Thermometer Study Attempts to Blame Schools for Covid Spread

Are children responsible for 70% of household COVID transmission? Doubtful.

When I first saw the recent study claiming that 70% of US household transmission came from a child, I just rolled my eyes. I figured it would fizzle out as most people saw it had weak methodology and wasn’t worth publicizing. But I underestimated how many people are still desperate to blame children and in particular schools on Covid spread. Mary Van Beusekom wrote an article for CIDRAP summarizing this study, which was put on blast by noted Covid hysteric Eric Feigl-Ding and others, and it picked up steam over the past couple of days, so now I feel it needs a documented response. After digging deeper into the study, I’m honestly somewhat surprised the study got published at all.

So far the CIDRAP article is the only media coverage of this study, which was published in JAMA Network Open. The CIDRAP author, Mary Van Beusekom, is familiar to me for publishing an article that grossly overstated Long Covid prevalence. That article was featured in one of my previous Substacks and was actually corrected by CIDRAP due to the error I identified. In this case, the summary of the study seems mostly correct in summarizing the study, but it naively accepts the study’s findings without any acknowledgement of the significant limitations, several of which were noted by the study’s authors. In fact, the word limitation is nowhere to be found in the CIDRAP article. Bad science journalism like this gets eaten up by people when the headline supports their priors.

If you don’t have time for the full article, know that this study is only about fevers — not Covid infections. Everything about the infections and the transmission is inferred from the timing of measured fevers, which introduces many problems with the data. There was no systematic checking for fever, and not everyone in the household used the smart thermometer and linked app, so many assumptions are being made about how well a recorded fever in the app captures all infections and correlates with the timing of the infection onset.

I honestly didn’t want to spend any time on this ridiculous study, because there are so many issues with the methodology and conclusions, and several are quite obvious to even a casual observer who is willing to think critically. But there’s been so much discussion about it on Twitter over the past two days, I decided to pull together my thoughts into this article. Here are several limitations and considerations that seriously affect the study’s conclusions…

Using Fever as a Proxy for Infection

I’ll start with the most obvious limitation of this study. It uses fever as a proxy for infection. That’s right, participants were never actually tested for Covid-19. Many fevers during the study period were probably due to Covid, but there are still many limitations and issues with inferring index cases from measured fevers. This approach has some significant limitations:

• Some fevers may have been other illnesses. Kids may have infected parents with other things that are being counted as Covid. Young children are more likely to be checked for fever, and have more fevers for various illnesses. This could also explain why children were more often identified as the index case during times when Covid prevalence was low.

• Many infected people do not have fevers. Many people infected with Covid show mild symptoms and may not run a fever. This methodology misses all household transmissions that didn’t result in a documented fever for 2 or more people in the household. Newman Nahas wisely pointed out that this flaw draws conclusions about people infected with Covid based only on data for those who were febrile. This study doesn’t have any data about potential Covid transmission from those without fevers. Perhaps children with fevers are more likely to be an index case than children with Covid who don’t run fevers, either due to their infectivity or family dynamics of caring for a febrile child.

• Not everyone becomes symptomatic on the same timeframe. A parent could have been infected first, but the young child developed a fever first. In this case, the child would be misidentified as the index case. This would also misattribute transmissions from outside the household, where multiple people in the household were exposed together, and a child was simply the first to be identified with a fever.

• Not everyone in the household used the smart thermometer. This misattributed transmissions that began with an adult who didn’t check their temperature. As an example, if Dad works outside the house and brings Covid home from work, but only Mom and kids used the smart thermometer, Dad is missed as the index case. The study notes that in households with multiple people using the thermometer, adult women (57.7%) were much more likely to participate than adult men (37.6%).

Ascertainment Bias & Other Related Issues

This is related to other issues of using fever as a proxy for infection, but I thought it needed a more lengthy discussion. This study was entirely observational, and the number of times people had their temperatures taken varied significantly. A large number of people were only checked for fever a few times, while others had their temperatures checked potentially hundreds of times. Kids are likely to have temperature checks more than parents, particularly when school is in session, and adults are more likely to check their temperature after their kids are known to be sick.

Many parents checked their child’s temperature daily as a requirement for daycare/school attendance. It would be interesting to see how many times people who were considered an index case had their temperature checked versus those who were not considered an index case. This would also impact the timing of when kids were considered the index case. Parents may be more likely to check their children’s temperatures when school is in session, either as a requirement for the school’s daily “health checks”, or because they were simply nervous about their child(ren) getting or spreading Covid at school.

Table 1

As David Zweig pointed out, Table 1 in the study shows that young children (ages 0-8) “on a per person basis, were tested 67% more often than adults” — a median of 5 vs 3 temperature readings. These numbers seem suspiciously low to me, especially considering there were an average of 12.1 readings per person in households with 2 or more participants (516,159 / 6,227,726). Some people had their temperatures checked a lot more than 12 times. Were these people more likely to be identified as the index case? The study did not provide data on the number of readings for the inferred index cases versus others.

The study mentions bias in checking children’s temperatures more than adults as a potential limitation, but then uses some seemingly misleading data to dismiss this concern, stating: “Perhaps parents are more likely to take their children's temperatures than their own, thus overestimating proportions with a pediatric index case. However, in this study, there were more temperature readings per person for adults than children.” Their own data from Table 1 in households with 2 or more participants seems to contradict this second sentence, so if both are accurate, I must assume the sentence in the limitations section is referring to the study as a whole, which included many households where only one adult used the thermometer. However, that data isn’t relevant to their household transmission results, so I’m not sure how it’s a relevant fact to counter that limitation.

Also, Kinsa sells three types of thermometers - traditional style (for oral, rectal, or underarm use), contactless forehead style, and an in-ear style. The study doesn’t provide any data on if one type of reading had a higher rate of identifying fevers. Rectal temperatures are known to be more accurate, and are typically only used in young children, which could further bias the data toward early identification of fever in the youngest children.

Finally, I’m not familiar with the Kinsa thermometers and smartphone app, but if you have the app, are all temperatures recorded in the app, or only ones where the adult chooses to send the data to the app? If so, this leads to an additional confounder where a mild fever in an adult might be ignored initially and not recorded, but then fevers in children were recorded for tracking purposes.

Additional Issues

There are many other issues with the study that need to be mentioned as well:

• This study conflicts with most of the existing literature on Covid transmission from children and schools, so it should be viewed with significant skepticism, even aside from the obvious issues with the methodology. The only reason it was uncritically embraced and promoted is because it supports a preconceived notion that many people have been pushing throughout the pandemic — that children are disease vectors. The study’s corresponding author brought his own bias into this study, which he gave away when he wrote “Not surprisingly, children are very frequently the index case,” in his tweet about the study. The truth is that prior studies found schools were not a significant driver of transmission, that children were not a major factor in household transmission, and that increased age and weight increase the amount of aerosolized virus.

• This study found that only ~15% of infections were from in-home transmission. This finding also goes against much of the existing literature, which further brings into question the rest of the findings. It seems likely that many cases where adults were the index case were simply missed due to many of the limitations described in this article. And even if it were somehow accurate that most infections come from outside the home, that would significantly reduce the importance of who the index case is within households.

• The study attempts to link the percentage of child index cases to school reopening, but the study doesn’t provide any data on geography, which was a huge factor in both school reopening and the seasonality of Covid spread, further confounding the data analysis. Some states opened schools much earlier than others, and Covid waves were very regional, with summer waves happening more in the Southern US. It would be very interesting to see where these thermometers are being used and how their inferred transmission data varied regionally.

• The paper doesn’t say how people obtained the thermometers, but refers to them as “commercially available.” The CIDRAP article states that the researchers “gave out” almost 850 thousand smart thermometers, but I suspect that’s not correct. As a result, we must consider that the population who would buy and use Kinsa smart thermometers (which range from $30-50 on the Kinsa web site) may not be representative of the US population as a whole. If the households are wealthier and more Covid conscious, the adults may have been more likely to work from home, avoid restaurants and large in-person gatherings, etc. This would skew the source of infections to children who leave the house to attend school for this population, versus lower income households where adults were often considered essential workers and who were exposed to Covid a lot more outside the home.

• The paper spends a lot of attention discussing temporal aspects of the percentage of inferred index cases from children, and then trying to tie this to school reopening and lifting of in-school mitigations, but the fever data is already so problematic based on the limitations already discussed, plus there are many other confounders with viral patterns related to increasing immunity from infection, vaccination, variants, geography, seasonality, etc. to even attempt to prove causality, so this entire aspect of the paper is just absurd.

And lastly… So what?

This last point isn’t so much a limitation of the study, but a statement about the value of the study’s supposed findings. Even if children did contribute to ~70% of household transmissions, what should we do with that data? The study authors claim their data suggests “a substantial role of school attendance in COVID-19 spread.” Are people pushing this study seriously arguing for more restrictions on school children? We know one of the authors argued for pre-emptive school closures in March 2020. Are people sharing this study still in favor of school closures to mitigate Covid spread, even after we have seen the extensive damage done to children’s education and mental health? Do they think parents should isolate sick children at home alone, instead of lovingly care for them, without concern if that means a parent might get an illness from their child? Do they want masks mandated in schools and daycares, ignoring the harms of masking children all day, everyday, while they are trying to learn and socialize?

We have always accepted that sending children to school and sports and other activities is important, and that yes, it might mean they get sick. That’s a trade-off most people accept because education and socialization is vital to our children’s futures.