[Skip to Content]
[Skip to Content Landing]
Views 1,488
Citations 0
Editor's Note
June 2014

Ensuring Correct Interpretation of Diagnostic Test Results

JAMA Intern Med. 2014;174(6):993. doi:10.1001/jamainternmed.2014.165

Today, physicians order a wide array of diagnostic laboratory and imaging tests for their patients, including genetic evaluations. To make sense of the growing number of diagnostic testing opportunities, one might expect that physicians, in turn, have grown in their ability to accurately interpret test results. The Research Letter by Manrai and colleagues1 finds this not to be the case. They replicated a classic study and found that only 23% of physicians and physicians-in-training correctly answered a single question testing their interpretation of a diagnostic test result. While this study was limited to a convenience sample from a single academic teaching hospital, it is not too far out on a limb to suggest that today’s physicians need to be better prepared to interpret diagnostic test results, including stronger training in statistics and clinical epidemiology.

In addition, this study reminds us that disease prevalence matters for testing, as does test accuracy (both sensitivity and specificity). However, these important pieces of information are often lacking at the bedside when we make a decision to order a test. Generally, prevalence is considered as only “rare” or “common.” And how often are physicians aware of diagnostic tests’ sensitivity and specificity? We need resources that make this information more easily and readily available.

In the meantime, before ordering any test, we must ask ourselves if it is even necessary. Assuming there are efficacious treatments for the disease being tested, what are our thresholds for “ruling out” disease on the low end and “ruling in” disease on the high end of probability, and then, what is the pretest probability of the disease? If your pretest probability falls between those thresholds, is the test accurate enough that a positive or negative test finding will result in a posttest probability that crosses these thresholds? If the test result is not going to change your clinical management, there is no reason for the patient to undergo testing in the first place.

The persistent inability of physicians to reliably manage this cognitive exercise implies that our educational programs need to do a better job at teaching numeracy skills. Because imprecise diagnostic decision making is leading to excessive testing, patient harm, and excessively costly care, we must raise the bar and master these cognitive skills.

Manrai  AK, Bhatia  G, Strymish  J, Kohane  IS, Jain  SH.  Medicine’s uncomfortable relationship with math: calculating positive predictive value [published online April 21, 2014].  JAMA Intern Med. doi:10.1001/jamainternmed.2014.1059.