COVID-19 VirusUpdate: It turns out that Maine’s CDC has received only 5 percent of the tests it expected from Abbot laboratories. An indeterminate number of additional tests should be delivered later this month, according to Abbot Labs. This is a serious setback, since only with widespread testing can we understand how to accurately start rolling back social distancing and other preventive measures.
With the advent of the new ID Now COVID-19 test that’s going into production by Abbot laboratories here in Maine I’m thinking now might be a good time to discuss a statistical view of our natural environment. Statistical, you say? Yes, we want a high confidence map showing estimates of the infected population and their boundaries.
But first, we need to know more about the test, that I have not seen reported. In other words, how accurate is it? Medical tests are assessed on the basis of their specificity (false positives) and sensitivity (false negatives). The former identifies most of the people who do not have the disease, and maybe a few who do. The latter identifies most of the people who have the disease, and maybe a few who do not. These can be illustrated with the following table :
| Test result | Disease present | No disease | Totals |
| Positive | Large numbers
(true positives) |
Small numbers
(false positives) |
|
| Negative | Small numbers
(false negatives) |
Large numbers
(true negatives) |
|
| Totals |
With this information we can use ratios to calculate the various likelihoods of each cell. A sensitive test is good at screening because it identifies most of the people who have the disease. In contrast, a specific test is good at diagnosing because it identifies most of the people who do not have the disease. How high the percentages for each of these test aspects are depends on several factors that impact the consequences for being wrong.
Since the FDA has fast-tracked approval for this test we’ll assume it’s up to the job. The next question is, what do we want to know? According to Dr. Fauci and other epidemiologists, we want a good estimate of the proportion of the infected population and where the disease is. To learn this we now have to sample the population, in this case Maine. I’ll summarize what I think would be one way to do it.[1]
Here are two possible methods for tracking COVID-19; both methods would require a large supply of tests that could be turned around within 24-48 hours. The first would involve collecting a sample using the list of Maine’s Census tracts. With this we could conduct a cluster sampling to randomly select some of these Census tracts. We would then randomly select within these tracts (possibly in multi-stages) to the point at which we have a list of randomly selected households. During this process we would publicize this sampling campaign to alert people to the possibility of their being contacted to take a COVID-19 test. How these contacts would be made would depend on the information Maine has in its data. People would be screened out if they had a previous positive COVID-19 test (however, prior tests should be merged with the sampling data to provide a more comprehensive picture). Subsequent testing could be conducted with this sample to determine disease spread over time.[2]
Alternatively, no survey would be conducted; instead, it would be well-publicized that anyone with COVID-19-like symptoms should get tested. This is the preferred method in most areas should all the necessary test kits become available.
Contact tracing (locating all the positive person’s contacts during the past two weeks and warn them to shelter in place) would be conducted with either method at the beginning and end of the outbreaks when there are relatively few positive cases. Tracing would also be conducted in sparsely populated areas. This would serve to limit the spread of disease; however, contact tracing is not feasible once the numbers of new cases begins to explode because there would not be enough personnel and monies to carry it out.
The sampling method would be more expensive but would have the advantage of providing information from people who do not go to a doctor or get tested because they either have mild symptoms or none at all. Without conducting a sample, these cases would remain unknown. Testing only those with COVID-19-like symptoms would be cheaper but would not show the full spread of the mild disease cases. This is important information since people with the disease who exhibit no or mild symptoms can spread the disease to people who are at greater risk for complications.
[1] The full methodology would likely take several pages, but since I don’t have all the information necessary from Maine and you probably aren’t interested I present only the ”broad brush strokes.”
[2] This method won’t work as well if the COVID-19 tests aren’t sensitive enough to detect pre-symptomatic cases within perhaps five days of infection. In that case we would just test the entire population presenting symptoms possibly related to COVID-19.
Stephen Fielding Images 