Sensitivity and specificity taken to the next level
In this video, we’ll be taking your expertise of sensitivity and specificity one step further.
In this video, we’ll be taking your expertise of sensitivity and specificity one step further. Among other things, you will learn how moving a threshold influences a test’s sensitivity and specificity.
PS: Unlike my previous videos, this one does not have an intro of me talking into the camera. Do you prefer videos with or without intros? Please let us know in the comments section.
[00:00:00] So, coming back to the blood pressure distributions we talked about previously. We said that moving the threshold of the test will change the proportion of people falsely and correctly classified as diseased and as healthy. Or let's pick another example relating to heart failure. Let's say we have a group of ten patients with heart failure and a group of ten healthy controls without heart failure. This is the concentration of BNP or brain natriuretic peptide, a marker of heart failure.
[00:00:30] So, people in the heart failure group, on average, have high levels of BNP. One's up here, one's here and so forth. Individuals without heart failure have lower levels but as we've seen previously, there will be some overlap. So, if we set our threshold at a lower level we'd correctly classify eight out of the ten diseased, two would be falsely classified as healthy. However, seven out of the ten healthy controls would be falsely classified as diseased. Whereas three would be correctly
[00:01:00] classified as non-diseased. What are the sensitivity and specificity? The sensitivity would be calculated as eight true positives, divided by all diseased folks or 80%. And the specificity would be calculated as three true negatives, divided by all non-diseased individuals or 30%. And what happens if we move the threshold to a much higher level? Well, in this case, we'd only pick up two diseased individuals or true positives and we'd miss eight people or false negatives.
[00:01:30] Turning to the non-diseased, we'd correctly classify a much larger proportion of non-diseased individuals. So, nine out of ten non-diseased would be correctly classified and only one individual would end up as a false positive. So, the sensitivity is two, divided by ten or 20% and the specificity is nine, divided by ten or 90%. So, moving the threshold is usually a compromise between sensitivity and specificity. As one goes up, the other one goes down and vice versa.
[00:02:00] Sensitivity and specificity are great for public health purposes because they tell us how many diseased individuals, in our population, we're going to pick up with the test and how many non-diseased we'll be able to rule out. However, the clinician needs to know slightly different indicators of validity because she's confronted with positive or negative test results and will have to decide what to do with them. And these additional indicators are going to be covered in the next lessons.