Roland, A. The missing link? Children and transmission of SARS-CoV-2, Don't Forget the Bubbles, 2020. Available at:
There is a huge amount of international interest currently in the role of children in the transmission chain of COVID-19, as many countries are looking to relax measures of lockdown including the possibility of school reopening. Despite significantly fewer cases of severe illness in children, and fewer detected cases in total, there are concerns about children being silent vectors of the disease and spreading it in the community.
Due to global interest, there has unsurprisingly followed a lot of hype, and some misleading claims. This post aims to outline all the evidence available so far on the potential role of children in transmission of COVID-19.
What do we need to know?
In order to understand children’s role in transmission, there are three things we need to consider:
- How easily children catch the disease
- How many children have the disease
- How infectious a child is once they have it
Let’s take a look at the evidence for each of these in turn
How easily children catch the disease?
On this front, we have five studies (three published and two pre-print) to help inform us. These studies all look more-or-less at the same thing, which is contact tracing. From cases that have been confirmed positive (an index case), they trace back all the people who that case has been in contact with over the recent past and test all of them for COVID-19 to see how many of them caught the illness from exposure to that index case. The proportion of people who have had contact that subsequently became infected is referred to as the Attack Rate (AR). Broadly speaking contacts can be split into two groups: household and non-household (this is important as obviously you are much more likely to transmit to someone in your house). We can also split them up according to age, and see if there is any difference in the number of children who catch the illness compared to adults.
A study from Shenzhen in China was the first to be released in pre-print in March and is now published in the Lancet ID. This study assessed 1286 contacts of 391 initial cases and showed children had a similar attack rate to the population average (7.4% vs 7.9%), but interestingly were much less likely to be symptomatic. This finding caused a lot of concern, but more data has emerged since.
A pre-print study from Japan was released shortly after. They examined 2496 contacts of 313 domestically acquired cases and found a much lower attack rate in children (7.2% males, 3.8% females) compared to adults (22% in people aged 50 -59 years).
Another pre-print study from Guangzhou in China examined 2017 close contacts of 212 confirmed cases. The overall attack rate was 12.6%, however, the attack rate in children was 5.3%. They calculated an odds ratio of acquiring infection in children of 0.27 (0.13 – 0.55) compared to adults >60 years of age.
A study published in Clinical Infectious Diseases assessed household contacts in particular. They assessed 392 contacts of 105 index cases in Wuhan, China (they had more stringent eligibility criteria to ensure they had correctly identified the index case in the household i.e. the person who brought the infection in). Of the 100 contacts under 18 years of age, only four became infected. This was compared to an attack rate of 21.9% among adult household contacts (making an overall attack rate of 16%).
A further study published in Science included some far-reaching assessments of transmission, but for our purposes, we will look at their findings regarding secondary attack rates in children. This was a contact tracing study from the Hunan CDC in China. They assessed 114 clusters (some clusters had more than one index case) and 7375 contacts. The authors performed regression analysis to adjust for other factors that influence AR (the type of transmission, travel history, etc) to determine the odds of becoming infected at different age groups. They found an odds ratio of 0.34 (0.24–0.49) for children under 14 years, compared to the reference group of 15-64 years (consistent across models).
Below is a table summarising the findings of all these studies.
|Author||Journal||No of index cases||No. of contacts||Secondary AR (children)||Secondary AR (adults)||OR secondary AR (children v adults)|
|Bi, Q||Lancet Infect Diseases||391 (32)||1286||17/233 (7.3%)||67/837 (8.0%)||0.82 [0.48-1.43] †|
|Zhang, J||Science||114 (1*)||7375||47/756* (6.2%)||606/6437 (10.4%)||0.34 [0.24-0.49]$|
|Mizumoto, K||medRxiv||294 (10)||2496||10/176 (5.6%)||153/697 (21.9%)||0.21 [0.11-0.41] †‡|
|Jing, Q-L||medRxiv||335 (5)||1938||10/244 (4.1%)||127/5831 (6.2%)||0.27 [0.13-0.55] ¥|
|Li, W||Clin Infect Dise||105 (unknown)||392||4/100 (4%)||60/292 (20.5%)||0.16 [0.06-0.46] †|
† Unadjusted OR
- Under 15 years
‡ This is based on assumption of peak rate for the 50-59 age groups applied to all adults (other data unavailable)
$ Adjusted OR in GLM for increased household exposure of children
¥ Children Vs adults age >60 years
Thank you to Grace Leo for help compiling this table
For comparison, see this study regarding influenza where children <5yrs had the highest secondary attack rate (22.6%).
How many children have the disease?
This is more difficult to assess. Most studies looking at case rates of COVID-19 have focused primarily on those presenting to hospital unwell with symptoms, and have tested them with PCR of nasal/oropharyngeal swabs. This risks missing out on a large population of children, if there were a large number of infected children in the community with few or no symptoms.
Where we can look for this data is in countries or populations where they have undertaken much wider community testing. South Korea for example has much more extensive community testing, and still has a minority of cases in children (5- 6%).
Iceland tested 6% of their entire population and found dramatically lower numbers of cases in children, including 6.7% children under 10 positive in “targeted testing” (symptomatic or high risk due to contacts) compared to 13.7% of those 10 and older, and found 0 children under 10 years positive in population screening (by invitation) compared to 0.8% of those over 10 years.
The Italian principality of Vo tested >85% of their population following their first death from COVID-19, and found no positive cases in children despite 2.6% of the population being positive. This finding was repeated when they tested again two weeks later – despite a number of children living in households with confirmed positive contacts.
Finally, a study in The Netherlands is undertaking community serology testing (looking for antibodies against SARS-CoV-2 as evidence of current or previous infection) and has released preliminary results. They have found 4.2% of adults are positive compared to 2% of those aged <20 years.
How infectious a child is once they have it
This is almost impossible to tell at the moment, as we have no direct experiments comparing exposure to an infected child to exposure to an infected adult – in particular as children appear to make up a small number of index cases.
For example, in the study from Guangzhou mentioned above, only 5% of index cases were noted to be children. In the previous study from Hunan, of the 114 clusters only a single child under 14 years of age was the index case in the household. In a study of an international collection of family clusters of COVID-19, a child was found to be the index case in a household in only 10% of clusters.
An issue with this data is that given schools were closed early in the outbreak, it is possible that this had an impact on the likelihood of a child becoming infected outside of the home and therefore becoming a household index case. Children rarely mix outside of the home other than at school.
What about viral load?
A high profile study emerged in the press recently examining viral loads of children and adults from Germany. This received a lot of press for its conclusion that there was no difference in viral loads between age groups, and that this would imply children are as infectious as adults. Due to the high profile nature of this study, we will consider this in detail.
What did they do? This is hard to say as their methods section is sparse, however, in brief, they assessed the viral load of all positive samples from patients with SARS-CoV-2 at some German hospitals by comparing the ct values (a surrogate measure for viral load based on PCR) and comparing these across arbitrary age groups.
Firstly, it is important to note there is far more to how infectious an individual is than merely their viral load as determined by PCR (which also detects dead virus, as opposed to viral culture), so it is not good practice to make many assumptions past this point.
Second, there is a big issue in determining which populations these samples have come from. We do not know if these are all symptomatic cases who were unwell and presented to hospital, or if this is taken from a community setting. This is extremely important, particularly given the most striking part of their main figure:
Note how few samples from children there are compared to adults. Let’s consider who this population might be, and assume their conclusion (there is no association between viral load and age) is correct.
- Symptomatic cases presenting to hospital: Then the correct conclusion is, “among patients with COVID-19 who were unwell enough to develop symptoms and present to hospital, there appears to be no difference in viral load. As there are considerably fewer children, it is possible many children with lower symptom burden are present in the community who may have different viral loads”.
- Community testing: Then the correct conclusion is, “among community populations with COVID-19, of those who test positive there are similar viral loads. We note there are considerably fewer children than adults, demonstrating that although viral loads are comparable, fewer children appear to be infected”.
This all stems on their conclusions being correct, however, there is another highly significant issue with their data analysis. The study question is whether there is an association between viral load and age. Age is a continuous variable, meaning it is a number that goes up continuously and doesn’t strictly have categories. If you make categories from it, you dramatically reduce the statistical power to detect differences. This is what they have done. Not only that, but they have applied a highly punitive correction for multiplicity (making lots of comparisons, which increases the probability of finding a statistically different difference by chance) which also dramatically reduces the chances of getting a “statistically significant” result.
Of note, in their paper they actually do use something closer to a more appropriate method of analysis which is a Kruskal-Wallis test. What did they find? A statistically significant association between age and viral load (p=0.01). This is not surprising if you actually plot the means of their data (courtesy of Johannes Textor)
Even by this inappropriate method of analysis, it looks extremely likely that if children weren’t so underrepresented in their sample the result would have been statistically significant.
What about schools?
Specific evidence in regards to transmission in schools is lacking, due to rapid shutdown at the start of the pandemic. A systematic review of the impact of school closures on the transmission of SARS and COVID-19 found only equivocal evidence for their impact in controlling transmission.
A study from an outbreak around a French secondary school has received some attention, as they found 40% of pupils and staff became infected with no difference between the two groups. What is important in this study, is that almost all the students in the study were aged 15-17 years of age, who appear to have similar disease characteristics to adults. Of the children 14 and under, a very small proportion got infected (it’s not clear how many were students and how many were family contacts). We cannot derive useful information from this study about younger children at present.
Another study from New South Wales in Australia demonstrated very low rates of infection in school children and low rates of spread, however, the absolute numbers are low (18 cases total, 12 secondary and six primary) so again it is difficult to draw firm conclusions from this study regarding spread in schools, despite the data being reassuring.
From the studies listed above we can determine:
- Children appear significantly less likely to acquire COVID-19 than adults when exposed
- There is reasonable evidence that there are significantly fewer children infected in the community than adults
- Children are rarely the index case in a household cluster in the literature to date
- It is not clear how likely an infected child is to pass on the infection compared to an infected adult, but there is no evidence that they are any more infectious
The most parsimonious explanation for all the above seems to be that children are less susceptible to becoming infected, therefore fewer of them have become infected, there are subsequently fewer infected individuals in the community, and children have therefore infrequently brought the infection into their homes.