Improved diet and regular exercise may be the best way to treat type 2 diabetes, but that has not stopped the proliferation of drugs to tackle the condition. These range from the old and familiar — insulin, metformin and the sulphonylureas — to relative newcomers such as Actos (pioglitazone) and Avandia (rosiglitazone).

No drug is without risk. Some studies have suggested a link between certain formulations of insulin and an increased risk of cancer. Metformin can cause fatal complications in patients with kidney failure. Thirty years ago, US regulators placed a warning on the sulphonylureas after studies suggested they increased the possibility of death from cardiovascular complications. And both Actos and Avandia can raise the risk of heart failure, with Avandia also having the potential to trigger a heart attack.

Last week, an advisory panel of 33 scientists, clinicians and statisticians convened by the US Food and Drug Administration (FDA) picked through the evidence of Avandia's heart risks, as they debated whether or not the agency should pull the drug from the market. They recommended by majority vote that the FDA should restrict the drug's distribution (see page 420). But it was a difficult decision based on imperfect data — not least a lack of clinical comparisons between Avandia and other diabetes drugs including Actos, which is in the same chemical class but does not seem to pose the same risk of heart attack.

Few experts believe that Avandia became a blockbuster because it was demonstrably better than more established diabetes drugs. In the absence of comparative studies, many experts suggest that the drug's success was driven by an aggressive advertising campaign by its maker, the London-based pharmaceutical giant GlaxoSmithKline, and by clinicians' desire to offer their patients something 'new'. In 2006, the year before its safety risks were made public, global sales of Avandia reached £1.66 billion (some US$3 billion).

Research suggests that heavy reliance on newer, patent-protected drugs is partly responsible for driving up the cost of treating diabetes, which in the United States rose from $6.7 billion in 2001 to $12.5 billion in 2007.

That does not mean there should not be a place on the market for Avandia. Many clinicians — even those sceptical of the drug's widespread appeal — say that it is a worthwhile treatment option for patients who cannot tolerate the side effects of the older drugs.

But Avandia is a clear example of why early comparative-effectiveness studies are needed for drugs that treat conditions for which therapies are already available. It is a lesson worth remembering as the United States begins to carry out the $1.1 billion in comparative-effectiveness research mandated by the American Recovery and Reinvestment Act of 2009, and as officials ponder whether and how the data collected should inform decisions made under the nation's newly reformed health-care system.

The issue is a highly contentious one in the United States, as many politicians there equate 'comparative effectiveness' with health-care rationing. But without these comparisons, as the Avandia example demonstrates, clinicians have only intuition and their own limited experience to guide their prescriptions. And when safety issues do emerge, regulators are forced to make decisions based on incomplete data. The studies need to be done consistently and early in the approval process, before the damage is done.