Are temperature checks a good screen?

Temperature checks have been the 'go to' symptom check for Covid-19. Do they provide useful information?

Welcome to Plugging the Gap (my email newsletter about Covid-19 and its economics). In case you don’t know me, I’m an economist and professor at the University of Toronto. I have written lots of books including, most recently, on Covid-19. You can follow me on twitter (@joshgans) or subscribe to this email newsletter here.


Temperature checks are pretty ubiquitous these days. Some places — such as doctors’ offices — conduct them upon entry. More often — as part of the school or work protocol — people are asked to check their temperature at home and certify they do not have a fever. No one is under any illusion that having a temperature is highly indicative that you have Covid-19. But the question is: can they help?

This question was asked of me by the University of Maryland scholar, Brent Goldfarb after he read this article in the Wall Street Journal.

But experts and medical groups increasingly say that isn’t a good gauge of Covid-19 as many infected children and adults don’t get fevers. Furthermore, variability in individual temperatures as well as questions about the accuracy of body-temperature scanners and infrared contact-free thermometers put such checks at risk of potential error.

The article lists a litany of issues with temperature checking from the accuracy of thermometers themselves to obsessive temperature checking picking up things other than Covid-19.

The Value of Temperature as a Screen

Like other symptom checks, seeing if you have a fever is a screen for Covid-19. This is because many people who have Covid-19 with symptoms also present at a high temperature. Moreover, the timing of this is overlapping with the time someone might be infectious (rather than just infected).

What does the data say on this? The WSJ article cited a CDC study of 300 children with Covid-19 of whom 56% had a temperature. This would be a calculation of what we call the sensitivity of temperature as a signal of Covid-19 (that is, the percentage of people who are truly positive for Covid-19 who have a temperature). What it doesn’t tell us is the specificity of temperature (that is, the percentage of people who don’t have Covid-19 who have a temperature) and that is likely to be some people because you can get temperatures for all manner of things. Nonetheless, if we assume that only a few kids are likely to have a temperature from other courses (say, a specificity of 95%), then if a child had a temperature (and the overall rate of Covid-19 amongst children was 1%), then the probability that they have Covid-19 is 9.5% (almost 20 times the probability of someone who did not have a temperature). That is a significant indicator of an issue. I would hope that we wouldn’t be letting people who have a one in ten chance of having Covid-19 into schools.

What about adults? I ferreted around for a study of this and the best one I found involved examining Swiss military personnel. They had 84 Covid positive people and took their temperature daily over a period of 14 days. The results are here:

You can see that Covid-19 patients did have a temperature especially between 2 and 6 days after diagnosis. But if you used 38 degrees as the threshold for a temperature, only 18% of the patients had that for some period of time. Drop that to 37.1 degrees and that percentage increases to 63%. Why not just use that lower threshold? Well, the number of false positives rises significantly with the specificity of the temperature check falling from 100% to 95%.

What does this mean? With a 38 degree threshold (with 99% specificity and 1% prevalence), a temperature would indicate a 15% probability that you had Covid-19 but the errors are large with few people being identified by this method and a small number of both false positives and false negatives. With a 38 degree threshold (with 95% specificity and 1% prevalence), a temperature would indicate an 11% probability that you had Covid-19 and you would likely pick up some people with Covid-19 without false negatives but false positives would be very high (5% of the population).

If the prevalence in the population rises, the usefulness of temperature checks really goes up. With 5% prevalence, a fever indicates that you have Covid-19 with 40% plus probability.

This all suggests to me that temperature checks are far from useless but we have to be cautious about precisely the consequences of high temperature. Should you stay away on that day? Probably. Should you be quarantined for 14 days or even sent for another test? That isn’t clear.

Humans, Duh!

An issue with these statistics is that the temperatures were taken under clinical conditions. A proper thermometer. At our home, we use one of those no-contact thermometers which gives a reading to the decimal space but take it a few times and you know there is some variance. It is also quite sensitive to what the person might have been doing just before the temperature was taken. So we have to take some of this with a grain of salt especially if we wanted to lower the threshold for what we regarded as a symptomatic temperature.

The bigger issue is, of course, what humans might do with temperature information. Here is an anecdote. The other day, I decided to take my daughter’s temperature before going to school. We have to then record whether she has a fever or not in an app and, if she does, she won’t be allowed entry to the school, will likely have to isolate for 14 days and may have to be sent for a test which could be many hours of time taken. So when I tested her and she came up 38.5 degrees, my first thought was not, “oh no, she has Covid-19. We must keep her home” but “if you miss class today, can you do it online.” The answer to that turned out to be no. Suffice it to say, this created a dilemma.

She had just had a shower so what I decided to do was wait and see if it was a bad reading due to that. Half an hour later I took her temperature and it was 36.4 degrees. So it had been a blip.

But if I am being honest, this would have been a big dilemma for me. If she has a temperature, I would like to keep her home from school. But missing a class when it is every other day is a big deal. What’s more, she certainly wanted to go and had no other issues (well, as it turned out, because she was fine). And if this led to 14-day isolation, as appeared to be the guidelines, that would have been pretty disruptive. Suffice it to say, I can easily imagine many parents not reporting honestly their kid’s symptoms especially if they can’t be verified at school. If schools are serious about this, then they need to perform temperature checks at school.

In reality, the problem is the guidelines. If there had been an easy way of keeping her home for just a day to see if the fever lasted or other symptoms developed, that would have been the right path. But there was no such option presented.

This is why the human equation really matters for screening. We want screens that can help us assess the potential riskiness of people who are to interact with others, but if the consequences of false positives are too dramatic, it is hard to leave it in the hands of people to manage that externality.


What did I miss?