Your doctor's best advice for a splashy new medical study might be to "take it with a grain of salt and call me in a decade." A review of 49 recent major medical studies found that nearly a third were either contradicted entirely or found to describe a smaller effect than was originally thought. The studies were chosen based on how often they'd been cited in later medical journal articles. A sampling:

• A 1991 study concluded that taking hormones after menopause was probably good for protecting women's hearts from coronary artery disease. Three years ago, an investigation designed to address that question famously showed just the opposite was true.
• A study based on a small number of trial participants, published in 1991, found that giving a patient a particular antibody could reduce the patient's risk of dying from some infections. A later, larger trial found that the antibody didn't help at all.
• A 1995 study found that giving recombinant tissue plasminogen activator (rt-PA) to stroke victims greatly aided their recovery, while later trials have found that rt-PA is useful only in some cases.

Not all studies, of course, are likely to be contradicted. For example, one 1998 review found that intensively managing Type II diabetes reduced a patient's risk of complications. That effect was so strong that it would be unethical to test it in a randomized trial where some patients' diabetes was less intensively controlled, according to clinical epidemiologist John Ioannidis, who carried out the analysis of these studies. Ioannidis is a professor at the University of Ioannina in Greece.

Ioannidis's findings appeared in last week's Journal of the American Medical Association, which also published many of the 49 studies he examined. He admits that, yes, in a few years another study could overturn his results, too: "Science is never final."