It turns out that 90 percent of the "very large" effects described in initial reports on medical treatments begin to shrink or vanish as more studies are done.

"If taken literally, such huge effects should change everyday clinical and public health practice on the spot," Dr. John Ioannidis of the Stanford School of Medicine in California told Reuters Health by email. "Our analysis suggests it is better to wait to see if these very large effects get replicated or not."

Ioannidis has made headlines before with research showing that studies in medicine are often contradicted by later evidence, a phenomenon that has been referred to as "the decline effect."

The new work, published in the Journal of the American Medical Association, pools more than 3,000 research reviews done by the Cochrane Collaboration, a prestigious international organization that evaluates medical evidence.

Nearly one in 10 of the analyses in those reviews showed a very large treatment effect - harmful or beneficial - in the first published trial. But usually the reports trumpeting astounding findings were based on small, less reliable experiments.

There are several possible explanations, but smaller experiments are generally more likely to yield extreme results by chance alone. As more data accumulate, they start to approach the average - something statisticians call "regression to the mean."

Out of the tens of thousands of treatment comparisons they looked at only one - a respiratory intervention in newborns - showed a reliable, very large drop in death rates.

So taking new research with a grain of salt may be appropriate, according to Ioannidis.

"Keep some healthy skepticism about claims for silver bullets, perfect cures, and huge effects," he advised.

The team also found that trials showing big effects more often looked at lab values such as bone density or blood pressure than did those with more moderate findings.

Such lab values are tied to health outcomes - say, bone fractures and strokes - and are easy to measure. They are also easier for drugmakers to target than the health outcomes themselves.

But that doesn't necessarily mean drugs that tweak your numbers will have the health effects you are looking for, said Dr. Andrew Oxman of the Norwegian Knowledge Centre for the Health Services in Oslo.

"Even if they reduce the lab value, you can't be sure they reduce your risk of heart attack or stroke or fracture," Oxman, who wrote an editorial about the new findings, told Reuters Health. "There are lots of examples where things start to be used and have entered the market based on surrogate outcomes and then actually proved harmful."

He mentioned the heart rhythm drugs encainide and flecainide, which for many years were given to people with acute heart attacks. But then trials showed they were actually bad for these patients.

"These drugs were by given well-meaning clinicians, but they actually killed more people than the Vietnam War did," Oxman said.

He said people should understand that while medicine can be lifesaving and reduce human suffering, it usually comes with side effects and the benefits are often modest.

"A lot of things don't have these slam-bang effects, so people have to look at the trade-offs, the harms," he said.

SOURCE: bit.ly/PoG8om and bit.ly/PoGkEf Journal of the American Medical Association, online October 23, 2012.

• A Permanent Economic Emergency