COVID-19 Testing - We need more of it -but we shouldn't overinterpret individual results
Coronavirus testing is hugely important to allow us to determine new outbreaks of COVID-19. But many people forget that test results must be interpreted based on context. There are many reasons why a test result might not indicate correctly whether someone has a disease.
Mislabeled specimens can yield to a false positive or false negative. This is shockingly common! (Here's an article about mislabeled prostate cancer biopsy samples)
Testing too early before viral load is high enough to pick up virus genetic material in the sample. COVID-19's incubation period is 2-14 days (median 5 days), so people tested day 4 after exposure might be negative, but might be contagious within a few days.
Poor specimen collection technique. Nasal swabs need to be inserted pretty deep, and it's uncomfortable. Dainty providers will see more false negatives)
Poor quality swabs. Most of the world's medical swabs were manufactured in Northern Italy - so there's been a wee bit of a shortage
Poor specimen handling. Swabs must be inserted in viral media and then refrigerated until they reach the labs. If they are not cold enough, the test might be a false negative. Even the point of care test could yield an incorrect result if not performed properly.
Test Characteristics: Tests all have less than 100% sensitivity and specificity. So even if all of the above problems are eliminated, a positive or negative test changes the pre-test probability of having disease, and doesn't unequivocally state that the person has that disease.
Sensitivity:Likelihood that someone who has the disease will have a positive test. Sensitivity of 90% means that of those who actually have the disease, 10% will be (falsely) negative.
Specificity:Likelihood someone without the disease will have a negative test. Specificity of 99% means that of those who really don't have the disease, only 1% will be (falsely) positive.
There are reports that the sensitivity of the PCR tests could be as low as 70% for nasal swabs, or 60% for throat swabs. That's not ridiculous - the rapid nasal influenza swab test has a sensitivity of around 55% in adults. Researchers also believe that stated sensitivity and specificity by the manufacturer usually overstates the accuracy of the test, and most early reports on accuracy will come from manufacturers.
The Chan Zuckerberg Initiative has reviewed Antibody testing. No test for acute illness (IgM) had a sensitivity of more than 50% in the first six days of illness. Specificity of the antibody test for convalescent COVID-19 (IgM) generally hovered around 95% - although one test had 16% false positives.
What does this mean for COVID-19?
COVID-19's incubation period is between 2 and 14 days. So if someone is infected on Monday, they are likely not to have symptoms until at least Wednesday - and the average person might not have symptoms for 5 days (Saturday). Two and a half percent won't have symptoms until late in the eleventh day after they were exposed. Many people tested during the period between infection and 2 weeks later will test negative, and a negative test doesn't necessarily mean that the person could not be contagious. It also doesn't mean that the person won't become contagious a number of days later. There are also some people with COVID-19 who have no symptoms at all - we don't know whether or when they will test positive.
The post-test likelihood of having COVID-19 is strongly correlated to the pretest probability. (This is wonky - and it's based on Bayes' Theorem - or probability theory - but it's enormously important in evaluating the result of a test. This is poorly understood even by many practicing clinicians. )
Let's assume that the test being used has an 80% sensitivity (probably optimistic) and a 99% specificity (might be true).
If someone has a "screening" nasal swab and has no symptoms, their underlying probability of having a "true" positive is probably 10% or less. In this example, one out of ten with a positive test is a false positive, and 2 out of every 100 with a negative test actually has COVID-19
Screening someone with no symptoms
Pre-test probability
10%
Sensitivity
80%
Specificity
99%
Have Disease
Do Not Have Disease
Predictive value of a Positive
Predictive Value of a Negative
Test Positive
800
90
90%
Test Negative
200
8910
98%
Total
1000
9000
This means that if you could screen people entering an airplane, and NONE of the 200 people entering the plane had symptoms, there could still be four people who had COVID-19 who test negative and be allowed to board the plane. If we had instead assumed a test sensitivity of 70%, we'd inadvertently allow six people with COVID-19 to board.
If someone has a test because they have a fever, cough and body aches in New Orleans or in New York City last week, the pre-test probability they have COVID-19 is as high as 90%. In this example, a positive test is almost certainly correct, while almost 2/3 of those with a negative test will actually have the disease. If the sensitivity is only 70%, then almost 3/4 of those with negative tests will actually have the disease
Diagnostic test on someone with symptoms of COVID in context of community spread
Pre-test probability
90%
Sensitivity
80%
Specificity
99%
Have Disease
Do Not Have Disease
Predictive value of a Positive
Predictive Value of a Negative
Test Positive
7200
10
100%
Test Negative
1800
990
35%
Total
9000
1000
This demonstrates the importance of excluding those with symptoms from the workplace in the context of community spread of disease, and the reason why access to tests would not be helpful. A negative test doesn't lower the post-test probability enough for us to feel confident that the person with the negative test is really not contagious.
Everyone wants an objective test to determine whether people have a disease, and it's natural to over-rely on such a test. It's important that we are able to convey to our clients the true value of a test - and help them design approaches that consider the accuracy of tests when determining who to exclude from the workplace.