This post was updated on 15th of September 2020
Introduction
As soon as it became clear that children suffer considerably milder disease from SARS-CoV-2 infection than adults, the focus naturally turned to their role in spreading the disease. Children are thought to have been significant drivers of previous influenza pandemics, and so the concern was that COVID-19 would be the same. However, transmission dynamics in children appear to be very different for COVID-19 as compared to flu, as described in a previous blog post. Much has changed since then, so we seek to explore the current evidence here on the transmission of COVID-19 by children.
Background
It is worth noting this issue is very controversial in many parts of the world. There have been political parties who have been extremely dismissive of COVID-19 altogether (including children’s role in transmission), and some prominent scientists who took and early stance on children being significant spreaders of infection. This has contributed to a polarisation of views with debate swinging between those who believe children cannot transmit at all, and those who are sure children are high risk for driving the pandemic. Emotions always run high when it comes to children, and this has been no different.
To compound the issue, studying transmission of COVID-19 in children in inherently difficult, due to a potentially significant (but unknown) proportion of children thought to be asymptomatic, and their contribution to the spread of disease remaining a bit of a mystery. Some gaps remain, but there are now increasing amounts of data to work with.
When we consider risks of transmission, we need to consider 2 classes of factors:
Non-modifiable: Biological susceptibility to acquiring the infection and passing it on, all other things being equal
Modifiable: Environmental/exposure considerations which may increase the risk of acquiring or passing on the infection
How easily do kids get infected?
This is probably the question in which we are the closest to having a good answer. That is because, in household, contact tracing studies we have a natural “experiment” we can examine, whereby most environmental considerations are kept relatively constant. Within a home, everyone is in the same environment, and gets more-or-less the same exposure to the infected individual. We have a number of these studies to draw on now; so many in fact that there are several meta-analyses/reviews looking at the topic (Goldstein et al, Viner et al, Madewell et al, Lei et al), which all come to the same conclusion: children are approximately half as susceptible to acquiring COVID-19 as adults, all other things being equal. Some studies looked, particularly at an age gradient.
There have been some attempts to try and explain this finding by other means, none-of which could plausible explain such a large effect size. Some examples include:
These studies aren’t relevant because children weren’t in school: Whilst this would certainly contribute to the number of children as potential index cases, it has no impact at all on the likelihood of children becoming infected in the home – in fact, if anything it would increase children’s exposure to household contacts since they would have fewer opportunities to be out of the house.
Children are testing negative because it’s too difficult to do nasopharyngeal swabs on them: Sounds plausible, and this will be true to some extent. However even mid-turbinate nasal swabs have been shown to be at least 90% as sensitive as NP swabs. If we assume a worst-case scenario, where all adults have a good NP sample and all children only get a mid-turbinate swab, this could account for a maximum 10% difference. Also, when tracing has been done by serology, the findings have been the same.
Children were not tested and missed because they are asymptomatic: The overwhelming majority of contact tracing studies tested all close contacts regardless of symptoms, and have come to the same conclusion
Children were a missed index case because of being asymptomatic: This is a complex theory that goes as such; a child gets the infection and is asymptomatic but brings it the infection into the home. They infect an adult who becomes symptomatic and gets tested. A contact tracer then comes into their home and tests everyone, but by this time the child tests negative because they have cleared the virus. Whilst an interesting theory, this would require the majority of household transmission clusters to have started with an asymptomatic child. Given our current best estimates are that up to 50% of children are truly asymptomatic, we should therefore see an accompanying large number of symptomatic children as index cases in households, which we don’t. An interesting theory, but not plausible.
There are three things worth noting on this point:
- Whilst children may be less likely to acquire the infection than adults, they can certainly become infected, and there are examples of large numbers of children being infected at institutions such as summer camps or secondary schools (although where precisely the children caught the infection can’t be certain, it seems reasonable to assume significant proportion caught it at these places)
- The relative reduction in susceptibility is multiplicative, i.e. in a transmission chain with lots of children, the relative likelihood of the infection getting passed down several generations is halved at each step compared to adults, i.e an adult transmission chain is 1 x 1 x 1 = 1, where-as for children 0.5 x 0.5 x 0.5 = 0.125 (these numbers are relative as the absolute risk will depend on the setting)
- Whilst the categorization here is often binary (child/adults) this is not how biology works; there is likely a graded increase in susceptibility – however, it’s unclear if this relationship is “S-shaped” (low in young children, rapidly increases in adolescence, high and consistent in adults) or linear (gradually increases from young children to the elderly).
Seroprevalence
Another way to try to determine how susceptible children are to infection is to look at seroprevalence; the proportions of children who have antibodies against SARS-CoV-2 indicated previous infection. The evidence here from the largest and most representative samples have shown from Spain, Switzerland, Italy, Germany, and the UK that children have evidence of lower rates of infection than adults. The differences are generally much more marked in younger children (<10 years) and disappear in older adolescence. The rates of infection tend to be between 50 – 75% of those of adults. This indirect evidence would support the outcomes of the household contact tracing studies that children are less susceptible.
There are several issues here to consider:
- These antibody tests have not been validated in children, so we cannot be completely certain of their accuracy (either picking up false positives due to cross-reactivity with other coronavirus antibodies or having false negatives due to the test processes designed to remove these false positives overcompensating)
- Rates of infection do not necessarily indicate susceptibility, as they may be a product of more or less exposure to the virus (e.g. young, working-age adults have the highest rates of antibody positivity due to higher rates of work/social contacts and exposure, not increased susceptibility)
There are a number of non-representative serosurveillance studies which must be taken with a grain of salt, as these can potentially introduce large biases (e.g. testing the children of healthcare workers, overrepresentation of areas of high incidence).
How infectious are children?
This has proven much more difficult to answer. Several countries have reported very few examples of child to child or child to adult transmission being found, but it’s not clear if this is because very few children were infected anyway, if children were asymptomatic but passing it on unnoticed, or if children are truly less infectious. A few different studies have attempted to help us understand.
Viral loads
Several studies have tried to quantify the amount of SARS-CoV-2 that infected children carry in their nasopharynx. This is generally done by inference from the cycle threshold (ct) count from rt-PCR testing – lower counts imply there was more viral genetic material present.
Results of these studies have been mixed. A small study of a mixed cohort of children from Switzerland found similar viral loads to adults, and confirmed the presence of live, culturable virus. One study looking at only symptomatic children in the US found most of them to have similar viral loads to adults, and younger children to have significantly more. Another US study found similar levels of virus among predominantly symptomatic children (there were some asymptomatic by the VL is not clear from the report, and ignore the strange comparison of viral loads in children at their peak to adult viral loads >7 days into the illness). Interestingly there was no variation according to age, or ACE2 expression. Another study of an undefined cohort of children from Germany found a slightly lower VL in children than adults, and a different, small study from Japan found significantly lower. A recent study including both symptomatic and asymptomatic children from the USA found comparable viral loads regardless of symptoms or age.
These results are somewhat mixed, but we can make some general comments that;
- Viral loads in children do not appear to be significantly different to that of adults
- Viral loads in children do not appear to differ by age or by symptom severity
- The presence of live virus confirms infectious potential of children with SARS-CoV-2
Could the implication be that children are as infectious as adults? Yes, it certainly could – however we must remember there is much more to infectiousness than a merely detectable virus in the nasopharynx. A detectable, viable virus is certainly a pre-requisite for being infectious, but it is only indirect evidence of infectiousness – what we want to know is what happens in the real world.
Clinical infectiousness
This is where we need to focus, as what happens in a lab is of no use if it doesn’t end up correlating with what happens clinically. Unfortunately, we don’t have huge amounts to go by at present, but we have some.
Much evidence is still indirect, such as observations from many countries that few children seem to be identified as index cases in households or as being responsible for many transmission events. For the reasons mentioned above, we cannot read too much into this (although it is more reassuring than finding the opposite)
The best data comes from South Koreas national contact tracing database, from which we (confusingly) have 2 separate publications on the same data regarding children. Both look at the number of secondary infections caused by children, predominantly within the household to calculate a Secondary Attack Rate (SAR) – the proportion of people exposed to the index case who became infected. Park et al also looks at adults and analyses the raw numbers to produce a SAR. Kim et al. take into account that many of the index cases actually shared the same initial exposure as some of their secondary cases, meaning that they most likely both became infected at the same time by the same 3rd party, but as one of them developed symptoms before the other, they were misclassified as an index case. They removed these cases from the analysis.
Please note – the numbers of children relative to adults in these studies are absolutely tiny, so take all findings with a pinch of salt as there is a possibility these cases are not representative of most children with COVID-19.
Park et al found the following household SAR per age brackets; 0-9y 5.3%, 10–19y 18.6%, 20y + between 7% (20-29y) and 18% (70-79y). There was a marked difference in the number of index cases in each age group, with only 29 aged 0-9, and 124 aged 10 – 19 (compared to 1695 aged 20 -29). These results suggest that young children appear to be significantly less infectious than adults, but that children aged 10 – 19 were just as infectious as adults.
Kim et al. corrected for shared exposure. Of the 41 secondary infections from the 248 contacts, only 1 did not share the same exposure, giving an SAR of 0.5% for children <18years in this cohort. Using the same methodology, a different paper found an SAR in adults of 7%. The estimate of 0.5% will almost certainly be biased downwards due to some of the contacts deemed as shared exposure having truly been infected by the child index case, however it seems unlikely this would be by a significant amount (in addition, the rate from younger children was much lower than adults even without this correction)
We cannot generalise the absolute attack rates from this setting, as South Korea uses extraordinary isolation and quarantine measures to prevent transmission from identified cases. We can however look at the relative differences between transmission from children and adults, and this gives us a sense that the risk seems to be lower from children.
In contrast with this data is a pre-print study from Trento, Italy. They again utilise an electronic database of contact networks/tracing of positive cases to determine secondary attack rates from index cases of different ages. An important point here is that the manuscript seems to suggest people were only tested if symptomatic, so there is already a significant bias here (given a significant proportion of children don’t develop symptoms). Also, once more the number of children (14) relative to adults (1475) is very small so take with a big pinch of salt.
The main finding of interest to us is the SAR from a child (<15y) index here was a whopping 22.4%, compared to between 10.6% and 17.1% for other ages. How do we reconcile this with the above?
Firstly, small numbers are prone to extreme results. Second, they haven’t adjusted for exposure type – as children’s contacts will almost exclusively be in the household (not on public transport etc) this will increase the proportion of contacts infected. Third, as we saw from the South Korean data there is a significant issue with shared exposure which can inflate the SAR and hasn’t been adjusted for here. Fourth, there is likely to be a big difference here in exposure of the contacts of infected children compared to adults. All these children are symptomatic. When you have a poorly child in the house, do you isolate them in their room? No. You cuddle them, you wipe their nose, and you clean up their vomit. This is not the same as a poorly adult with COVID-19 who will be sentenced to sleep in the spare room by themselves. Increasing exposure will of course increase transmission rate, not because the child is more infectious, but because a symptomatic child gets more intense contact, not less (in contrast to South Korea, where the children get isolated in a medical facility and their caregivers must wear full airborne PPE including respirator masks!).
In summary, we do not have a huge amount of direct evidence at present. What we do have would lean towards children perhaps being less contagious than adults, although within a household environment when symptomatic being capable of infecting their caregivers to a significant degree.
Are we finding the right children?
The issue of course with almost all the evidence above is case identification. Almost all the children who manage to be identified and included in these studies tend to have been so because they were symptomatic. We know children suffer much milder disease than adults, and increasingly it looks as though more of them are asymptomatic (best estimates currently ~50%).
If we are only studying children with the most severe symptoms, this introduces a huge source of bias to our studies. There is some evidence that the infectiousness of an individual is well correlated with the severity of their symptoms, with asymptomatic people appeared to contribute little to transmission. If we are not studying the asymptomatic children (which maybe half or more of all infected children) then we will not capture this in our studies. This is the evidence gap that currently needs to be filled, which is inherently difficult as how can you identify an asymptomatic child until they’ve infected someone who’s become symptomatic, and once this happens how can you tell who infected who?
Modifiable factors
Because of how different the world has been since the pandemic, with school closures and lock downs, we cannot be certain how environmental and behavioural factors will influence the transmission of COVID-19 by children. Evidence from countries where restrictions have lifted have generally been positive, with no significant upticks in community transmission notable as a result of cautious school re-openings.
An important factor is the number of social contacts. This is one reasons why schools are considered high risk: children mix with lots of other children, which increases the opportunities to transmit to several individuals (thereby increasing the R0). Modelling by Zhang et al seemed to suggest that under normal circumstances, this level of mixing in school would balance out the favourable effects of children’s reduced susceptibility to infection. This however does not account for the possibility of children also being less infectious if asymptomatic.
There is little other evidence to guide us at present regarding other modifiable factors influencing transmission specifically in children; predominantly regarding mode and duration of exposure (and conversely, measures to mitigate these, such as cohorting, ventilation, masks, etc.). Evidence will mainly be derived from studies in adults, which is beyond the scope of this review.
What about schools?
The primary reason that childrens role in transmission has been such a contested topic, is that it is an important factor in decisions regarding the opening or closing of schools as measures to contain the spread of SARS-CoV-2. For pandemic influenza, school closures have been an important staple in managing the spread of disease, due to children’s disproportionate contribution to disease spread (mainly as a function of their lack of pre-existing immunity, meaning they catch it more readily, become more symptomatic, and spread it more efficiently). It is clear however that SARS-CoV-2 does not have the same relative adult:child phenotype to flu. Let’s take a look at the studies so far.
First, let’s take a look at some studies in areas of high community transmission. A study of a secondary school in France using serology to test for past infection, found over 40% of pupils and staff from the school to be positive; much higher than the family members of the pupils (10.9%) indicating that the transmission occurred within the school. Interestingly, only 1 of the 37 children aged 14 or under in this study was positive. From a primary school in the same region of France, only 8.8% of pupils were seropositive compared to 12% of their family members; indicating transmission predominantly occurred within the home. Of the 3 children attending school whilst likely positive, no evidence was found of onward transmission from these pupils. At a private school in Chile, following an index case in a primary school teacher they found 9% of pupils and 16% of staff to be seropositive (unclear how much transmission occurred in the school as opposed to outside). Finally, a secondary school in Israel closed after 2 unlinked cases in pupils with symptoms, and subsequently tested all staff/children and found 13.2% of pupils and 16.6% of staff members positive on PCR.
So this shows us in areas with a large burden of disease and no (or limited) transmission mitigation, significant amounts of transmission can also occur within schools; albeit with staff seemingly more affected than pupils, and younger children less affected than older.
What about in areas with transmission better controlled? Studies from Ireland and Singapore found no evidence of transmission at all from a handful of positive cases in children introduced into school environments. A large study from New South Wales (Aus) found limited evidence of transmission within schools; from initial 25 cases (15 children and 12 adults) only 18 secondary infections were identified, despite 44% of contacts being screened regardless of symptoms. The highest secondary attack rate was found between adults (4.4%). Following cautious reopening of schools in the UK, from over 20,000 institutions serving >1mil children, only 30 outbreaks (consisting of 2 or more cases) were identified in schools, alongside 67 isolated positive cases. Of the 30 outbreaks, 22 consisted solely of transmission from an adult (either to other adults, or to a child). They also found outbreak size to be strongly correlated with levels of community transmission. In a study of child care facilities in Rhode Island, in 29 facilities which identified positive cases there was no secondary transmission within the facilities in 22 of these. Note should be made that all these studies (except Ireland) were undertaken with precautions in place; either social distancing/small class sizes, mask wearing, cohorting, or closures with test, trace and isolate in place.
There has been significant discussion about whether it ‘safe’ for children to return to school. The studies discussed so far make it fairly clear that ‘safe’ is the wrong lens through which to view this debate. The contextual factors are significant behind these decisions and it would be better to stratify via degrees of risk rather than binary “safe or not safe”. In situations where some degree of social distancing can be assured (large school estate, small class sizes) and low rates of COVID19 in the community, it would seem logical that low susceptibility and low chance of infectivity would mean that children are low risk to return to school. However where there are high rates of community transmission, crowded schools with older children, then it is more likely to see spread within schools. This doesn’t necessarily mean the latter is unsafe, just that transmission potentially more likely. The frequency of adult to adult transmission seen in countries where schools have reopened still suggests that measures should be directed here to avoid spread rather than be consumed with concern about children.
Conclusion
- Children are approximately half as susceptible to acquiring SARS-CoV-2 as adults given the same exposure. This is most clear for younger children (<10yrs) and an increases during adolescence to adult susceptibility.
- Children have highly variable amounts of SARS-CoV-2 virus detectable in their nasopharynx, broadly similar to that of adults (often from samples of children which may not be representative).
- Children may be less infectious than adults, but there is little direct evidence.
- There is emerging evidence that asymptomatic individuals may be less infectious than those with symptoms, which given potentially high rates of asymptomatic infection in children may reduce their contribution to community transmission.
- It is unclear how environmental/modifiable factors will contribute to children transmitting COVID-19, and this will likely depend on international variations in social restrictions and infection prevention measures deployed in schools.
I don’t see anywhere on this webpage, the date this article was published?
That would be useful information.
Are all “Don’t Forget the Bubbles” undated?
Agree, please update with new studies and findings!
Do the authors intend to update this?
Plain and simple children are as likely to catch it and spread it as adults….they are simply not being picked up by testing. They are not being picked up by antigen tests because they are not presenting with symptoms and because of the practical difficulties of swab testing young children. This may be because the immature immune system according to one report is more suited to deal with novel infections and may explain why children are generally asymptomatic and may have a smaller window for positive antigen testing.
Secondly recent reports suggest that asymptomatic cases do not develop a sufficient antibody response to provide a positive antibody test. So on both fronts positives are being missed when children are tested which leads in my view to erroneous conclusions in regard to their susceptibility to contacting the virus and subsequent ability to spread it.
Our data show that initial VLs at diagnosis in symptomatic children is comparable to those in adults, and that symptomatic children of all ages shed infectious virus in early acute illness. Infectious virus isolation success was largely comparable to that of adults, although two specimens yielded an isolate at a lower VL (1.2×104 and 1.4×105 copies/ml) than what was observed in adults. SARS-CoV-2 shedding patterns of culture competent virus in symptomatic children resemble those observed in adults. Therefore, transmission of SARS-CoV-2 from children is plausible.
Hi Alisdair / Damien,
thanks for the good work. What is your thought on this study?
https://www.medrxiv.org/content/10.1101/2020.04.27.20076778v1
All the best
Kai
I think that the above discussed high profile study by Jones and collegues has far more concerns and limitations than mentioned and I strongy agree that it actually provides evidence in favour for the notion, that the viral load differs between children and adults (for many reasons). I just want to add one of these reasons: As mentioned above in this study a significant Kruskal-Wallis test was obtained, indicating differences regarding viral load between age groups. Dunn‘s post hoc test indicates significant differences between age groups „Kindergarten“ and „Mature“ as well. This result is simply ignored by their conclusion (it is not even discussed in the results or discussion section). Their conclusion seems to be based primarly on the non-significant Tukey HSD post hoc analysis for which the requirements (e.g. homogeneity of variance) ist not met and therefore should not have been conducted at all. I‘m stunned, that these (obviously wrong) conclusion of this study is cited so many times (at least in Germany) without beeing criticized for all its methodological errors and limitations.
Thanks Alasdair. If children are more likely than adults to have asymptomatic infection then it seems possible that testing only symptomatic people may make it appear that the attack rate in children is less than in adults. It would be good to see more studies that indicate whether the testing was on asymptomatic children but as you say the data so far looks good for children.
Thanks for your reply Maryza. Whilst the issues you mentioned might influence the absolute attack rates in the studies, they shouldn’t influence the attack rate of children relative to adults, as the same testing procedures were performed on all cohorts. Therefore the OR should be uninfluenced. Hope that makes sense!
Hi Alasdair, thank you for a great summary. Regarding “How easily children catch the disease?” it seems that all contacts in the Shenzen study were tested by PCR at start and end of quarantine period, contacts in CID paper tested at start and midway through quarantine and in Science paper were tested once but it’s not clear when. For the other two studies it’s not clear to me if all contacts had PCR done irrespective of symptoms or if the attack rate is a clinical attack rate. So this may explain why the attack rates in children vary in these studies? Thanks
comorbidity children has to be less no ? type 2 diabetic etc
children run around lick thinks hug each other suck fingers pick noses , shout and scream cry share crayons share seats and desks spit food when they eat ,,dont wash hands correctly , cant even look after PE kit let alone a mask , share food drinks sweets , snog , only 2 % i think are at school at this time in England , stand by your beds when that is even 20%, Talking about receptors and studies etc, forgets that kids are kids you try changing that , they get ill quick and recover quick try testing for that let alone trying to protect a teacher from a vomiting child without a military grade NBC suit,, and the effects of the disease may not present as they do in adults could be a multi organ inflammatory response , not a clue ,, last time i checked is not that ethical to test on children and the approach in England seems like that ..
anyone with kids know they have many viral infections recover in 12 mins are back eating ice cream and you get “it” and are buggered for 3 weeks , i have no idea take care all
Thanks for your comment Rob. Yes I have seen the article. Many details are unclear at the moment, but we should not be surprised to see examples where significant numbers of children become infected. If anyone is exposed for prolonged periods of time, they will become infected. It is best to consider the evidence as a whole rather than to focus on individual anecdotes to help keep perspective. Thanks again!
So in summary, despite over 3 million confirmed cases, we are really not seeing outbreaks spread by kids. In the Montreal example (quoted above) of 12/27 kids likely reflects they caught from an infected worker, not each other. This is compelling Epidemiology to say kids are not super spreader, if spreaders at all.
Just a news article and a small sample, but this is concerning: https://montrealgazette.com/news/local-news/coronavirus-12-children-four-employees-infected-at-mascouche-daycare/wcm/9655f3d1-79c0-4efe-8f92-1f72883c5cf4/ . Outbreak in a daycare centre for children of health workers. Attack rate of 44% among the children.
Agree. Fantastic review. Cheers from Costa Rica
Thanks a lot “mate”
Great review
In the German paper you find the following quotation:
“To enable an estimate of infectivity in children, we analyzed viral loads observed during routine testing at a large laboratory testing centre in Berlin (Charité Institute of Virology and Labor Berlin). Charité Institute of Virology was the first laboratory qualified to test for SARS-CoV-2 in Germany and until early February 2020 was the only SARS-CoV-2 testing facility in Berlin, a city of ca. 3.8 million inhabitants. Labor Berlin is a large medical laboratory services provider in Berlin, owned by the senate of Berlin and serving Charité as well as other large hospitals in Berlin and beyond. Labor Berlin serves public testing centres that mainly see adult outpatients. It also tests out- and inpatients from several hospitals, and serves practitioners and public health agencies submitting samples taken during household-based contact tracing.”
I understand it as saying that the samples came from all kinds of tests, both community and hospital based.
Thank you so much for putting this together. When I saw the data from the German study and realized there was at least a 2Log difference in viral load between the highest value in the under 6yr old group and highest value in the over 45yr old group (or even the over 25yr old group), I realized there was something odd with the statistics and conclusion.
I think these data present us with an opportunity to dial in the particular innate mechanism that may be at play here. Or could it all be related to lower ACEII receptor in children? Will someone care enough to investigate this?
We should be counting our blessing children are not as affected or symptomatic.
The usual adult pick up rate of 70% likely to be lower in children as the smaller they are difficult for us to get a swab from the back of throat. In children they seem to be shedding the virus in their stools and if they are combined with nose and throat swabs should result in better pick up rates. Children are also recently presenting in toxic shock, typical and atypical Kawasaki’s with some having significant cardiac pathology.
Thank you for your comment Emma.
I am currently not aware of any studies specifically assessing test characteristics of swabbing in children. I have not seen anything to suggest 50% specificity of swabbing, including from adult literature.
Most respiratory viral PCR tests are assumed to perform as well in children as adults (operator dependent of course).
Alasdair
Is there any evidence about the sensitivity of the tests in children?
Our microbiologist says that the swans have a 50% false negative rate. So are very unreliable.
Is there any information about how reliable they are in children? Could it be harder to get a positive swab?
There is also some confusion about timings of swabs. Could this be an issue in children?
Both of these would mean we are getting inaccurate data.
I can see no information about this in the studies.
Emma Blake
Paediatric Consultant