THURSDAY, Nov. 10, 2005 (HealthDay News) -- The recent withdrawal of the blockbuster painkillers Vioxx and Bextra begs the question: Could researchers have uncovered heart risks linked to these drugs earlier?
According to a new report in the December issue of Health Affairs, the use of a more detailed method of analyzing clinical trial data might have made the difference.
The study found that, of 108 high-profile trials reported in three top medical journals in 2001, just 4 percent employed this well-known methodology. Called multivariable or risk-stratified analysis, the sophisticated method sheds light on hidden treatment risks and benefits that do not always emerge under standard analysis.
"What you're trying to say is, 'Are there high-risk patients who would benefit much more by a treatment?' Or, turning it over to the safety side, as with Vioxx -- could you identify patients at high risk for a heart attack, for example, because they are smokers or overweight? That's what you're trying to do when you do this risk-stratified analysis," explained Stuart Pocock, a professor of medical statistics at the London School of Hygiene and Tropical Medicine.
Although he was not involved in the study, Pocock agreed with its basic premise: That those who design and publish clinical trials need to include risk-stratified analyses in their results. This type of analysis helps identify subgroups of patients with characteristics that could make them especially prone to either a benefit -- or harm -- from a particular drug.
Many people understand the structure of the typical randomized, controlled clinical drug trial: Give group A a new drug and Group B a harmless placebo, then follow their health outcomes for a specific amount of time.
This type of standard analysis can give a broad picture of a medication's overall safety and effectiveness, experts say. But too often risks for particular patients -- such as the cardiovascular trouble seen in a subset of Vioxx users -- can go unnoticed.
A treatment's benefits can be exaggerated, too. For example, if a small percentage of study participants see their health greatly improve after using a medication, standard analysis spreads that benefit over the group as whole -- even though many will have experienced no change in their health status.
In one prime example, researchers in 1993 used standard methodology to analyze data on the use of the clot-busting drug tPA for patients admitted to a hospital with a heart attack.
The study showed that, overall, people fared much better if tPA was administered soon after admission.
"However, this treatment also has a risk of harm in that it can cause people to bleed, have serious strokes and even die," said the lead researcher of the Health Affairs study, Dr. Rodney Hayward, a professor of medicine and public health at the University of Michigan, in Ann Arbor.
Noting this tPA-linked stroke risk, another researcher re-analyzed the original tPA data years later. This time, he used risk stratification to factor in characteristics that raised patients' risk for bleeding.
"For people with high risk for heart attack, the treatment is very beneficial," Hayward said, "so that any risk of stroke is more than offset. But there was also this population who had risk factors for bleeding, and a much lower risk of heart attack."
The data re-analysis confirmed that these high-risk patients "should not receive tPA, from a safety standpoint," Hayward said. "And it's only by doing the multivariable risk-stratification that you could detect that."
So, if risk-stratification is such an effective means of uncovering patients who might be harmed by a particular drug, why isn't it used in every clinical trial?
"First of all, some of these models have only been developed over the past 15 years," said Hayward, who is also director of heath services research at Ann Arbor Veteran's Administration Health Care System.
"Then there are very human reasons -- researchers have an inherent bias of wanting to view their results as positively as possible," he said. " If you've spent the past 20 years of your life trying to prove that something works, are you going to analyze the data with something that's [more] complicated and less familiar to you, to show that it didn't work in a lot of the people?"
Drug companies also see little reason to use risk-stratification analysis, Hayward said. Standard analyses that spread a medication's benefits for a select few across the group as a whole "make it look like everybody's benefiting at least a moderate amount, and therefore maybe 10 million people should have this medication," he said.
Again, a deeper statistical analysis focused on patients at specific levels of risk would probably show that that's just not the case, he said.
"I don't think blaming industry for this is the right thing to do, though, because the responsibility for setting standards rests on the profession and the journals' editorial boards," Hayward said. "It's up to editorial boards and professional organizations to step up and require this."
Pocock believes both standard data analysis and more detailed risk-stratification methods should co-exist. "Any trial should first report the overall result," he said. "And then you explore secondarily what the benefits or harms are by different risk strata."
But does this mean the average overworked doctor should crunch a complex series of numbers before he writes a prescription, or orders a particular procedure?
Hayward said that for really big decisions -- a chemotherapy regimen or the use of a $10,000 drug -- weighing individual patient risk factors into the balance is probably a good and ethical idea, "because so much is at stake."
But for more routine care, a simple computerized model might be more appropriate. "The clinician might punch in the factors -- 'This is a woman, her cholesterol level is this, this is her age' -- and it might say, 'This person will get a high benefit from the treatment, so definitely treat,' " Hayward said. "Or, 'Only modest benefit -- patient-doctor to share decision.' "
"I think most patients just want to get a general idea. They don't want a bunch of numbers," he said.
More information
To find out more on how clinical trials are conducted, head to the U.S. National Institutes of Health.
