The real problem is the biomedical ignorance of statisticians

Heath’s review supports a book that gives a “devastating dissection of the statistical illiteracy of doctors,”1 when the real problem is the devastating biomedical ignorance of statisticians.

Why should a doctor need to know how to calculate the chance of breast cancer in a patient with a positive screening mammogram, given “a prevalence of 1%, a sensitivity of 90%, and a false positive rate of 9%”? What the doctor needs is a test that gives a straight yes or no answer, or something close to it. A positive test that reveals only a 1/10 chance is, at best, just a screen for more investigation; it is not a test on which an answer can, or should, be given to the patient. It is a bad “test,” and no amount of statistical juggling will improve it. The clinician needs better tests not better statistics—the better the test the less statistics is needed for its understanding; the usefulness of a test is inversely related to the statistics required for its interpretation.

The misguided reliance on statistics in contemporary health studies strangles their meaning. The problem is not that clinicians need more statistical nous to deconstruct them, but that clinical studies need to be more quantitative, methodologically rigorous, and above all imaginative, so that their science is understood and their conclusion obvious without having to peel away obtrusive, oversupportive, statistical packaging.

Statistics has wormed its way to the core of clinical research; it should return to being the simple ancillary aid that it once was.