A few buzz words (describing mitigation techniques) have been floated during this time. I think that there are a fair amount of misunderstanding when using a lot of these words/techniques - compared to what they mean in a epidemiological setting. And since I have spent a fair amount of my time last three months discussing, explaining and listening to my friends and family telling me what limitations are involved with "testing", I would like to write a post about them (might have to be several).
First, a difference between tests. Two tests are mentioned a lot; PCR and serological (antibody). PCR test will indicate if there is RNA from the virus inside of a person.
Serological (antibody) will show if the person has developed antibodies against the virus (and therefore can not be sick again from the same virus). That is "the person has already 'seen' the virus".
Both tests have details attached to them that most people wouldn't know or care about. My primitive opinion is that people shouldn't really need know these details since FDA/EMA/country's each agency that approves test should know these details and only approve according to high standards. It might be obvious that I have an issue with how tests have been approved since I made this sentence in italics.
To evaluate biological tests for diagnostics there are three things to remember (there are a lot of others but these three are fundamental for how to view the test accuracy and validity - big words to say "does the test work well"):
- Sensitivity (how likely is a negative test to be really negative)
- Specificity (does the test detect the right thing that it's looking for)
- Limit of Detection (LOD) (related to sensitivity however slightly different and especially when talking about a whole chain of things together)
Before I go through these parameters, let's throw in another caveat with tests in general - especially since it goes hand in glove with these three fundaments. A test result should for all intent and purposes be seen as a probability and not a certainty. The golden and easy example for everyone to talk about is pregnancy tests. Those are, even if they are really really good, still a question of probability and certainty. The easy thing - the great thing - with pregnancy tests are that they measure one hormone (hCG) that only exists if the woman has a growing fetus inside. Then the test becomes only the question of "how low of a level of hCG can the test detect" (LOD). That's usually made into an equivalent of an average level of hCG at certain number of days after implantation, which is why tests have "early detecting, 5 days before period" (very low amounts of hormone can be detected) or "detection as early as first day of missed period" (a decent amount of hormone can be detected). If only all tests could be this clear cut.
With virus and PCR there is another issue that isn't much talked about outside the world of virus and microbiology. The fact that "just because RNA is detected doesn't mean that there is an infectious virus present". Dr Raciniello has described this on his blog a long time ago (here). The main take home message is "just because you find RNA, doesn't mean you find a virus particle that can infect." Further making it complicated, it will also depend where in the infection phase the person is, and what was investigated". It's been clear looking at published aggregated case studies from hospitals in regards to Sars-cov-2 virus is the likelihood of a positive test at different times in the infection within a person. The amount of virus, more specifically the amount of virus loose in the body available for detection and reinfection, varies a lot during the infection. There is a lot of virus day before symptoms show, there's evidence that 4 days after first symptoms there is more likelihood of testing positive. Combining the points (existing RNA but not active virion) seem to support the findings that 30 days after first symptoms, even if a positive PCR test, doesn't mean shedding (spreading) infectious virus. Brings back to the idea "it's a probablity, not a certainty". Also why hospitals where patients test negative on PCR but have distinctive 'glass lung' on CT are treated as covid19 positive.
(Side note for extra points; there is an assay that virologists use to evaluate virus load based on "plaque forming units" pfu which is dead cells that a virus have infected and killed, and that would give a "how many viruses who can infect exist in a sample" <- .="" p="" virions="">
So, back to the Sensitivity, Specificity and LOD.
When you take a PCR test - especially for an RNA virus - there is alway an overall chance/risk it will end up as a negative if you fail running the test itself. Why? Because the test is looking for a detection of something that will have to be going through at least 3 steps where the actual RNA might disappear before showing up in the test results. That's why there's always a control sample run together will all the real samples to make sure that all the steps worked out.
- Sensitivity is generally descried as how good the test is to find the true positives of what the test is looking for. That is to say, only the people who really have RNA from the virus should test positive when using the PCR test. Similar for the serotest (antibody), only people who have encountered Sars-cov-2 and developed those specific antibodies should test positive.
- Specificity is generally described as how good the test is to show true negatives of what the test is saying it looks for. To simplify, if you test for sars-cov-2 RNA it's important that anyone with any old cold doesn't show up as positive (the common cold is part of the coronavirus family and share some parts of the RNA with sars-cov-2). If you don't have the virus, you are negative when testing.
There are repercussions of this part failing as well since looking at serotests (looking for presence of antibody in the blood) and if this shows antibody titers but it's the wrong type of antibody (say, from the common cold) then the test will have a low specificity.
- Limit of Detection (LOD) is the lowest concentration/level of test subject that can be detected in a sample. The reason for my addition to the list is that when it comes to certain tests, the LOD will depend on a number of different processes than "just the level in the blood/saliva" as is the case for a lot of samples. Again the pregnancy test as an example, pee on a stick. The hormone is in urine. The test is designed for levels that are found in urine (side note, the blood test is more sensitive and have a lower limit of detection). There is no part of the test that requires any manipulation/handling of the urine (or blood). However for a PCR test of an RNA virus - there are a lot of details involved. First of all, there is the base level of "amount of viral RNA present in the nose swab" that has to be present to be picked up on the swab. Then there are different ways to get the RNA (virus) off the swab, extracted into liquid and being able to even "run the PCR on" (PCR=polymerase chain reaction, a mix of stuff in a little tube to amplify what is in the tube from the start). And the limitations on how big of a volume was the RNA extracted into and how little was brought on into the PCR reaction etc.....
I'm aware I didn't do a great job on the LOD discussion if nothing else I hope that I made it clear that it is complicated and also super important. Even if you don't know how to calculate this step - it should be super clear in the FDA approval process since the LOD will determine the limits of the usefulness of the test.
All three of these parameters will decide if the test is useful. It will further determine if the test is useful on an individual or population view. Why this difference? Because of statistics. If the test is only to be roughly an estimate on where the population - a higher lever of uncertainty (higher possible wrong results) are more acceptable since stats work that many tests together with a fairly high certainty turns into an ok stats.
The same test can have a detrimental effect on an individual basis. As previously stated, imagine going for a PCR test to determine you don't have the virus before going to visit your grandmother. You get a false negative (you think it is negative), and then go visit grand mother. That's a huge risk. Similar with testing if you have antibodies (the serotest) and it comes back positive (you have antibodies). You then might start doing relaxing and not wash your hands as much as before, and then you get sick.
All of these parameters are things that I (we) anticipate and expect that the FDA (EMA/country of your living) have evaluated and deemed acceptable. However, it's important to remember that in this case of the pandemic - there were a lot of changes in the acceptance criteria for a lot of tests. In order to get an Emergency Use Authorization (EUA) the documentation from companies were relaxed a lot and most of all, not required to clearly state the limits on the test themselves.
You can argue that there isn't time to go through rigorous testing and documentation in a pandemic with an unknown virus - and I would agree whole hearted that time is of the essence. However, I would say that these details, the basis of the validity of the entire test, are extremely important to clearly state. If nothing else exactly to use when calculating the results on a population basis since the errors will compound/become very different based on these base numbers.
From an individual point of view - I wouldn't take a serotest without wanting to know the exact specifications since I wouldn't know what to do with a positive test unless I knew the cross-reactivity number. And that's not even mentioning the issue what kind of levels (or types) of antibodies that might be protective against sars-cov-2.