Don’t panic

The FDA was well aware of the cardiovascular data with rosiglitazone. The cardiorenal advisory board has a calm and sedate head, unlike Steve Nissen.

An editorial published online May 23, 2007 in the Lancet weighs in on the debate, urging calm [3]. It points out that the two most reliable studies to inform decision-making are ADOPT and DREAM. The DREAM study showed MI rates of 0.6% for rosiglitazone group vs 0.3% for controls, and the MI/stroke/cardiovascular composite occurred in 1.2% of rosiglitazone vs 0.9% of control patients, with neither result reaching statistical significance. The only significantly relevant finding in the ADOPT trial was an excess of congestive heart failure episodes for rosiglitazone-treated patients compared with glyburide (22 vs 9 events), the editorial adds.

“Taken together, these results, although based on very small numbers of events, certainly raise a signal of concern and indicate the need for more reliable information about rosiglitazone’s safety. But the FDA, physicians, and patients can reasonably await the results of RECORD, a phase 3 trial designed specifically to study cardiovascular outcomes. Until the results of RECORD are in, it would be premature to overinterpret a meta-analysis that the authors and NEJM editorialists all acknowledge contains important weaknesses. To avoid unnecessary panic among patients, a calmer and more considered approach to the safety of rosiglitazone is needed. Alarmist headlines and confident declarations help nobody,” the Lancet editorial concludes.

Inaccurate estimate of risk
On the limitations of Nissen’s analysis, Haffner explained that one of the main concerns he has is the fact that it used a statistical technique that weighted the studies by the number of subjects included. “This would be acceptable if the studies were of a similar length, but the analysis included two large-scale studies with long-term follow-up and 40 smaller studies with much shorter follow-up. Under these circumstances, the smaller studies have been overweighted,” he said.

Haffner added that if the 43% increase in risk shown in the analysis had occurred in the ongoing RECORD study with rosiglitazone, it would have been stopped a year ago. “I think if a proper meta-analysis were done with rosiglitazone, it would probably show some increased risk, but this would not be significant. The hazard ratio would not be 43%—but it may be 30%. That doesn’t mean it has a clean bill of health, but it doesn’t justify pulling it off the market, either. It’s ludicrous to suggest taking a drug off the market based on flawed data showing p values of 0.03 and 0.06.”

Strom also feels there has been an overreaction to these data. “Nissen did a good job with the data he had, but this meta-analysis is very far from being definitive. If this is all the evidence they have for harm, then people are overreacting. Yes, this meta-analysis is important—but it should be viewed as raising questions, not providing answers.” Alarmist headlines and confident declarations help nobody,

Strom makes the point that a meta-analysis cannot deal with studies in which no outcomes were seen, which is why such studies were excluded, but from a safety point of view, these studies give important information. He also highlighted the lack of individual data as a huge drawback. “They recognized this and stated it as a limitation in their paper, but you can’t do the correct analysis without the individual data, and maybe they shouldn’t have tried. They ended up doing the wrong analysis. These are very borderline findings—p values of 0.03 for MI and 0.06 for CV death, and doing a different analysis could easily have generated a different answer. Their findings are therefore only suggestive at most.

“I’m not panicking about this. I think these drugs are overused, but I do have some patients on rosiglitazone, and I will not take them off it just because of this study. It is an interesting hypothesis but far from conclusive,” Strom added.

What is the absolute increase in risk?
McGuire believes a cardiovascular safety signal does exist within the rosiglitazone data set, but he adds that “to present the data as point estimates of risk without the clear context of the absolute signal is irresponsible and alarmist.” He explains that the worst-case scenario from these data is that rosiglitazone would elevate risk for MI from 1.4% to 2%. “So, a safety signal—yes. An emergency call to remove the drug from the market and cause global chaos among patients taking the drug? No. At least three large-scale trials currently under way will have much more robust power to evaluate these effects. With this small safety signal, we can afford to wait for the additional data being accumulated.”

Dr Storm sums it up: “Drugs are given because they have benefits. Where there’s a scare, people stop taking them. So you can do more harm than good if the scare is not based on correct information. I do believe this meta-analysis should have been published, but a less sensational editorial would have been better!