Friday, February 06, 2009

Can Anyone Spare A Statistics 101 Text?

We know statistics is not everyone's strong suit. Take the folks at Sequenom, which later this year hopes to market a blood test called SequreDx to screen for Down syndrome.

Yesterday, the diagnotic test maker acknowledged some errors in a presentation given the previous week. Such as? Well, Sequenom misused technical terms to describe the test's accuracy, and miscalculated the percentage of unresolved results from its RNA-based test, according to the Associated Press.

To be specific, Sequenom said its test had a positive predictive value of 99.9 percent, but meant to say specificity was 99.9 percent. Positive predictive value is the proportion of samples with positive test results who have the disease, while specificity is the proportion of healthy patients that are correctly diagnosed, the AP notes. The presentation also confused negative predictive value and sensitivity. In these samples, however, both rates were 100 percent.

And so a contrite Harry Stylli, Sequenom's ceo, tells the AP the mistakes were "editorial" and the underlying test data was correct: in 858 samples, the SEQureDx correctly identified 28 cases of Down syndrome in high-risk pregnancies. There were no false negatives, or samples the test failed to detect the presence of Down syndrome, and one false positive, in which the test returned a positive result and further testing showed the fetus was not carrying the condition.

The correction, he said, "led to folks basically saying we were cooking the books, and that led to the slide in the stock." Indeed, Sequenom shares fell nearly 17 percent at one point this week and Lazard Capital Markets analyst Sean Lavin wrote investors that the changes hurt management credibility.

What are the odds of something like this happening again? We have no idea. But we do have one suggestion - don't ask Sequenom management to make the calculation.

image from flickr user mac steve used under a creative commons license.

No comments: