Meta-analysis stands trial with Avandia

The FDA's recent Avandia re-evaluation provides a "useful lesson in the complexities and uncertainties of modern medicine," says the Wall Street Journal. It's really statistical techniques and the design of randomized trials that are at issue.

The FDA panel's votes--12 for market withdrawal; 10 for use restrictions with label revisions; seven to revise the label and add warnings--demonstrate a regulatory tightrope walk over an ocean of data, statistics, and analyses. Perhaps the regulator's thinking takes a familiar saying and gives it a twist:

"There are three kinds of statistics: statistics, damned statistics, and lies."

Meta-analysis, the technique used by Avandia opponent Steven Nissen (photo) of the Cleveland Clinic in making his case for why the drug should be pulled from the market, is often "unreliable or misleading because it merges into one data set many different studies with different levels of quality and size, and which are meant to measure different outcomes," the paper says.

Nor does meta-analysis allow for the possibility that different patients react differently to different drugs.

Nissen contends, of course, that meta-analysis is the "most robust available approach" for determining the drug's safety, and it provides a way to overcome limitations of earlier studies.

With the panel's vote last week, the Avandia debate has advanced by one ambiguous step. But the larger debate--concerning the value of data analysis and meta-analysis as safety and efficacy predictors on which to base regulatory decisions--is far from over, especially given FDA advocacy of risk-based methods in both the clinic and the manufacturing plant.

- here's the article

Suggested Articles

There's no evidence personal patient information leaked during the 11-week breach, but the same can't be said about Sangamo's own secrets.

Through a new online tracker, AllTrials names sponsors who fail to report clinical trial results on time per the FDAAA Final Rule.

The new solution aims to streamline the incorporation of human genomic data into clinical trial designs.