HOW TO READ ARTICLES ON DIAGNOSTIC TESTS
This document is an introduction to how to think about diagnostic tests; it is meant to be self-contained and readable at some time when you may need a short refresher course. With respect to “how to read an article about a diagnostic test” look at section VI, which draws upon the discussion and lists the questions that you should ask about any article that describes a diagnostic test. I. What do we mean by “diagnostic tests”?
For purposes of this discussion, diagnostic tests include everything physicians do to diagnose disease. Examples include physical examination for ascites, blood culture for evidence of sepsis, head CT to rule out intracranial bleed, the CAGE screening test for alcoholism, and electronic fetal monitoring for signs for fetal distress. II. What is a gold standard for diagnostic test?
The gold standard is the best single test or combination of tests that is relevant to the particular diagnosis. Thus, when thinking of exercise tolerance tests as a diagnostic test for coronary artery disease, the gold standard would be angiogram or autopsy. The gold standard is different for different tests, and may be impossible to obtain except under unusual situations (autopsy, for example). III. What do the terms sensitivity, specificity, positive predictive value, and negative predictive value mean, and how do they apply to a clinical situation?
Definitions: a. Sensitivity is the likelihood that the diagnostic test will indicate the presence of disease when the disease is actually present. b. Specificity is the likelihood that the diagnostic disease will indicate the absence of disease when the disease is actually absent. c. Positive predictive value is the likelihood that a positive test result actually means that the disease is present. d. Negative predictive value is the likelihood that a negative test result actually means that the disease is absent. Statistics for a test are usually given in percentages. For example, a 70% sensitivity for a particular culture would mean that, on average, the test would read positive for 70% of women who truly have infections.
It is important to understand the clinical meaning of these concepts. Suppose I decide to perform a Microtrak test for Chlamydia on an asymptomatic female patient. In this setting, a. sensitivity will be the chance that if a woman had a chlamydial infection it would be picked up by the test.
b.
specificity will be the chance that the test will indicate no infection if, in fact,
the woman has no infection.
c.
positive predictive value is the chance that the positive result that comes
back actually represents chlamydial infection.
d.
negative predictive value is the chance that the negative result that comes
back actually represents the absence of a chlamydial infection.
IV.
What is a 2 x 2 (read two by two) table, and what is it used for?
A 2 x 2 table is an easy way of summarizing and calculating all of the information about a diagnostic test. To make a 2 x 2 table for the Chlamydia example, draw a four-square table as follows, with test results + or – on the side, and true infection status + or – at the top. Then fill in the numbers from your data, putting true infections which the test labeled positive in the upper left corner, and so on.
TRUTH
Disease Present Positive a Disease Absent b
TEST
c Negative d
The 2 x 2 table allows easy calculation of the important values by the following formulas: sensitivity = a / a + c specificity = d / b + d positive predictive value = a / a + b negative predictive value = d / c + d
V.
What is Bayes’ theorem, and what are its implications for diagnostic tests?
In words, Bayes’ theorem states that the predictive value of a test will depend on the prevalence of the disease. For diseases with high prevalence, the positive predictive value will be high; as the prevalence drops, however, the positive predictive value will decrease. The negative predictive value moves in the opposite direction. In practical terms, given a particular test, if a clinician uses a diagnostic test in a high prevalence setting, a positive test will be more likely to be truly positive than in a low prevalence setting. For example, San Francisco gay males have an HIV seroprevalence estimated at >70%; asymptomatic low risk people from rural north Carolina have a prevalence of disease of <<1%. A positive HIV screen in North Carolina is very likely to be a false positive; in a gay San Franciscan, the chances of a true positive are much greater. Many clinicians initially find Bayes’ theorem counterintuitive, but understanding what it means is essential for a rational approach to diagnosis, especially in population screening. 2 x 2 tables can be used to demonstrate the effect. VI. What questions should be asked of every article in which a diagnostic test is described? A. B. C. What is the disease being looked for? What is the gold standard? Is it reasonable, and what are its limitations? Does the article define the sensitivity, specificity, and predictive values of the test? If the article does not explicitly give the values, calculate them yourself. Toss out any article that doesn’t give you the data you need to get these values. What is the prevalence of disease in the population used to validate the test? How different is that population from yours, and what effect would that difference have on the predictive values of the test?
D.
F:\bft5e\Worddata\diagnostictests.doc