questionable research … they’re supposed to do a freaking investigation,” Egilman said, adding that Keller should be able to present evidence on his behalf.
Poses added that Brown’s lack of public discussion of the study, despite widespread interest, is potentially embarrassing for the University.
“Someone (at Brown) should be talking about it, because the rest of the world is talking about it,” he said.
Jureidini agreed that Brown should investigate Keller’s involvement. Study 329 “can only be interpreted as deliberately misrepresenting what was found,” he said.
A lawyer at Baum Hedlund – who is now representing the families of patients who claim to have been harmed by Paxil in class action lawsuits against GSK – said the University’s attitude toward his firm’s efforts has been “less than cooperative.”
Lawyers for the University, George Murgatroyd added, have “fought us every inch of the way.”
Murgatroyd said Brown claims it no longer had the memos that Keller had sent to the IRB, Brown’s research ethics panel.
The University, however, “can’t discuss particular cases of possible claims of wrongdoing and what we do about them,” Provost David Kertzer ’69 P’95 P’98 said – ending the possibility of a public inquiry.