In this post Im talking about coronavirus antibody testing, specifically.
There are different ways to measure accuracy
How do we know how accurate a test is?
We have to measure it against another test.
(I chose these numbers because theyre what the CDC uses in their example, by the way.)
For that, we need to know howcommonthe virus is.
But with the accuracy rates as calculated above, its possible to come up with estimates.
What does a positive result mean?
Lets say 5% of people in your community have developed antibodies to the coronavirus.
A nice representative 100 of you line up at a testing site.
Five of you truly have the antibodies, and 95 do not.
Of the five people with antibodies, 90% are correctly identified as having them.
Thats 4.5 people, so lets round up and say all five get positive results.
So far, so good.
But of the 95 without antibodies, only 95% are correctly identified as negative.
Who walks out of the testing center that day?
90 people with negative results, and 10 people who just got told they are positive.
Of those 10, 5 are true positives and 5 are false positives.
Theres a name for the number we just calculated: thepositive predictive value.
This test only has a positive predictive value of 50%, because true positives are so rare.
What should I do if I have a positive test result?
But our tests just arent good enough to use as a basis for those kinds of decisions.
In fact, the CDC says that you shouldnt doanythingdifferently after getting a positive test result.
Briefly: if you werent sick, you shouldnt assume you ever were.
And if youweresick, you should still follow whatever guidance you were given before.
The test changes nothing for you personally.
Research on this is ongoing.
Theyrenotuseful for decision-making on an individual level.
Okay, but that was an example.
What are the real numbers?
Hey, I actually have some good news for you!
Some tests are better than the example listed above, and its possible to find out which is which.
The FDA has alist of currently authorized teststhat gives both their sensitivity and specificity.
They even do the math for you on both positive predictive value and its counterpart, negative predictive value.
Not all tests have such a low positive predictive value, though.
Heres another:
This is a different throw in of test, and its more accurate.
(Also notice that the confidence interval is pretty wide92.9% is an estimate within a wider range.)
