Besides a vaccine, the Holy Grail in the Covid crisis that every government has been reaching for has been the setting up of ‘test and trace’ schemes.
Testing and tracing have been credited with the lower incidence of Covid in places like Italy, when it used to be one of the worst-hit countries, Germany or even the Philippines.
But these kinds of schemes are reliant on widespread testing. So the $64,000 question is: how good are the tests?
Dr. Robert Verkerk, director of the Alliance for Natural Health, and a columnist on our magazine What Doctors Don’t Tell You, and his colleagues have addressed just this question. Recently they offered blow by blow dismantling of the accuracy of the Covid test.
This involves deconstruction a few types of statistical analysis, so I’ll try to relay Rob’s brilliant findings as simply as possible.
Currently, aside from the clinical symptoms (like loss of taste and smell) most cases of Covid are confirmed through use of the RT-PCR test. That alphabet soup stands for ‘reverse transcriptase polymerase chain reaction,’ which means that the test works by using an enzyme called reverse transcriptase to turn a piece of RNA into a matching set of DNA.
The PCR or ‘chain reaction’ portion then starts crazily replicating the DNA exponentially, with a fluorescent signal added to these DNA doppelgangers to easily identify them.
Kary Mullis, the offbeat inventor of the test, was awarded the Nobel prize for his discovery, which has been used to detect, among other things, disguised moose meat in hamburger, the brain of a 7000-year-old human mummy, the true identity of the outlaw Jesse James and even the fur of a cat named Snowball, which turned out to be crucial in identifying the perpetrator of a murder.
A highly sensitive test in the lab, no doubt, but by Mullis’ own admission, no good when it came to diagnosing disease out there in the real world.
Rob quotes the biochemist Dr. David Rasnick, who says that the PCR test is a ‘great scientific research tool,’ but a ‘horrible tool for clinical medicine.’
A bad test for Covid
Why could such as sensitive test be rubbish at detecting a virus? When attempting to work out the accuracy of a test like this, researchers use two yardsticks: a test’s sensitivity –its ability to detect true positives – and its specificity – its ability to detect true negatives.
So let’s say that a test has a sensitivity of 99 percent. That means that out of 100 tests, only one of the tests will pick up a false negative. As for its specificity, out of another 100 tests, only one of the tests will show a false positive.
And all the makers of the Covid PCR tests claim high specificity and sensitivity.
So far so good. Despite the fact that governments like to say we are in the midst of an epidemic of Covid, the fact is that a very small percentage of us have this virus at any particular time.
So in that situation, says Rob, scientists have to rely on a statistical method called ‘Bayesian probability theory’ which says that if we know the ‘disease prevalence’ – ie, the proportion of the population in any community, state or country who are infected with the virus – we can better assess how accurate the test result is.
Here’s where it gets interesting. The lower the prevalence of the Covid virus, the wildly less accurate the test becomes in accurately identifying a true positive. The opposite is the case with false negatives; the higher the prevalence of disease, the wildly less accurate the true negative test results.
To assess the accuracy of any given test result, says Rob, you also need another statistic. This one makes use of Baysian probability to work out the Positive Predictive Value (PPV) and the Negative Predictive Value (NPV), both used by the late Doug Altman, a professor of Statistics at Oxford University, to determine the accuracy of diagnostic and screening tests.
Assuming that the tests have a 95 percent accuracy in sensitivity and specificity – which most companies producing these tests claim – just watch what happens to the PPV – the chance you get an accurate positive reading – when the percentage of the population with Covid falls.
These are Verkerk’s calculations (he is a PhD from Imperial College, essentially one of the world’s premier universities in mathematics and science, so he knows his onions) for parts of the UK.
For the Northwest, where 0.21 percent of the population supposedly have Covid, representing the highest prevalence of the virus, the chances of an accurate positive test stand at 9 percent. In the West Midlands, the area with only 0.04 percent of the population infected, the chances you are going to get an accurate test is essentially 1 percent.
What about the US, the country with the world’s highest number of cases, which reported 302,715 cases in the last seven days as of this writing?
That is, essentially 1 in every 1057 Americans, or a prevalence of 0.09 percent. According to Verkerk’s statistics, that means that the accuracy of those positive tests is, at best, about 15 percent.
You can calculate this yourself, says Rob, by plugging in the numbers in Medcalc, which is statistical software for non-scientists: https://www.medcalc.org/calc/diagnostic_test.php
Why is this so? Verkerk describes it as analogous as looking for a needle in a haystack. “As real needles are so few and far between, the chance of finding things that look like needles but aren’t increases.”