Household transmission of seasonal coronavirus infections: Results from the Flu Watch cohort study

Background: In the context of the current coronavirus disease 2019 (COVID-19) pandemic, understanding household transmission of seasonal coronaviruses may inform pandemic control. We aimed to investigate what proportion of seasonal coronavirus transmission occurred within households, measure the risk of transmission in households, and describe the impact of household-related factors of risk of transmission. Methods: Using data from three winter seasons of the UK Flu Watch cohort study, we measured the proportion of symptomatic infections acquired outside and within the home, the household transmission risk and the household secondary attack risk for PCR-confirmed seasonal coronaviruses. We present transmission risk stratified by demographic features of households. Results: We estimated that the proportion of cases acquired outside the home, weighted by age and region, was 90.7% (95% CI 84.6- 94.5, n=173/195) and within the home was 9.3% (5.5-15.4, 22/195). Following a symptomatic coronavirus index case, 14.9% (9.8 - 22.1, 20/134) of households experienced symptomatic transmission to at least one other household member. Onward transmission risk ranged from 11.90% (4.84-26.36, 5/42) to 19.44% (9.21-36.49, 7/36) by strain. The overall household secondary attack risk for symptomatic cases was 8.00% (5.31-11.88, 22/275), ranging across strains from 5.10 (2.11-11.84, 5/98) to 10.14 (4.82- 20.11, 7/69). Median clinical onset serial interval was 7 days (IQR= 6-9.5). Households including older adults, 3+ children, current smokers, contacts with chronic health conditions, and those in relatively deprived areas had the highest transmission risks. Child index cases and male index cases demonstrated the highest transmission risks. Conclusion: Most seasonal coronaviruses appear to be acquired outside the household, with relatively modest risk of onward transmission within households. Transmission risk following an index case appears to vary by demographic household features, with potential overlap between those demonstrating the highest point estimates for seasonal coronavirus transmission risk and COVID-19 susceptibility and poor illness outcomes.


Background
Understanding household transmission parameters is important for outbreak modelling and response as coronavirus disease 2019 (COVID-19) becomes established in communities worldwide. While studies are currently underway, developing robust estimates for household transmission parameters will take some time. Evidence from other human coronaviruses may therefore be useful, given similar routes of transmission 1,2 .
Outbreaks of other emerging coronaviruses, i.e. severe acute respiratory syndrome (SARS) and Middle East respiratory syndrome (MERS), appear to be primarily based in health care settings without substantial community transmission and household transmission studies are limited. Estimates of transmission risk to at least one other household member following a SARS index case range from 12.3%-13.5%, with estimated secondary household attack risks between 4.6%-10.2% 3-5 . In index cases, younger age and healthcare worker status were associated with lower risk of household transmission. A single study of MERS found that 44% of households with an index case experienced onwards transmission, with a secondary attack risk of 24% and risk factors for developing secondary infection in contacts including being adult, male, and having long-term conditions 6 .
Seasonal coronavirus infections also appear to cluster within households 7 , with proportions of onwards transmission to at least one household member following child index cases ranging between 8% and 33% 8,9 . A recent community cohort study based in Michigan 10 found secondary attack risks ranging from 7.2%-12.6% for circulating seasonal coronavirus strains and clinical-onset serial intervals between 3.2-3.6 days.
Here we aimed to report key characteristics of seasonal coronavirus household transmission in a population-based UK cohort, using data from the Flu Watch study 11 . We present the proportion of laboratory-confirmed seasonal coronavirus infections acquired within and outside the household, the household transmission risk and secondary attack risk for symptomatic cases, the clinical-onset serial interval, and transmission risk stratified by demographic household features.

Method
Study design and procedures Data were drawn from the Flu Watch prospective cohort study of acute respiratory infections in English households 11 . We included the three winter seasons (2006-2007, 2007-2008, 2008-2009), during which all samples were systematically tested for coronaviruses. The Flu Watch study methodology and cohort profile are described in detail elsewhere 11,12 .
Whole households were recruited annually to the study following random selection of a household member from GP practice lists. From 2008-2009, participants from the previous cohort were also re-invited to participate. Inclusion criteria were that all household members agreed to participate, and all members over 16 agreed to provide blood samples for other Flu Watch research. Exclusion criteria were household size >6 members, severely incapacitating or terminal illness in any household member, and heavy involvement in other research 12 .
Participants provided demographic data at seasonal baseline following recruitment, and were followed-up weekly via telephone or online throughout each season to report any symptoms of acute respiratory infection. If participants experienced any symptoms, they were requested to provide a nasal swab on the second day of illness and to provide a daily diary of symptoms from the first day of illness until the symptoms resolved. Realtime PCR was carried out to screen nasal swabs for a panel of viruses, including three circulating seasonal coronavirus strains (229-E, NL63, and OC43) 11 .

Definitions and analyses
Guided by an estimated incubation period of 2-5 days and further 2-18 days of symptomatic illness with viral shedding for seasonal coronaviruses 13,14 , we defined index cases as the first PCR-confirmed infection in a household or >23-days following a prior case, co-primary cases as potential index cases of the same strain arising within ≤2 days, and secondary cases as infections with the same strain occurring >2 days and ≤23 days from exposure to the index case(s). We assumed that household transmission to any secondary cases had occurred if the strain was unknown, but criteria for transmission were otherwise met.
We estimated the proportions of total PCR-confirmed coronavirus cases in the study likely acquired in the community (i.e. index, co-primary, or single-person household cases) and acquired in the household (i.e. secondary cases). The proportions were weighted to the English national structure of age and region 11 .
For households in which secondary transmission was possible (i.e. excluding single-person households and episodes of co-primary infection affecting all household members), we calculated the symptomatic household transmission risk-the proportion of households that experienced at least one secondary case -and the household secondary attack risk -the proportion of total exposed participants who became secondary cases. To avoid multiple inclusions of the same episode where there were co-primary cases, transmission risk and secondary attack risk were estimated based on a randomly-selected single index case. Due to the testing protocol, it was not possible to detect asymptomatic primary or secondary cases. We calculated the clinical-onset serial interval as the time in days from onset of reported symptoms in the index case(s) to onset of reported symptoms of the secondary case(s). All co-primary or multiple secondary cases in this cohort reported symptom onset on the same day.
We stratified the household symptomatic transmission risk by the following household features potentially relevant to transmission: the age structure of the household (adults between 16-64 only, older adult(s), adult(s) and children), age of index case (child <16, adult 16-64, older adult 65+) for households with children (not stratified for other household structures due to limited variation), the number of people and number of children in the household, whether the household contained current smoker(s), index of multiple deprivation, long-term health conditions in household contact(s), and the index sex, index healthcare worker status and index transmission-preventive hygiene behaviour. Index case hygiene behaviour was classified as a binary variable according to adherence to prevention guidelines 15,16 i.e. covering the mouth while coughing or sneezing, using a single-use tissue, and washing the hands habitually after coughing or sneezing and at least moderately-frequently throughout the day (>=5 times according to median split and previous literature 17 ). Due to the small number of household transmissions, we present descriptive stratified analyses only and do not describe by strain sub-groups. Where features of the index case or index-exposed dyads were investigated, we included only those households where there was a clear index case rather than co-primary cases. Any instances of missing demographic data are noted in Table 1.  7.0 (6.0-21.0, 5.0-21.0) 7.0 (6.0-7.5, 6.0-8.0) 6.5 (6-8.75, 6.0-14.0) 7.0 (6.0-9.5, 5.0-21.0) *All strains include cases where strain was unknown (n=3); **due to unknown index strain, n=5 exposed contacts were included with no outbreak strain recorded.
To estimate potential non-detection of household transmissions due to non-adherence to swabbing protocol, we measured the proportion of exposed participants who reported symptoms within one month of an index case but did not submit a nasal swab. Analyses were performed in Stata Version 15. . Due to unknown index strain, outbreak strain was unavailable for five exposed participants.

Proportion of infections acquired in the households and features of transmission
The median clinical-onset serial interval was 7 days (IQR = 6-9.5 days, range 5-21 days). This median value was consistent for 229E and NL63, with some between-strain variation in range, and similar to the OC43 median serial interval of 6.5 days. There were two households for which the exact date of symptom onset of the exposed case could not be traced (only the beginning date of the week of illness), which were excluded from the serial interval calculation. Adherence to swabbing protocol during household outbreaks Among the 275 exposed household members, there were 75 distinct episodes of respiratory symptoms within 23 days of a household coronavirus index case. For 4 of these episodes (5.33%), the participant did not provide a swab at any point during their illness.

Discussion
The study describes a range of important characteristics of seasonal coronavirus transmission that have not previously been reported. The great majority of infections (91%) were acquired outside the household. Following a symptomatic index case within the household, onward transmission to at least one other household member occurred in 14.9% of households, with a secondary attack risk of 8.0%. Risk of onwards household transmission fell within previous estimates for seasonal coronaviruses 8,9 and were higher than the limited literature suggests for SARS 3-5 , but lower than for MERS 6 . The secondary attack risk fell within the range identified for seasonal coronavirus strains in a recent US-based longitudinal cohort study 10 and was somewhat lower than preliminary estimates of the household secondary attack risk for COVID-19 -ranging between 10.5% in USA 18 , 13.6% in Taiwan  Demographic and health-related characteristics with the highest point estimates for symptomatic transmission risk overlapped with risk groups for susceptibility and/or poor outcomes in COVID-19, namely households comprising older adults, current smokers, exposed members with chronic illness, and those in deprived areas 29-31 . Although children had a higher point estimate for transmission risk, index cases were distributed relatively evenly across age groups, in contrast with a Michigan community cohort study in which children comprised the majority of index cases for most strains 10 . Good hand and respiratory hygiene by index cases may be helpful for disrupting transmission, possibly overlapping with the lack onwards transmission from healthcare workers. The findings should be interpreted with caution given relatively small numbers of positive transmissions and, related, wide confidence intervals for transmission risk estimates. We were underpowered to conduct multivariate analysis required to disentangle the interrelationships between potential demographic and health-related risk factors and this is a relevant area for further research concerning seasonal and pandemic coronaviruses.
This study had a number of limitations. Data were limited to winter seasons. Only symptomatic cases were swabbed and detected. While there appeared to be good adherence to the swabbing protocol among household members exposed to an index case, five percent of those who developed symptoms following exposure failing to provide a swab so secondary transmission may be underestimated. It is possible that the nasal swabbing protocol, which was developed primarily for influenza detection, did not provide ideal sensitivity for these strains of coronavirus. We also cannot exclude that index cases in the initial weeks of the study could have acquired the case through undetected household transmission. The analyses did not account for background rate of infection in the community and consequently some apparent secondary cases may have been acquired incidentally outside of the household. All risk factor analyses were limited by a relatively small number of onward transmissions.
To our knowledge, this study presents the first investigation into household transmission in seasonal coronaviruses in a population-based UK sample. In typical winter seasons with individuals freely able to continue typical activities and social contacts, household transmission appears to be lower than in COVID-19. However, this difference may reflect different levels of population immunity and behaviours rather than intrinsic differences in the transmissibility of seasonal coronaviruses and the pandemic strain. This project contains the following underlying data: -Household_CoV_acquired.csv (data required to compute the proportion of cases presumably acquired outside of the household versus and the proportion acquired from household transmission. Each row represents an anonymised PCR-confirmed seasonal coronavirus case) -Household_CoV_acquired.dta (above data file in .dta format) -Household_CoV_TransmissionRisk.csv (data required to compute the risk of symptomatic onward household transmission following a seasonal coronavirus index case, and perform stratified descriptive analyses) -Household_CoV_TransmissionRisk.csv (above data file in .dta format) -Household_CoV_SAR.csv (data required to compute the seasonal coronavirus secondary attack risk overall and by strain. Each row represents an anonymised exposed-index pair from a given outbreak) -Household_CoV_SAR.dta (above data file in .dta format) -HH Transmission Serial Interval.csv (presents available, anonymised data required to compute the median clinicalonset serial interval overall and by strain for each household outbreak) This is a well-written and timely article on household transmission of seasonal coronaviruses with implications for the current pandemic. This presents original work using a population-based cohort in England which has previously been described. This prospective cohort was contacted weekly during the winter season, which is when seasonal coronaviruses circulate. Symptomatic participants in each household kept a daily log of symptoms allowing for granular data collection. PCR and strain level characterization of coronavirus diagnoses were included which was a major strength. Household transmission risk is presented and stratified by relevant characteristics including household structure, number of children, smoking, and hygiene. The findings include description of clinical serial interval, secondary attack risk, and household transmission risk stratified by strain type. The comparison of this study's household attack rates to other coronaviruses (including SARS and MERS) is very valuable and adds to the cogwheel of data informing household transmission of COVID-19. The authors do a good job of noting caution in interpretation of results given that the risk factor section is underpowered with overlapping confidence intervals.

Major Concerns:
A particularly interesting result is the increased risk of transmission among males, households with older adults, smokers, and chronic illnesses which is also seen as risk factors for severe disease in COVID-19. But looking at Table 1, it is difficult to know if these trends are statistically significant given the overlapping confidence intervals. Additionally, for "older adults in the household" it is unclear whether this may simply be a proxy for more individuals in a household or for having children in the household (such as a multigenerational family). The limitation of wide confidence intervals is raised in the discussion section, but this potential relationship also appears in the abstract conclusion as a major finding. Given that there are so many caveats and need for further work in this area, I suggest removing the finding from the abstract and keep it described with all the caveats in the discussion section. The fact that there were no household transmission when the index case followed hand and respiratory hygiene protocols seems to be a really important data point and finding of the study to highlight.

Minor Concerns:
Please define "deprived areas" or IMD (indices of multiple deprivation) used in Table 1 and comments in the discussion (bottom left paragraph) in the methods section. 1.
In the methods section, the inability to identify asymptomatic primary or secondary cases is raised. It would be helpful to include this limitation in the discussion section and provide any underlying literature on asymptomatic transmission of seasonal coronaviruses. This is an important piece in how such data is linked to transmission of the current pandemic strain.

2.
One of the strengths of the paper is the description of strains, but the discussion does not include any background on known differences or comments/hypotheses regarding whether strain type may play a role in household transmission risk.

3.
When discussing "households with older adults", does this also include children and could this simply be a proxy for more people in the household?

4.
It may be valuable to discuss possible background immunity to common hCoVs a bit more in the discussion, and its possible impact on household transmission findings. This is where there is also a big difference between the common hCoVs and SARS-CoV-2. Authors allude to this throughout, but it may benefit from some attention in the discussion section.

If applicable, is the statistical analysis and its interpretation appropriate? Partly
Are all the source data underlying the results available to ensure full reproducibility? Yes

Are the conclusions drawn adequately supported by the results? Yes
Competing Interests: No competing interests were disclosed.

Melissa Rolfes
Influenza Division, Centers for Disease Control and Prevention, Atlanta, GA, USA The authors provide a report on seasonal coronavirus infections detected in the Flu Watch cohort in the UK and examine household and demographic features associated with subsequent infection in other household members. In general, the rationale and implications of the study are clearly described and further elaboration of the methods and results would strengthen the manuscript overall. This is important work for the public health community, grappling with the ongoing SARS-CoV-2 pandemic; however, one limitation of the Flu Watch study, and its conclusions and relevance for the current pandemic, are that all detected infections were symptomatic because only symptomatic participants contributed nasal swab specimens for PCR testing. I look forward to further studies that can build on this study design to innovatively capture asymptomatic seasonal respiratory virus infections to better understand the drivers behind onward transmission.
Specific comments that I hope the authors can address in the next version are below.
Major comments Please include further description of the weekly telephone/internet survey, including the list of symptoms that were asked about. The description should include enough detail to answer the following questions: Was the question addressed to one member of the household or did each member have to respond about symptoms? Was the list of symptoms different if the participant was a young child (and below what age were these different symptoms solicited)?

○
The methods section did not include a definition of the indices of multiple deprivation. A short description of what the indices are based on would be important for re-creating this work and analysis. Additionally, a reference to the index would be helpful for readers.

○
Please include the method you used to estimate the confidence intervals for the household transmission risk, the symptomatic secondary transmission risk, and the community infection risk…well, all really.

○
The results section could be elaborated with more information about the cohort size. How many people (and how many households) participated in each season? What were the demographic, household, geographic, and socioeconomic characteristics of the participants?
In the 3 rd paragraph of the results section, the authors state that 22 households had transmission (out of a possible 134) = 16.42% household transmission risk. But in Table 2, the number is 20 out of a possible 134 = 14.93%. Please make the appropriate corrections in the table and text.

○
The literature on household transmission of SARS-CoV-2 is moving quickly, but it'd be useful to update the Discussion section and references to include more recent estimates of household secondary transmission risk for SARS-CoV-2.
Yes onward transmission if the primary case was a healthcare worker or if they practiced good hand and respiratory hygiene.
The methods used to calculate confidence intervals need to be explained more clearly. When I try to replicate them, I get similar results but not exact matches.

5.
Given the uncertainty in the point and interval estimates, a single decimal point in the percentages is probably sufficient.

6.
One other limitation needs to be mentioned: The method of calculating the secondary attack risk implicitly assumes that all secondary cases are infected by the primary case, ignoring the possibility of multiple generations of transmission within the household. 7.