A test that is positive for coronavirus, aka Covid-19, confirms the presence of infection, if you have clinical signs and symptoms of infection. If you don’t, it doesn’t. Alternatively, a test that is negative is helpful in looking at a large, healthy population but it means much less for someone who is ill. Even worse, negative tests in sick people result in infection and death of healthcare providers and family. Confusing? Yes. 

There has been a lot of discussion and concern about the late arrival of testing for Covid-19. In order to shed a little light on this issue, let’s look at the current understanding of the test currently in use for detection of the disease. The test is properly done by sampling mucus from the nasopharynx, the back of the nasal cavity above the uvula, the little “punching bag” seen at the back of the mouth. The RNA of the virus is amplified by the polymerase chain reaction until there is enough to measure. The test is highly dependent on two things, amount of virus in the site sampled and sampling the correct site with sufficient (meaning unpleasant) vigor. 

Machines are tested on standard positive and negative samples. The machine characteristics must deliver accurate results 99% of the time or higher. However, samples submitted are not like the standards.

Data from Covid-19 testing in a real life scenario are few. A study published in the journal Radiology looked at testing done in China on about a thousand patients who presented with symptoms typical of the Covid-19 infection. 600 PCR tests were positive and 900 chest CT scans were positive in this group. It could be assumed that 300 PCR tests were false negatives and that the CT scans were highly accurate.

Using this data set as a springboard, let’s look at two vastly different uses of testing to detect infection.  First, consider a cohort similar to the one published, with a 90% prevalence of disease and a test that has a specificity of 99% and a sensitivity of 70%. A 4 x 4 plot of true positives, false positives, and true and false negatives would look like this:

Test + Test – Total  (Key)

Disease present 600 300 900 TP FN

Disease absent 1 99 100 FP TN

The positive predictive value in this situation (TP/TP+FP) is 99.8. The negative predictive value (TN/TN+FN) is 0.25. Thus a positive test is helpful but a negative test clearly does not exclude disease and is useless in changing any aspect of management when the prevalence is high. In this scenario, the false negatives may result in infection of many others including care providers.

Now look at a second cohort with a prevalence of disease of 0.1%. This would be using the test to screen for disease. (At this writing, the number of cases in the US is almost 300,000 in a nation of 330,000,000, or roughly 0.1%) In this case, let us screen 100,000 people with the same test, of whom 100 will be infected.  The 4 x 4 plot follows.

Test + Test – Total  (Key)

Disease present 70 30 100 TP FN

Disease absent 1,000 98,900 99,900 FP TN

The positive predictive value here is 70/1,070 = 0.07, which in ordinary English means that only 7% of the positive results occur in people with infection.  The negative predictive value is 98,900/99,200 = 99.7.  Suppose a leader proposed isolating all people with a positive test in this situation.  93% of those isolated would not have the disease and would be isolated unnecessarily. Further, testing for infection in screening vastly overestimates the prevalence of disease, in this case by about ten fold. (100 with disease but testing positive in 1,070) In the US, if we tested about one third of the country, one hundred million, we would quarantine a million people unnecessarily. This might be acceptable to some governments but a lot less so to Americans who understand these numbers. We would not sentence 1,000,000 innocents to a month in jail to ensure that 70,00 criminals did time for their crime. Additionally, there is a growing suspicion that about 25% of those infected are asymptomatic or minimally symptomatic yet spread virus for weeks. We have no idea how accurate that figure is because false positives are blended with the silently infected.

The CDC and the FDA cooperate to an extent on development of tests for epidemics. Considerable regulation exists with the intent of ensuring high quality and accurate performance of laboratory testing. With this coronavirus epidemic, the hue and cry of the masses (mob?) has forced the suspension of many of the procedures that foster accuracy, just so a test is available. We will soon have tests flying off the shelves to be used to diagnose this condition. It is extremely unlikely that these instantly created machines will have better operational characteristics than the existing tests. It appears that the likely less accurate results will be ready a lot sooner.

On a more positive note, there are tests that measure antibody to the virus, a protein that shows up weeks following infection. These tests are generally more accurate with far fewer false negatives. This testing can give a better notion of the prevalence of disease but after the fact. It will help clarify the false positives versus asymptomatic carrier question.

The late appearance of testing for Covid-19 in the US began as a consequence of false statements from China that led to a misunderstanding of the characteristics of the viral transmission and underestimation of the problem, and obstructive regulations in place for years that might be acceptable in times of health but fail in times requiring an urgent response. Federal health agencies could adopt a more functional approach to respond to epidemics, one that provides a nimble response and production of testing mechanisms by private industry that is both massive and as accurate as possible, recognizing the limitations and strengths of acute testing. On the other hand, testing for viral presence in the nasopharynx will always have the limitations noted here. An argument could be made that such testing is of limited value in containing an epidemic and that the focus should be in changing behaviors and practices that limit spread until a vaccine can be developed a year or two later. Numbers from testing certainly help the news media industrial complex while giving many people anxiety and provoking other psychiatric reactions. Fear is a double edged sword. It motivates many to comply with infection control measures while it stimulates fraud and crime in others. It’s impossible to say at this point—or at any point—what effect on infection and mortality rates has been derived from government-dictated termination of selected business activities as opposed to a lighter touch with enhanced guidelines regarding occupation limits, spacing, masks, cleaning, etc. Just like we will know more about infection rates a year from now, we might better understand how devastating government actions have been on the economy, provided we have control states or economies where the restrictions have been less onerous.

Coronavirus will not be the last epidemic. All epidemics will not necessarily be viral as there is a rapidly expanding list of antibiotic resistant bacteria. Microbes owned our planet for two billion years. They want it back.