Quantifying population contact patterns in the United States during the COVID-19 pandemic

The COVID-19 pandemic underscored the critical role of interpersonal interaction in disease transmission. To curb the spread, many countries implemented unprecedented physical distancing measures. In the US, these measures varied across regions, creating a complex landscape for evaluating their effectiveness. This study aimed to quantify changes in interpersonal contact patterns in the US during the pandemic's early stages. Understanding these patterns is crucial for informing epidemiological models, predicting transmission dynamics, and identifying populations at greatest risk. The Berkeley Interpersonal Contact Survey (BICS) was developed to collect detailed data on the number and characteristics of contacts individuals had, including the age, sex, relationship, and location of interactions. This granular information is valuable for precisely modeling transmission dynamics and targeting interventions to high-risk groups. Age-structured contact rates are particularly relevant due to age-related variations in COVID-19 outcomes. The study sought to describe changes in contact rates and patterns across the pandemic's evolution, identify correlates of close interpersonal contact, and evaluate physical distancing policy effectiveness by estimating the impact on the reproduction number (R0).

Prior research highlighted the importance of social contact patterns in the spread of infectious diseases, particularly respiratory illnesses. Several studies have utilized social contact surveys to parameterize models and estimate transmission parameters for various pathogens. However, existing data on pre-pandemic contact patterns in the US were limited. The study leveraged existing data from a 2015 Facebook survey and the UK POLYMOD study as pre-pandemic baselines for comparison. These baselines provided context for interpreting the reductions in contact observed during the pandemic. While acknowledging the limitations of using non-probability samples, the researchers justified the choice by noting the cost and logistical constraints of conducting probability sampling during a pandemic.

The BICS study comprised four waves of data collection: Wave 0 (March 22–April 8, 2020), Wave 1 (April 10–May 4, 2020), Wave 2 (June 17–23, 2020), and Wave 3 (September 11–26, 2020). A total of 9743 respondents participated. The survey employed a quota sampling approach using online panels, aiming for representation across various demographic groups and including oversampled data from specific cities (New York, San Francisco Bay Area, Atlanta, Phoenix, Boston, and Philadelphia in Wave 1). To account for the quota sampling, the researchers employed calibration weighting, adjusting for variables such as age, sex, race, ethnicity, education, household size, and location (rural/suburban/urban). Detailed information was requested about up to three contacts per respondent, which included the contact's age, sex, relationship, and the location of the interaction. Contact-level weights were developed to handle situations where respondents reported more than three contacts but provided details for only three. Statistical analysis was performed using negative binomial regression models, accounting for overdispersion in the count data and handling censored observations (those reporting "10 or more" contacts). A Bayesian approach was used for model fitting. The primary outcome was the number of daily contacts and non-household contacts. To study the effect of physical distancing on transmission, age-structured contact matrices were estimated for each wave. The dominant eigenvalue of the contact matrix was used to estimate the reproduction number R0 relative to pre-pandemic levels, using both Facebook 2015 data and the UK POLYMOD study as baselines. The analysis also considered mask usage data (available in Waves 1-3) to explore its effect on R0 estimation. The study utilized the R packages *autumn* and *leaf-peepr* for weighting and statistical analysis, and Stan for Bayesian model fitting.

The study revealed a dramatic reduction in interpersonal contact during the first wave (Wave 0), with an 82% decrease in average daily contacts compared to pre-pandemic levels (using the 2015 Facebook study as a baseline). Subsequent waves demonstrated gradual increases in contact rates: 74% reduction in Wave 1, 68% in Wave 2, and 60% in Wave 3. These changes translated to substantial declines in the estimated R0: 73% in Wave 0, 57% in Wave 1, 48% in Wave 2, and 36% in Wave 3. The R0 estimates were above 1 in Waves 1-3, indicating the potential for continued transmission even with reduced contact rates. Analyses using the UK POLYMOD study as a baseline yielded qualitatively similar results. The increase in contact rates was associated with increases in work contacts, and contacts at stores and businesses, suggesting that the reopening of the economy might have played a major role. Demographic analysis revealed that younger individuals (under 45) and males consistently reported higher contact rates compared to older individuals and females. Contact patterns varied across cities and over time, with some cities exhibiting steadier increases in contact than others. The incorporation of mask usage data suggested that widespread mask adoption could further reduce the estimated R0, but the study lacked Wave 0 data on mask usage. The age-structured contact matrices showed a persistent pattern of assortative mixing by age, even during periods of reduced overall contact.

The findings demonstrate the substantial impact of early physical distancing measures in curbing the spread of COVID-19 in the US, reflected in the significant reduction in R0 during the initial phase of the pandemic. The subsequent increase in contact rates and R0 values highlights the challenges in maintaining effective control of the virus in the absence of widespread adherence to distancing measures. The observed demographic differences in contact rates suggest that targeted interventions could be more effective in reducing transmission within those groups. The study's findings on the impact of work-related and business-related contacts should be considered for the development of policies aimed at reducing transmission as the economy reopens. The findings contribute to a growing body of research on the impact of non-pharmaceutical interventions (NPIs) on COVID-19 transmission, with particular implications for modeling and policy-making.

The BICS study provides valuable insights into changing population contact patterns during the initial phases of the COVID-19 pandemic in the United States. The findings clearly show the effectiveness of early physical distancing measures in reducing transmission but also the need for ongoing monitoring of contact patterns and adaptability of intervention strategies as the pandemic evolves. Future research could investigate the effectiveness of other non-pharmaceutical interventions, including mask use and vaccination, in reducing transmission dynamics further. The ongoing BICS study, with continued data collection, will provide crucial data on the effectiveness of future interventions.

The study acknowledges several limitations, including the use of a quota sample from an online panel, potential recall and social desirability biases in survey responses, the exclusion of non-English speakers and individuals under 18, and the inability to account for age-specific differences in susceptibility or infectiousness. These factors could influence the generalizability and accuracy of the findings.