Association of close-range contact patterns with SARS-CoV-2: a household transmission study

Background

Households are an important location for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission, especially during periods where travel and work was restricted to essential services. We aimed to assess the association of close-range contact patterns with SARS-CoV-2 transmission.

Methods

We deployed proximity sensors for two weeks to measure face-to-face interactions between household members after SARS-CoV-2 was identified in the household, in South Africa, 2020 - 2021. We calculated duration, frequency and average duration of close range proximity events with SARS-CoV-2 index cases. We assessed the association of contact parameters with SARS-CoV-2 transmission using mixed effects logistic regression accounting for index and household member characteristics.

Results

We included 340 individuals (88 SARS-CoV-2 index cases and 252 household members). On multivariable analysis, factors associated with SARS-CoV-2 acquisition were index cases with minimum Ct value <30 (aOR 10.2 95%CI 1.4-77.4) vs >35, contacts aged 13-17 years (aOR 7.7 95%CI 1.0-58.2) vs <5 years and female contacts (aOR 2.3 95%CI 1.1-4.8). No contact parameters were associated with acquisition (aOR 1.0 95%CI 1.0-1.0) for all three of duration, frequency and average duration.

Conclusion

We did not find an association between close-range proximity events and SARS-CoV-2 household transmission. It may be that droplet-mediated transmission during close-proximity contacts play a smaller role than airborne transmission of SARS-CoV-2 in the household, due to high contact rates in households or study limitations.

Funding

Wellcome Trust (Grant number 221003/Z/20/Z) in collaboration with the Foreign, Commonwealth and Development Office, United Kingdom.


Conclusion 55
We did not find an association between close-range proximity events and SARS-CoV-2 household 56 transmission. It may be that droplet-mediated transmission during close-proximity contacts play a 57 smaller role than airborne transmission of SARS-CoV-2 in the household, due to high contact rates in 58 households or study limitations. 59 Introduction 64 South Africa has experienced five waves of severe acute respiratory syndrome coronavirus 2 (SARS-65 CoV-2) infection, with over 4 million laboratory-confirmed cases by August 2022 1 . The true burden is 66 highly underestimated, since based on seroprevalence data, after the third wave of infection, 43% to 67 83% of the 59.5 million South African inhabitants had already been infected, varying by age and 68 setting 2,3 . 69 SARS-CoV-2 transmission is mainly via the respiratory route, with droplet-mediated transmission 70 thought to be the most important but airborne transmission also occurs 4,5 . Infection from 71 contaminated surfaces has also been described 4 . Although infection risk is highest from 72 symptomatic individuals 6 , with the most infectious period one day before symptom onset 4 , 73 asymptomatic individuals can still transmit SARS-CoV-2 7,8 . Households are a focal point for SARS-74 CoV-2 transmission 9,10 , especially during peaks of non-pharmaceutical intervention (NPI) 75 restrictions, when movement outside of the household was limited 9 . Transmission within 76 households can in turn lead to spill over to the community 11 . 77 Prior to the widespread availability of SARS-CoV-2 vaccines, most countries relied on NPIs to reduce 78 the transmission of the virus, including wearing face masks, social and physical distancing. While 79 mobility and contact survey data showed that the implementation of NPIs led to a reduction in 80 community contacts 9,12 and in turn opportunity for infection, it is still unknown what the role of 81 contact patterns are in the transmission of SARS-CoV-2 in the household. Most analysis relating 82 contact patterns and SARS-CoV-2 transmission done to date has been based on low resolution data 83 collected from contact tracing 13 , mobility data 9 and contact surveys 12 . To obtain high-resolution 84 contact data, devices broadcasting and receiving radio frequency waves can be used to measure the 85 frequency and duration of close-proximity contacts. This has been used previously to collect contact 86 data in among others, schools 14 , workplaces 15 , hospitals 16 and households 17 , which can, in turn, be 87 used for modelling disease transmission. Specifically, for SARS-CoV-2 so far, high-resolution contact 88 data were collected on cruise ships to identify areas of high contact, and to investigate the 89 usefulness of NPIs 18 . 90 Understanding the drivers of SARS-CoV-2 transmission in the household, especially contact patterns, 91 can help inform NPIs for future SARS-CoV-2 resurgences and potentially future emerging pathogens 92 with pandemic potential. We aimed to assess the association of household close-range contact 93 patterns with the transmission of SARS-CoV-2 in the household using proximity sensors deployed 94 after the identification of SARS-CoV-2 in the household. 95

96
Screening, enrolment and follow-up 97 We nested a contact study within a case-ascertained, prospective, household transmission study for 98 SARS-CoV-2, implemented in two urban communities in South Africa, Klerksdorp (North West  99 Province) and Soweto (Gauteng Province) from October 2020 through September 2021. Sample size 100 calculations were performed for the main study, but not the nested contact study. For the main 101 study we aimed to assess a significant difference in the household cumulative infection risk (HCIR) 102 between household contacts exposed to SARS-CoV-2 by a HIV-infected vs HIV-uninfected index case 103 for a 95% confidence interval and 80% power. The resulting total sample size was 440 exposed 104 household members. Detailed sample size calculations and methods for the main study have been 105 reported previously 19 . In short, symptomatic adults (aged ≥18 years, symptom onset ≤5 days prior) 106 where we were unable to classify the sample as a variant of concern due to primary testing done 121 elsewhere, low viral load or poor sequence quality. Households with multiple SARS-CoV-2 variants 122 circulating at the same time (mixed clusters) were excluded from the analysis. We also collected 123 serum at the first and final household visit for serological testing, using an in-house ELISA to detect 124 antibodies against SARS-CoV-2 spike protein 20 and nucleocapsid protein using Roche Elecsys anti-125 SARS-CoV-2 assay. Individuals were considered seropositive if they tested positive on either assay. 126 Individuals sero-positive at the start of follow-up with no rRT-PCR confirmed SARS-CoV-2 infection 127 during follow-up were excluded from the analysis as they may have been protected from infection 128 21 . 129

Contact pattern measurements 130
At the first or second visit during follow-up, we deployed wearable radio frequency (RF) proximity 131 sensors 15 for two weeks to measure close-range interactions (<1.5 meters) between household 132 members. The proximity sensors exchange low-power radio packets in the ISM (Industrial, Scientific 133 and Medical) radio band. Exchange of packets and Received Signal Strength Indicator (RSSI), suitably 134 thresholded, are used to assess proximity between the devices. A contact interval between two 135 devices is defined as a sequence of consecutive 20-second intervals within which at least one radio 136 packet was exchanged. Each sensor had a unique hardware identifier that was linked to participant 137 study identifiers. Sensors were worn in a PVC pouch either pinned to clothing on the chest, or on a 138 lanyard around the neck based on participant preference. We requested participants to wear the 139 device while at home, to store them separately from other household member sensors at night, and 140 to complete a log sheet every day for the periods the sensors were put on and taken off. During each 141 household visit during the sensor deployment period, field workers confirmed sensors were worn. A 142 deployment log was completed for each household to link the sensor identifier to the participant 143 identifier and to log the date and time sensors were deployed and collected. After sensor collection, 144 batteries were removed to prevent further package exchange between sensors. Sensors were 145 transported to the study office where each sensor was connected to a computer and data 146 downloaded. 147

Data analysis 148
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. We assumed the first individual with COVID-19 compatible symptoms in the household (individual 149 screened at clinic) was the index case. Any household member testing positive for SARS-CoV-2 150 within the two weeks from the last positive result for the index case was considered a secondary 151 SARS-CoV-2 case. Contact event data were cleaned using an automated pipeline. We excluded any 152 close-range proximity events outside of the deployment period and that occurred during a 5-minute 153 time slice that the accelerometer did not detect any movement of the sensor. Due to a technical 154 error, some sensors at the Klerksdorp site did not have a valid time stamp and needed additional 155 processing to align the time series of close-range proximity events. This was achieved by computing, 156 for each pair of tags X and Y, the temporal shift that maximizes the correlation between the time 157 series of the number of packets per unit time transmitted by X and received by Y, and the reciprocal 158 time series of the number of packets per unit time transmitted by Y and received by X (operation 159 that can be efficiently carried out working in the frequency domain via Fourier transformation). This 160 allowed us to build a temporal alignment graph between sensors and -as long as there was at least 161 one sensor with a valid timestamp in the household -to use such graph to propagate the valid 162 timestamp to all other sensors, thus recovering global temporal alignment. For the analysis, we only 163 considered close-range proximity events that occurred one day after deployment and one day 164 before collection. Where no timestamp was available, we used data collected from one to ten days 165 after deployment. 166 We assessed three contact parameters: 1) duration (median daily cumulative time in contact in 167 seconds), 2) frequency (median daily number of contacts with the index/infected individuals over 168 the deployment period) and 3) average duration (cumulative time in contact divided by the 169 cumulative number of close-range proximity events over full deployment period). Median values 170 were preferred over mean values due to the rightly skewed data, and the different number of days 171 with measured contact data for each household after data cleaning. We assessed contact 172 parameters in two ways: 1) median number of close-range proximity events with the presumptive 173 index case and 2) median number of close-range proximity events with all SARS-CoV-2 infected 174 household members (as confirmed by rRT-PCR). The latter assessment was to take into account that 175 the transmission could have been from any of the infected household members, and not necessarily 176 the index case, or that the index case was misclassified. 177 We constructed contact matrices by combining the median duration and frequency of close-range 178 proximity events for all participants between each age group, respectively. To normalize the matrix 179 based on number of participants, we divided the cumulative contact duration and frequency by the 180 total number of individuals in the two age groups being investigated in each cell. 181 We assessed the association of contact parameters with SARS-CoV-2 household transmission using 182 the Wilcoxon rank-sum test (considering p<0.05 as significant) and through logistic regression 183 controlling for individual characteristics associated with transmission. To assess factors associated 184 with SARS-CoV-2 household transmission, we performed logistic regression with a mixed effects 185 hierarchical regression model to account for household-and site-level clustering. For the analysis 186 with a defined index case (i.e. investigating close-range proximity events with all presumptive index 187 cases, first person with COVID-19 symptoms), we included only household contacts with their SARS-188 CoV-2 infection status as the outcome, assessing both index (transmission) and contact (acquisition) 189 characteristics. For the analysis with no defined index case (i.e. investigating close-range proximity 190 events with all SARS-CoV-2 infected household members), we included all enrolled household 191 members (originally considered presumptive index and household contacts), assessing only their 192 own characteristics. For the analysis with close-range proximity events with all SARS-CoV-2 infected 193 household members, we included an offset term in the model to account for the number of SARS-194 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The overall median duration of daily close-range proximity events was 1,095 seconds (18 minutes, 223 IQR 398-2,705 seconds), with a 39 second (IQR 32-49 seconds) average duration per contact event, 224 and a median of 26 (IQR 10-58) close-range proximity events per day amongst household members 225 (Figure 1, Supplementary Table 2). The highest median daily contact duration was observed between 226 individuals within the <5 year, 5-12 year and 35-59 year groups (Figure 2 A, D). Similar patterns were 227 also seen for median daily close-range proximity duration and frequency in children aged 5-12 and 228 13-17 years (Figure 2 B-F. 229 We did not find any association between any of the contact parameters (either with the index case 230 or all SARS-CoV-2 infected household members) and SARS-CoV-2 infection in the household using 231 the Wilcoxon rank-sum test (p-values ranging 0.2-0.9, Table 1). 232 When assessing factors associated with SARS-CoV-2 transmission from presumptive index cases and 233 acquisition in household members, none of the contact parameters were associated with SARS-CoV-234 2 transmission on univariate analysis. Sleeping in the same room as the index case was also not 235 associated with transmission (OR 0.94, 95%CI 0.47 to 1.88). On multivariable analysis after 236 controlling for index age and SARS-CoV-2 infecting variant, factors significantly associated with 237 higher SARS-CoV-2 transmission and acquisition was index case minimum C t value <30 (aOR 10.2 238 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint duration (aOR 1.0 95%CI 1.0-1.0) with SARS-CoV-2 infected household members was associated with 250 acquisition (Table 3). 251

252
In this case-ascertained, prospective household transmission study we did not find an association 253 between the duration and frequency of close-range proximity events with SARS-CoV-2 infected 254 household members with transmission in the household. Singapore, it was found that sharing a bedroom with an index case and speaking to the index case 263 for more 30 minutes or longer increased the risk for infection 26 . We did not see similar results when 264 assessing sharing a bedroom with the index case, and this may be due to the already high level of 265 crowding in included households. Although we observed an increase in infection risk with higher 266 average contact durations with the index case on univariate analysis, this association was no longer 267 seen when adjusting for age and other index and contact factors associated with 268 transmission/acquisition. 269 There are several possible reasons why we did not observe an association with close-range proximity 270 events and SARS-CoV-2 transmission, these can be classified as related to transmission dynamics or 271 study limitations. One possibility is that along with droplet-mediated transmission during close-272 proximity contacts, airborne 4,5 , and to a lesser extent, fomite-mediated transmission 4 may also play 273 a role in the transmission of SARS-CoV-2 in the household. More evidence is becoming available 274 showing that aerosol transmission may be a more important transmission route for SARS-CoV-2 than 275 initially anticipated, especially so in poorly ventilated indoor environments 5,27 . Households in these 276 communities do not have central air-conditioning or heating 28 , and during the winter months 277 ventilation may by poorer than in summer, although we did not measure this. Furthermore, sensors 278 only measure face-to-face interactions, and if individuals were close to each other but not directly 279 facing one another for extended durations, we would not have measured this, although sharing of 280 the same air may have occurred. The ventilation within households should be considered in future 281 studies, as this can be a target for intervention strategies to reduce secondary transmission. The high 282 level of interaction in relatively crowded South African households may already be above the 283 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. threshold for transmission risk, with host-characteristics like index viral load and contact age being 284 more important to determine infection risk in this context. It is of interest that close-range proximity 285 patterns within the household did not fully account for the differences in transmission based on age; 286 with teenagers and adults experiencing the highest infection risk, but children aged 5-17 years 287 having the highest contacts. 288 Our study had limitations both in design and execution. Due to the nature of the case-ascertained 289 study design, we would have missed the period when the index case was most infectious, just before 290 symptom onset 4 , and the close-range contact patterns measured during the study may have been 291 different after the household members were aware of the index SARS-CoV-2 case (leading to 292 reduced contact), and again once secondary cases were informed of their infection status (leading to 293 increased contact). We also did not collect any information on possible NPI usage in the households, 294 like wearing of masks. A study from South Africa showed that individuals staying at home were less 295 likely to wear a mask 29 , but these data were not ascertained during a time when a household 296 member was infected with SARS-CoV-2. We also did not consider where contacts took place (indoors 297 or outdoors), which relates to ventilation and may have influenced transmission. We may have also 298 misclassified the true index case if they were asymptomatic, and did not consider tertiary 299 transmission chains in the index-directed analysis. To adjust for possible misclassification, we 300 performed a grouped assessment investigating close-range proximity events with all SARS-CoV-2 301 infected household members. This grouped analysis may also have diluted possible associations with 302 the true infector. We also did not consider multiple introductions within the household, although we 303 did exclude households with more than one SARS-CoV-2 variant detected. During the peaks of waves 304 of infection in South Africa, one variant was responsible for the majority of the infections 30 , and the 305 additional introductions within the household were likely to have been the same as the initial 306 variant. Higher resolution sequencing data may be useful to more accurately identify chains of 307 transmission within the household. Combining contact data with clinical and 308 virological/bacteriological data have been shown to be useful to reconstruct transmission networks 309 31 , and we will consider this for future analyses. Our measurement of close-range contact patterns 310 was also limited by compliance, as during the cleaning process we identified 73 sensors that were 311 not worn, based on accelerometer data. We also had limited data in some households, where some 312 individuals did not consent to the contact aspect for the study, or where we were unable to retrieve 313 data due to hardware failure, lost or damaged tags. 314 In conclusion, we did not observe an association with close-proximity contacts and SARS-CoV-2 315 transmission in the household. A case-ascertained, prospective household transmission study may 316 not be well suited to investigate this question. A possible other study design to consider is randomly 317 selected prospective household cohorts, but deployment of sensors for extended periods of time 318 may be logistically challenging and lead to participant fatigue and households in a cohort may not 319 experience infection episodes unless the community attack rate is very high. High-resolution 320 contacts in other settings like schools or workplaces where contacts are less frequent could be 321 useful to identify the type of contact events that may lead to SARS-CoV-2 transmission. It may be 322 that aerosol transmission plays a more important role than droplet-mediated transmission, which 323 would make ventilation within households can also be an important consideration for future studies, 324 and increased ventilation can be a method to reduce secondary transmission in households. 325 Nevertheless, our study provides high-resolution household contact data that can be used to 326 parametrise future transmission models. 327 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Communicable Diseases of the National Health Laboratory Service, Johannesburg, South Africa.) 370 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint    October 2021. Teal denotes lowest value, purple highest, white no data for age group combination 397 . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted December 22, 2022. ; https://doi.org/10.1101/2022.12.22.22283843 doi: medRxiv preprint