Monday, May 21, 2012

More Thoughts on Sensitivity and Specificity

Recall that the sensitivity and specificity of a diagnostic test refers to its ability to either correctly identify people with a condition of interest (sensitivity) or correctly identify those without the condition of interest (specificity). There are some implications to this. The mnemonic SnOUT means “sensitivity rules out.” Translated, this refers to the fact the greater the sensitivity a test has, the better able it is to correctly identify people with a condition of interest, the more likely a negative test truly rules out that condition. And the mnemonic SpIN means “specificity rules in” In other words, the better able a test can rule out condition of interest (the higher its specificity), the more likely a positive test is truly positive.

Think of it this way. Consider a test where there is a cut-off point. Above that cut-off, everyone is said to be positive, or to have the condition of interest; below it, they do not. For example, we might say that everyone above a temperature of 98.6 degrees is “sick,” and everyone below is “not sick.” But we also know that there will be some people who are sick who do not have an elevated temperature (false negative) and some who are above our cut-off point but are not sick at all (false positives). Now, let’s shift the cut-off point. Let’s move the point where we call people “sick” to 101 degrees. One thing is for sure; everyone with a temperature above 101 degrees is pretty much sick. That is, our test now has a very high sensitivity, because with a cut-off of 101 it can correctly identify sick people very well. But, guess what? The specificity falls now- we have a higher percentage of people with temperatures under out cut-off point who really are sick but our test does not identify them very well.

And the reverse happens when we lower our cut-off point. If we drop the temperature to 97 degrees, virtually everyone under 97 will be correctly identified as “not sick,” increasing the test’s specificity. But we will see a decrease in sensitivity because some people will now be identified as “sick” when they really are not. In any test, as we increase sensitivity we decrease specificity, and vice versa. So, we need to decide whether we want to increase sensitivity at the expense of specificity or specificity at the expense of sensitivity.

So, why would this matter? Tests with high sensitivity may be suitable when the consequence of reporting a false positive finding to the patient is minor. For example, we tell them they have high trigylcerides and they shift to a healthier diet. But, tests with high specificity are better when false positive findings could lead to painful or expensive treatments; i.e. a test that leads to a surgical intervention. In this case, false positives should be minimized.

No comments: