Day One: Data Deluge at FDA Rosiglitazone Review Provides More Questions Than Answers

Shelley Wood

July 14, 2010

July 14, 2010 (Gaithersburg, Maryland) — Long predicted to be the most exciting FDA meeting of recent memory, day one of the rosiglitazone (Avandia, GlaxoSmithKline) safety meeting opened with a bang and ended with a whimper after an 11-hour data deluge from 19 separate presentations.

FDA Commissioner Margaret Hamburg provided the bang by addressing the panel at the outset--something that, by her own admission, is not typically done at advisory committee meetings but warranted in this case, given the sheer "amount of attention this meeting and this issue have been receiving."

Then, setting the tone for the rest of the day, Hamburg succinctly stated what many observers have already noted: "For some advisory committee meetings, all of the FDA opinions line up nicely with each other, the issues are clear, and all the agency scientists are in agreement. And as you know, that is not the case today."

What followed was layer upon layer of scientific study, with presentations from GlaxoSmithKline employees, outside experts, and FDA scientists and statisticians, the rare candid opinion offering the only respite from the barrage of data, much of it seemingly contradictory. Presenters were frequently warned that they were running out of time, and the few question periods sprinkled throughout the day were often cut short, with many members of the panel staying silent for the entire day, either dumbfounded or slow to gather their thoughts before the next presentation began.

Early in the day, presentations on behalf of the sponsor restated GlaxoSmithKline's position that there is no reliable signal of increased risk of mortality or MI with rosiglitazone and that only a randomized controlled trial will provide the answers.

Things Heat up

Next, things warmed up considerably with presentations from Dr Steven Nissen (Cleveland Clinic, OH), who called upon members of the panel by name, reminding them that as clinical trialists they should reject the RECORD results on the grounds of poor design and conduct. "Clinical-trial conduct has to count for something. . . . We cannot and should not allow data of this quality to affect decisions that affect lives."

The FDA's own Dr Thomas Marciniak, saying he "likes to get down and dirty," provided details on what he called "mishandling" of adverse events in the RECORD trial. He also pointed to two critical changes to end-point definitions--to exclude silent MIs and to count all unknown deaths as cardiac--made after the company had access to patient code data. His conclusion: RECORD cannot be trusted to provide trustworthy data on MI or cardiovascular deaths.

"What I think all this is suggesting is that from the study conduct biases, any CV estimates of risk you get from RECORD should not be viewed as precise statistical estimates, but as lower bounds," Marciniak concluded. The true CV risks are "likely to be higher."

But immediately after Marciniak's presentation, his colleague at the FDA's Center for Drug Evaluation and Research (CDER), Dr Ellis Unger, told panel members that while he respected Marciniak's work and "tenaciousness," he himself saw no statistically increased risk of MI, no matter what analysis of RECORD was used, and suggested the best data to look at were those for all-cause mortality, which are "more difficult to fudge."

During questions, however, panel member Dr Nancy Geller (National Institutes of Health, Bethesda, MD) took issue with Unger's position, saying, "fudging" data isn't difficult, not "if you report the wrong follow-up date, and not if you withdraw patients from the trial prior to their deaths," to which Unger conceded, "fair enough."

Then again, in further FDA presentations, Marciniak's revelations seemed to lose their luster, with FDA medical officer Dr Susan Leibenhaut presenting results of the FDA's inspections of RECORD sites and conduct. Pressed to explain whether or not her inspection turned up any kind of trial violations out of the ordinary, Leibenhaut hedged, finally saying only that they were "typical of what we can find" and "not considered serious violations."

This kind of back-and-forth positioning epitomized the FDA presentations throughout the day, particularly those concerned with the conduct, potential bias, and final results of RECORD, leaving some panel members to ask just what they were to make of the to-and-fro.

"Identifying bias is one thing, but implying malfeasance is another," panel member Dr Sanjay Kaul (Cedars Sinai Medical Center, Los Angeles, CA) pointed out.

Seeing the Forest for the Plots

Late in the afternoon, panelists heard more from the FDA, most notably the incendiary data from the FDA's Drs Kate Gelperin and David Graham, plus analyses from FDA statistical reviewers. Gelperin's presentation, in particular, bristled with spiky forest plots, the upshot of which showed AMI, congestive heart failure, and mortality to be higher in rosiglitazone-treated patients than in pioglitazone (Actos, Takeda Pharmaceuticals)-treated patients. "In my view, it's highly likely that rosiglitazone therapy is associated with increased risk of adverse cardiovascular outcomes," she concluded.

Graham, meanwhile, presented the results from his recent Medicare analysis, then took an additional 15 minutes to present his "personal perspective," slamming the RECORD trial, saying that if it appeared before an advisory panel for a new approval of a drug it would be rejected as "garbage."

"You can't trust it. And if you trust it, you are engaging in the willing suspension of disbelief."

Graham also decried what he called the "unethical" premise of the ongoing Thiazolidinedione Intervention with Vitamin D Evaluation (TIDE) trial, which was designed based on the premise that rosiglitazone had increased risks. What ethical physician would enroll a patient in such a trial? Graham asked. "The presence of informed consent does not render ethical a study that is intrinsically unethical," he noted.

Finally, Dr Fiona Callaghan (CDER) presented the FDA's own meta-analysis of rosiglitazone--an update of its 2007 analysis--concluding that while overall major adverse cardiovascular event (MACE) rates miss statistical significance, rates of serious myocardial ischemia and total myocardial ischemia now reached statistical significance. A separate FDA meta-analysis of pioglitazone, presented by Dr Bradley McEvoy (CDER), showed no comparable signal of these kinds of problems.

Presentations from BARI-2D and VADT rounded out the day, both showing no evidence of risk with rosiglitazone. In the face of voluble criticism by mathematically minded panel members, presenters of both sets of trial data admitted that their analyses were "seriously flawed," at least for answering the questions at hand.

A Handful of Questions

Just a handful of the panel members asked regular questions throughout the day. Dr Marvin Konstam (Tufts-New England Medical Center, Boston, MA) regularly returned to the question of whether all of the risks of rosiglitazone were driven by comparisons against placebo and not against active controls. "I think later on we should really get into what the implication of that is, vis-à-vis, you know, [other] drugs that are currently on the market for the management of diabetes," he said.

Dr Thomas Fleming (University of Washington, Seattle), likewise, piped up frequently with hard questions on the statistical rigor of the analyses, saving his most searing criticisms for the analyses from the BARI-2D study.

Kaul also critiqued study methodology at points throughout the day but let fly with his biggest salvo late in the afternoon, when he questioned Nissen about how his meta-analysis dealt with zero-event trials, a point he hammered three years ago in his notorious rereview of Nissen's 2007 paper.

"Such fragile data are deserving of frugal interpretation, not liberal interpretation," Kaul said. "And the regulatory and clinical implications of these interpretations could be profound; hence caution is warranted before drawing definitive conclusions."

Nissen, turning the tables, said he fully agreed: "You're absolute right, the data are fragile, and the question is, whose fault is that? We have a drug that's been on the market for 11 years. The company has had every opportunity to do large outcomes trials, adequately powered and properly run to answer this question. They didn't do it."

For the full minute-by-minute from the day's hearings, read's account on twitter, and follow our ongoing coverage Wednesday.