In The Conversation, medical professor Norman Paradis gives a primer on the poor reliability of even some of the best screening tests.
“Tests in which you cut the patient open and examine tissue under a microscope have the best performance, with nearly perfect sensitivity and specificity,” writes Paradis. “Imaging tests, such as CAT scans and MRIs, provide millions of visual data points and also have very good performance. But by the time we get down to measuring the concentration of molecules in blood, problems develop. Such tests should not be used without a thorough understanding of the incidence of the disease. At the very bottom of the hierarchy of performance are psycho-social survey instruments – tests in which a series of questions are asked with the intention of making psychological diagnosis. Some experts have asserted that once publication bias (the tendency to publish only positive results) is removed, most if not all such instruments will be found to lack any predictive performance whatsoever.”
Could better tests have predicted the rare circumstances of the Germanwings crash? Probably not (The Conversation, May 27, 2015)