Part II defined the two types of testing that are being used to test for SARS-Cov-2 virus infection.
(2) Antibody testing ... cheap but not very reliable
Antibodies are proteins that your immune system makes to fight infections. New types of infection typically result in new types of antibodies. These are in your blood stream where they will be looking for cells infected by the virus. So an antibody test is done using a small pin prick of blood.
The detection of the proteins is based on their shape. Two proteins will stick together if they have the right shape ... like a hand fitting into the right sized glove. So the people making the tests have to come up with another protein that will fit the antibody like a hand fits a glove. Then they need something else to detect the fact that the hand is in the glove!
Imagine, ghoulishly, that you had a box of hands and wanted to see if a particular hand was in the box. You could do this by designing a glove that would only fit that particular hand, and no other.
What could possibly go wrong?
There are other corona viruses besides SARS-Cov-2 and a couple of them commonly infect people and cause cold symptoms. Does your body produce antibodies to these? Damn right. Might they look a bit like the antibodies to SARS-Cov-2? A little. But whether its antibodies to other corona viruses or antibodies to some other infection, the prime goal is to make your glove not fit anything else.
Antibody tests tend mostly to be rather imperfect. They are used because they are quick and cheap and they are often used to measure the percentage of a population with an infection; rather than to give a definitive answer about a particular person in the population.
To userstand this we need some jargon.
A) Sensitive. Suppose you have 100 people who you know have SARS-Cov-2 (because you did the RNA test on them!) and your antibody test successfully says the 99 of them have the virus. We say the test is 99% sensitive.
B) Specific. Suppose you have 100 people who you know definitely don't have SARS-Cov-2 and your test only makes 3 mistakes; saying that 3 people have the virus. We say that the test is 97% specific.
The accuracy of many tests is defined by these two measures: sensitivity and specificity.
The test sounds terrific. Until you think hard about what happens when you run it on a population where very few people have the virus.
Suppose only 1 in 100 people have the virus. Your test will improperly say that about 3 people have it when they don't. Do 1000 tests in this population and most of the positive results will be for people who don't have the virus. If this isn't clear, then stop reading and think about it until it's clear.
Just think about it. 1000 tests is 10 lots of 100. We should get about 10 positive results from people who actually have the virus, but another roughly 30 positive results from people who don't; because of that 97% specificity. The chances of having the virus given you test positive is just 10/40. You can call this 25% of a probability of 0.25.
Which brings us to the major point of this little series of posts. The following image. If you are more used to percentages than probabilities.
What about if we do a second test on the same person and get another positive result? In this case the probability of the person having the disease is no longer 1 in 100. What is it?
Consider a population of 10,000 and a rate of disease at 1 in 100. Test everybody and we will get 99 positive results from the 100 people with the disease and 297 positive results from the 9,900 people without the disease. So the rate of disease in the people with positive test results is 99/396. That's 1 in 4. So doing a second test is like doing a first test in a population with a 1 in 4 chance of having the disease. To solve this we just imagine a population of 400 where 100 have the disease and 300 don't. The test specification tells us that testing them all will give us 99 positive results from the 100 with the disease and 9 from the 300 who don't. So the probability of having the disease given a second positive test is 99/(99+9)=0.91. The result is in this second chart.
As you can see. Tests which sound really accurate need to be handled with care when testing for rare diseases. You don't want a person having a positive test result thinking that their life is in danger until you get it confirmed with a gold-standard DNA test.
Current antibody tests for SARS-Cov-2 are mostly terrible. The study just hyperlinked tested 14 such tests and found sensitivity and specificity in the range of 80 to 100 percent. Only 3 of the tests had close to 100 percent for both and even then, it depended on when the test was done. Sensitivity in the 3 best tests was highest a few weeks after the onset of symptoms.
Tests like these are great for testing a population, where you can take account of the defects in calculating infection rates. But for individuals wanting a diagnosis, then RNA tests are required.
Comments