Bad Stats, Bad Medicine

The recent ruckus over the safety of the pain relievers Vioxx and Celebrex makes the opinionated Web site Improving Medical Statistics a timely read. Eric Roehm, a cardiologist from Round Rock, Texas, exposes statistical gaffes, shoddy study designs, and unwarranted conclusions that slipped past peer review and into the pages of top journals. For example, the doctor's warning that pregnant women should abstain from alcohol stems from a flawed 1984 study that didn't factor out the effects of smoking. Even the 2001 paper that first raised questions about the safety of Vioxx and Celebrex has a weakness: The researchers compared the treatment group from one study to placebo groups from other trials.

Improving Medical Statistics

(from Science magazine)