Percent here, percentage there: how is it with speed tests, not only in Germany

Even poor quality rapid tests can find a new place in the world in the face of new coronavirus epidemics. The error rate may not be the longest in the test.

In the Czech Republic, the first project of a large-scale testing of a large sample of the population is being prepared, most of them in memory of the fiasco with the Czech rapid tests, which according to the regional hygienic station in the Moravian-Silesian region had an error rate of 80 percent.

Changing errors in rapid tests and tests for area control of promo (how much according to our virus um) work on the same principle. So let’s expect a question: are such tests at all for him?

Yes, even those very mistakes. The original threatening appearance can be mitigated by a good explanation and you do not have to evoke the remaining pessimism. If someone utters such information, they should be able to explain it so that it does not cause distrust in the public. Today’s test of accidents for all people, even the seemingly very wrong ones can be considered a certain place in the city, if they are treated in the right way.

How it is tested

Today, in the Czech Republic and in the rest of the world, the laboratory is always tested by RT-PCR (polymer chain reaction with reverse transcription). Its principle is to recognize the genetic information of the virus, e.g. from a sample of the nasopharyngeal saliva and thus the presence of the virus itself in the patient’s body. The sample processing itself takes about 6 and 8 hours, if the capacity of the staff and test kits is free.

This type of test revealed infection and two days before the onset of the symptom and then for as long as the patient is infected. It should only be as long as I have symptoms (ie usually 57 days), but it can also be very important (and 5 weeks). It is important that the results are not affected by the patient’s condition and the course of the disease. After passing, the test is negative.

On the other hand, the infamous speed tests imported from it belong to another group of tests, the so-called serological, ie counter-tests. Detect the presence of antibodies (specifically the so-called immunoglobulin IgG and IgM).

A blood sample is taken for the test itself. The body begins to form after the onset of the disease, when symptoms appear (the disease can occur even without symptoms). IgM are the first lines of immune response to viral infection, detectable approximately three days after the onset of the disease. Kdeto IgG are specific antibodies with an important role in the immune system. Recent studies suggest that these IgG antibodies can be detected after eight days.

So where is the difference? If the patient is positive for the PCR test and negative for the rapid test, the disease has broken out and the immune system has not developed any antibodies. If both tests are positive, the patient is still infectious and thinks he is ill for about a week. If only the rapid test is positive, we can be sure that the disease has occurred, but the patient is not infected.

Therefore, it does not make sense to use rapid tests for the detection of the disease in the sweating phase, because it is negative all the time when the patient shows symptoms of the disease and is infected. The best use of the rapid test is in people who are likely to have been infected as a control. In the case of a positive result, a PCR test is performed on the site to rule out the patient’s infection.

However, the change of the hygienic station also tested these tests on the patient in the first stages of the outbreak of the disease, which caused that significant error rate of 80%. Without context, this information is very pessimistic, despite the fact that according to the epidemiologist Roman Prymula, the error rate in fact is only between 20 and 30%.

What about tmi sly?

In the field of diagnostic tools in medicine (according to, for example, in psychology and in police tests for alcohol or other substances), several terms and statistical procedures are used, which at first glance may not seem completely understandable. However, their understanding can help me to explain what the change in the error rate of the test really means, and for this, the seemingly threatening was not such a crucial value. Let’s help with the following table:

Table umo

This table allows you to compare the test results with the facts. Mainly, the color fields are in the middle. Green fields indicate the results that correspond to the actual state of health. Specifically, those cases where the test first revealed the infection (so-called true positive cases, SP) and even if it was first evaluated that the patient is not infected (true negative, SN). These should be as large as possible, but this test is not perfect and it is not possible to correctly recognize and rule out the infection in all those tested.

Conversely, red fields indicate errors. First, there are cases where in a healthy hunter the test incorrectly evaluates that he is infected (false positive result, FP). Secondly, there are cases when the infected test does not recognize the infection (false negative result, FN).

Reliability and error rate of the test in the table no. The error rate increased by setting both red fields and removing this as a percentage of all tested. On the contrary, reliability is obtained from the green field (ie again by summation and then output as a percentage of the total number of people tested). Reliability and error rate together are therefore always 100%.

They are clear and clearly understandable, but a clear bag about the quality of the test will not tell them everything. Do not provide information about the source of the error rate.

The quality of the test is therefore usually assessed according to other values. Sensitivity indicates the probability that the test will be correctly diagnosed with this test as the infection, ie how much of the total sweat the test will reveal. Specificity, on the other hand, is the probability that a test in unaffected people will really show that they are healthy. Otherwise, how many true (true) negative results have resulted in all really healthy people.

That’s not all. Another important indicator is the so-called positive predictive value (PPV). This value indicates what percentage according to the test positive people is in fact nakaench. This is often interesting information for practical use.

In contrast to sensitivity and specificity, this value also includes what percentage of the population in the total population is nakaench (so-called prevalence). If there are few such people, it is difficult to detect them. Even good tests (ie tests with high sensitivity and specificity) may not be sufficient in such a case.

Simple treasure: look for a single nakaenho in a million inhabitants. There is a really reliable test available, which, for simplicity, can be found by any real patient (sensitivity 100%). He then mistakenly identifies a healthy person as ill in only one percent of cases (ie 99 percent specificity).

This looks at first glance very very favorable. In practice, this would mean that after the first round of testing, you will have to look for your only patient among 10,000 people who mistakenly report the test as ill. In some cases, very good tests also lead (ie even these have a very low positive predictive value under certain circumstances).

In general, for quality tests, all changes in value must be as high as possible. With low sensitivities, some infections will remain undetected, so they will move freely and increase the possibility of the infections of others, even if they know about it. On the contrary, with low specificity, we will report many of those who do not actually have it, and we will unjustifiably quarantine them.

Two percent between them

We’ll look at the concrete treasure. Imagine that the healthcare system would significantly increase its testing capacity and be able to test 1,000,000 people, with 20,000 of them infected with coronavirus at that time. Thus, according to the selected population (prevalence) it would be 2%.

Data on the sensitivity and specificity of the currently used RT-PCR tests are relatively difficult to trace, however, one laboratory in Chicago states that the sensitivity of the laboratory tests used is approximately 93.7% and the specificity even 99.9%. So we will use them for this example, but take them with a grain of salt. They will only serve to make the interpretation somehow close to realistic slm. When we include these values ​​in the table change, it would look like this:


The numbers in the table at a glance show that such a test would be quite suitable for use on the surface, or its error rate is only 0.2% (we sowed the fields in the red fields and divided them by a very tested million). Pesto nm this test out of a total of 20,000 nakaench did not reveal 1,260 people.

If such were to be relevant, it would indicate the need for at least some of the current measures, as 1,260 falen negative cases are at risk of penetrating the virus. At the same time, we sent 980 healthy people to a 14-day quarantine unjustifiably due to a false positive result. However, given the overall population, this is really important and we can consider it acceptable in this context.

The test has such a high positive predictive value, specifically 95% (18,740 / (18,740 + 980) = 0.9503). This means that 95% of people who have tested positive with a PCR test will actually be infected.

We should have laboratory PCR tests. But what will it look like with those speed tests? As already mentioned, their actual error rate at first treatment is between 20 and 30%. So let’s look at one of the monch tables with the same population as in the previous example for an optimistic variant of the error rate of around 20%. (In fact, there are many variants, but let’s mention one of them for illustration, so take it with a grain of salt. If you want to try out what the situation would look like with different sounds, you can use our interactive application).


The rapid test with such results has a sensitivity of about 70% and a specificity of 80%, which are noteworthy values ​​not for the previous example. Then the people do not look satisfactory, 6,000 nakaench nm escaped by this test, on the contrary, we first found in 196,000 people that they had contracted the coronavirus infection.

The biggest problem with such a test is that it is only perceived by the real patient. The positive predictive value (how many people with a positive test result are actually treated) is only 6.7%.

Therefore, if the error rate is stated by the epidemiologist Prymma truth, it does not seem that these rapid tests would be suitable for the area of ​​the test population, which, after all, the hygienic station stated. This does not mean that my quality tests are completely unnecessary, when we have them available.

my lpe

For example, it is possible to monitor the number of people tested, ie in our case to choose people who are known to have contracted coronavirus infection. For example, those who drank in contact with an infected person.

One example for all: out of a million people in the previous case, the hygienists will select 100,000 for the test room. At that moment, the positive predictive value rises to 46% (14,000 / (14,000 + 16,000) = 0.467). This is certainly not irregular, but you can work with it.

This change has tried to point out that different percentages may mean something different in different situations and that it is therefore necessary to put them into context and get enough explanations. Where there was a pessimistic error rate of 80% at the arrest, it is finally admirable that our speed test is used. even those with coronavirus infection in proli).

Correction: In the text, the specificity and sensitivity were changed in one case during editing in the editorial office. We have fixed the error and I apologize for it.