STATS, the statistics-based media watchdog, has published its annual Dubious Data Awards, flagging the year’s worst science stories. It is, as usual, an impressive list: There’s the tale of the San Francisco mayor, who banned city agencies from buying water in plastic bottles, partly because they’re made of a chemical that sounds a lot like a known carcinogen but isn’t. Or the one-man advocacy group that announced without the benefit of evidence that fire-retardants in mattresses could kill 300 million people—more than cancer is expected to kill in the next 10 years.

Ranking right up there on the list was the reporting on the New England Journal of Medicine’s study claiming that GSK’s Avandia (rosiglitazone) increases the risk of heart attack by 43 percent. The STATS take:

On closer inspection, many of the cases of heart attack driving the statistical increase in risk came from one study where the diabetic participants already had congestive heart disease. Worse, Nissen excluded studies where there were no heart attacks. If you want to look at the safety of a drug, studies that don’t find adverse outcomes are as statistically important as studies that found adverse outcomes. Nissen also wasn’t able to control for the amount of time patients were exposed to Avandia relative to when they had a heart attack, which further undermined the [certainty] of his result. Even though a number of studies were published in medical journals criticizing the methodology, the media ignored them and continued to report the magic increased risk of 43 percent.

As Brian Strom, chair of the Department of Biostatistics and Epidemiology and director of the Center for Clinical Epidemiology and Biostatistics at the University of Pennsylvania noted: “You should not draw any conclusion based on this analysis. I would not recommend taking patients off of Avandia because of this.”

Here’s more from STATS on Avandia.