What is it about our profession that causes us to cling so tenaciously to numerical cutoffs? Every introductory medical school statistics course teaches that for any assay interrogating disease, a population suffering from the disease forms a Gaussian, normative or "bell," curve when its test results are plotted in a histogram, and that curve nearly always collides with the bell curve derived from the population free of disease. The larger the collision, the lousier the test, but that fact is nearly always muted after a threshold is set and we settle into using the assay in our clinics.
展开▼
