Assessing the Association Between Social Gatherings and COVID-19 Risk Using Birthdays | Public Health | JAMA Internal Medicine | JAMA Network
[Skip to Navigation]
Figure 1.  Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household
Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household

Absolute adjusted difference in COVID-19 rates per 10 000 individuals between households in which at least 1 member had a birthday compared with households in which no member had a birthday in a given week between January 1 and November 8, 2020. Estimates were obtained from a multivariable linear regression of COVID-19 diagnosis within a household (binary variable) as a function of having a birthday in the household in the 0- to-2-week period prior (binary variable) and covariates including the number of people in each household, an indicator for any children in the household, and fixed-effect controls for county and week, which adjusted for differences across geographies and general trends in the COVID-19 pandemic, respectively. Households in counties in the first and second decile of COVID-19 prevalence were the reference. Error bars indicate 95% CIs. See eMethods in the Supplement for additional details.

Figure 2.  Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household, Child’s vs Adult’s Birthdays
Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household, Child’s vs Adult’s Birthdays

Absolute adjusted difference in COVID-19 rates per 10 000 individuals between households in which at least 1 child vs adult member had a birthday in a given week compared with households in which no member had a birthday, between January 1 and November 8, 2020. Estimates were obtained from a multivariable linear regression of COVID-19 diagnosis within a household (binary variable) as a function of having a birthday in the household in the 0- to-2-week period prior (binary variable), interacted with whether the birthday was for a child or an adult, allowing for a formal test of interactions. Covariates included the number of people in each household, an indicator for any children in the household, and fixed-effect controls for county and week, which adjusted for differences across geographies and general trends in the COVID-19 pandemic, respectively. Households in counties in the first and second decile of COVID-19 prevalence were the reference. Error bars indicate 95% CIs. See eMethods in the Supplement for additional details.

Figure 3.  Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household, by Presence of Precipitation During Week of Birthday, Milestone Birthdays, Republican vs Democrat Voter Share in 2016 Presidential Election, and Presence of County Shelter-in-Place Orders
Differences in Rates of COVID-19 According to Presence of a Birthday in an Individual’s Household, by Presence of Precipitation During Week of Birthday, Milestone Birthdays, Republican vs Democrat Voter Share in 2016 Presidential Election, and Presence of County Shelter-in-Place Orders

Absolute adjusted difference in COVID-19 rates per 10 000 individuals between households in which at least 1 member had a birthday compared with households in which no member had a birthday, between January 1 and November 8, 2020, and by counties with precipitation on the Saturday of that week (A), households with milestone birthdays (16th, 18th, 20th, 30th, 40th, 50th, and 60th) vs non-milestone birthdays (B), counties that voted for Hillary Clinton vs Donald Trump in the 2016 presidential election (C), and counties with a shelter-in-place policy in effect (D). Estimates were obtained from a multivariable linear regression of COVID-19 diagnosis within a household (binary variable) as a function of having a birthday in the household in the 0- to-2-week period prior (binary variable) interacted with each of the above variables in separate regressions for each subgroup. Covariates included the number of people in each household, an indicator for any children in the household, and fixed-effect controls for county and week, which adjusted for differences across geographies and general trends in the COVID-19 pandemic, respectively. Households in counties in the first and second decile of COVID-19 prevalence were the reference. Error bars indicate 95% CIs. See eMethods in the Supplement for additional details.

Table.  Comparison of Individual and Household Characteristics
Comparison of Individual and Household Characteristics
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Views 76,394
    Citations 0
    Original Investigation
    June 21, 2021

    Assessing the Association Between Social Gatherings and COVID-19 Risk Using Birthdays

    Author Affiliations
    • 1RAND Corporation, Santa Monica, California
    • 2Castlight Health, San Francisco, California
    • 3Department of Health Care Policy, Harvard Medical School, Boston, Massachusetts
    • 4Department of Medicine, Massachusetts General Hospital, Boston
    • 5National Bureau of Economic Research, Cambridge, Massachusetts
    JAMA Intern Med. 2021;181(8):1090-1099. doi:10.1001/jamainternmed.2021.2915
    Key Points

    Question  Is there an association between household birthdays, which likely correspond to informal social gatherings, and COVID-19 infection?

    Findings  This cross-sectional study used administrative health care data on 2.9 million households from the first 45 weeks of 2020 and found that, among households in the top decile of county COVID-19 prevalence, those with birthdays had 8.6 more diagnoses per 10 000 individuals compared with households without a birthday, a relative increase of 31% of county-level prevalence, an increase in COVID-19 diagnoses of 15.8 per 10 000 persons after a child birthday, and an increase in COVID-19 diagnoses of 5.8 per 10 000 among households with an adult birthday.

    Meaning  This study suggests that events that lead to small and informal social gatherings, such as birthdays, and in particular, children’s birthdays, are a potentially important source in SARS-CoV-2 transmission.

    Abstract

    Importance  Many policies designed to stop the spread of COVID-19 address formal gatherings, such as workplaces and dining locations. Informal social gatherings are a potentially important mode of SARS-CoV-2 transmission, but studying their role in transmission is challenged by data and methodological limitations; birthdays offer an opportunity to empirically quantify the potential role of small social gatherings in COVID-19 spread.

    Objective  To assess the association between social gatherings and SARS-CoV-2 transmission by studying whether COVID-19 rates increase after birthdays in a household.

    Design, Setting, and Participants  This cross-sectional study used nationwide data from January 1 to November 8, 2020, from 2.9 million US households with private insurance to compare COVID-19 infections between households with and without a birthday in the preceding 2 weeks, stratified according to county-level COVID-19 prevalence in that week and adjusting for household size and both week- and county-specific differences. The study also compared how birthday-associated infection rates differed by type of birthday (eg, child vs adult birthday, or a milestone birthday such as a 50th birthday), county-level precipitation on the Saturday of each week (which could move gatherings indoors), political leanings in the county, and state shelter-in-place policies.

    Main Outcomes and Measures  Household-level COVID-19 infection.

    Results  Among the 2.9 million households in the study, in the top decile of counties in COVID-19 prevalence, households with a birthday in the 2 weeks prior had 8.6 more diagnoses per 10 000 individuals (95% CI, 6.6-10.7 per 10 000 individuals) compared with households without a birthday in the 2 weeks prior, a relative increase of 31% above the county-level prevalence of 27.8 cases per 10 000 individuals, vs 0.9 more diagnoses per 10 000 individuals (95% CI, 0.6-1.3 per 10 000 individuals) in the fifth decile (P < .001 for interaction). Households in the tenth decile of COVID-19 prevalence had an increase in COVID-19 diagnoses of 15.8 per 10 000 persons (95% CI, 11.7-19.9 per 10 000 persons) after a child birthday, compared with an increase of 5.8 per 10 000 persons (95% CI, 3.7-7.9 per 10 000 persons) among households with an adult birthday (P < .001 in a test of interactions). No differences were found by milestone birthdays, county political leaning, precipitation, or shelter-in-place policies.

    Conclusions and Relevance  This cross-sectional study suggests that birthdays, which likely correspond with social gatherings and celebrations, were associated with increased rates of diagnosed COVID-19 infection within households in counties with high COVID-19 prevalence.

    Introduction

    In addition to large gatherings of people, small social gatherings are thought to be an important source of SARS-CoV-2 transmission.1,2 Although several evaluations of state policies that limit social gatherings—limitations on gathering size, curfews, or shelter-in-place orders—have found that these policies are associated with reduced growth in COVID-19 cases,3-7 it is empirically challenging to disentangle the potential consequences of any specific policy associated with COVID-19 cases given the many confounding factors at local, regional, and state levels. Although it is well known that SARS-CoV-2 spreads primarily through person-to-person contact, estimating the risk entailed with small gatherings—and, therefore, the risk reduction associated with not gathering—is also empirically challenging given that large-scale data are required on when individuals gather and whether those who gather are more likely to develop COVID-19. Even with such data, those who are more likely to gather may be at higher risk for COVID-19 for unobserved but associated reasons, a problem of confounding.

    We analyzed the potential increased risk of COVID-19 around small social gatherings by studying changes in COVID-19 rates after important life events, specifically birthdays. To the extent that birthdays provide an important reason for people to gather; can be identified and linked to COVID-19 diagnoses in large, administrative health care data; and should not be associated with COVID-19 risk because cases are randomly distributed across households—thereby addressing the problem of confounding in who does and does not socially gather—birthdays occurring during the ongoing COVID-19 pandemic offer an opportunity to empirically quantify the potential role of small social gatherings in COVID-19 spread.

    Methods
    Overview of Approach

    We quantified the likelihood of COVID-19 infection associated with small social gatherings by studying changes in household COVID-19 infection rates after the presence of a birthday in a household. We hypothesized that birthdays—which may be identifiable for each household member in insurance claims data and can be linked to administrative diagnoses of COVID-19 in these data—should otherwise not be associated with COVID-19 risk because they are randomly distributed across households. Studying changes in household COVID-19 infection rates after a specific, plausibly exogenous event such as birthdays therefore allows for an assessment of the increase in a household’s COVID-19 risk after a social gathering. To the extent that the presence of a birthday in a household, on average, increases the number of social gatherings in that week compared with households in which no member had a birthday that week, a comparison of COVID-19 infections between these 2 groups in the weeks after the birthday would be informative as to the increased risk of COVID-19 associated with greater social gathering. This study was approved by the RAND institutional review board with a waiver of informed consent because the data were collected for nonresearch purposes and are deidentified. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

    Data Source

    We identified household birthdays and household COVID-19 infections between January 1 and November 8, 2020, using a large commercial insurance database from Castlight Health,8-10 which provides health navigation services for approximately 200 US employers that provide employer-sponsored insurance. The employers range in size—from a few thousand to more than 500 000 employees—and industry (eg, education, communications, retail, and government). The data contain linkages for all household members enrolled on a single insurance plan, which allows for analyses that link members to households. We examined the representativeness of our study population for the broader commercially insured population nationally by comparing the study population with the population in the American Community Survey, administered by the US Census Bureau (eTable 1 in the Supplement).

    Outcomes

    For each household, we identified weekly COVID-19 diagnoses at the household-level using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes U07.1, B34.2, and B97.29. The sensitivity and specificity of administrative diagnosis codes for COVID-19 have recently been reported to be 98.0% and 99.0%, respectively.11 As a result of using administrative diagnoses of COVID-19, we potentially missed asymptomatic cases or infections that did not require interactions with health care professionals. Thus, our analysis should be interpreted as examining the association between household birthdays and moderate-to-severe COVID-19 infections.

    Statistical Analysis

    Within each household, we first identified weeks in which any member of the household had a birthday, including children and adults. To account for both differential timing of birthday-related social gatherings (ie, some gatherings might occur several days before or after the actual birthday) and the SARS-CoV-2 incubation period,4 we defined the period in which birthday-related COVID-19 cases could emerge as the 2-week period after the birthday. As described below, we conducted a falsification analysis in which COVID-19 infections were assessed in the 4 and 8 weeks prior to the birthday as well, hypothesizing that no association should exist between the presence of a birthday in a household and COVID-19 cases in the 4 or 8 weeks prior.

    The likelihood of COVID-19 infection in a given household would be expected to be associated with both social distancing behavior of household members and COVID-19 prevalence among social groups in which any gathering occurs. Birthday-related social gatherings that occurred in January and February 2020, prior to the declaration of the COVID-19 national emergency, or in areas with less COVID-19 prevalence would be less likely to be associated with COVID-19 infection. Put differently, the absolute magnitude of any association between birthdays and COVID-19 risk would be expected to be larger in counties and time periods with greater COVID-19 prevalence. To allow for the association between having a birthday and COVID-19 infection to be nonlinearly modified by COVID-19 prevalence in a given county in a given week, we calculated deciles of COVID-19 infection using data from USAFacts.12 Specifically, we calculated COVID-19 prevalence for each county-week in 2020 and assigned county-weeks into deciles based on prevalence; therefore, a given county early in the pandemic could be in the bottom decile of COVID-19 prevalence for county-weeks and in a higher decile later in the pandemic, when reported cases in that county were greater. As a sensitivity analysis, we repeated our analysis using quintiles of COVID-19 prevalence instead of deciles.

    We assessed the association between household birthdays and COVID-19 infection rates by estimating a multivariable linear probability model (specified below) of COVID-19 diagnosis in a household in a given week as a function of a birthday occurring in the household in the 2 weeks prior (indicator variable), county and week fixed effects (ie, indicator variables for each county and each of 52 calendar weeks), and family-level covariates. We interacted the household birthday indicator with indicators for each decile of individuals per 10 000 with a COVID-19 diagnosis in each county, to allow for the association between having a birthday and household COVID-19 infection to be nonlinearly modified by COVID-19 prevalence. We estimated multivariable linear probability models rather than logistic models owing to the large number of fixed effects in our regression (2454 fixed effects for counties and 42 fixed effects for weeks in our model) and failure of the logistic model to converge in this setting.

    For each household i and week t, we measured the binary outcome of a COVID-19 infection as:

    Image description not available..

    In this model, the coefficients measure the association between COVID-19 diagnosis and the joint interaction of having a birthday in a household in a specific decile of COVID-19 prevalence (the first 2 deciles were combined and served as the reference category owing to slow growth of COVID-19 in some areas). We adjusted for the number of members in the household, to account for increasing rates of spread among larger households, and an indicator for whether the household had any children, to account for potential COVID-19 transmission from children. We also included fixed effects for county and week, which adjusted for time-invariant market characteristics (eg, income, education status, and population size and density) and general trends in the COVID-19 pandemic, respectively. Specifically, week fixed effects adjusted nonlinearly for the general trajectory of the COVID-19 pandemic and county fixed effects adjusted for unobserved, time-invariant differences across geographical regions. Owing to these fixed effect controls, the δ coefficients of interest measured the difference in household COVID-19 infection rates among households in the same county and in the same week, but between households with and without a birthday.

    Subgroup Analysis

    We assessed the association between birthdays and COVID-19 risk in households with a child’s vs an adult’s birthday, because the likelihood and nature of gathering (eg, degree of physical distancing and masking) may depend on whether a child or adult had a birthday; in households with a milestone birthday (eg, turning 40, 50, or 60 years), which could make gatherings more likely; when heavy precipitation occurred during the Saturday of the birthday week (the day of the week we assumed that most birthday-related gatherings might occur),13 which could move gatherings indoors (in a sensitivity analysis we also categorized rainfall at the weekly level and identified weeks as having heavy precipitation if precipitation in a given week was above the national weekly mean in 2020); in areas with a large share of voters for President Trump in the 2016 election,14 to assess the association of differences in social distancing and masking preferences and policies with outcomes; and in counties with shelter-in-place orders during the birthday week, which may limit interactions associated with COVID-19 transmission and therefore lower the magnitude of association between birthdays and subsequent COVID-19 risk within households.15 Additional details are in the eMethods in the Supplement.

    Sensitivity Analysis

    The key econometric identification assumption with our approach was that household birthdays occurred quasi-randomly (ie, a researcher did not randomize households; instead, randomization occurred by nature) and, aside from any association of birthdays with gatherings and subsequent SARS-CoV-2 transmission, would otherwise not be associated with factors in COVID-19 risk. If birthdays were more likely to occur in areas with increased COVID-19 prevalence or in later stages of the COVID-19 pandemic, then our results could spuriously reflect differential birthday geographical location and timing, rather than increased risks that arise from this potential source of social gatherings.

    To address this concern, we conducted 3 analyses. First, we compared individual and household characteristics of individuals with birthdays in each calendar quarter. Based on the individual’s quarter of birth, we compared individual age, sex, birthday day of week, US Census region (Northeast, Midwest, South, and West), household size, the share of households with any children, and the industry of the household’s primary insurance subscriber.

    Second, we tested for differences in COVID-19 infection rates in the 4 and 8 weeks before a household birthday. Although some birthdays may be celebrated 1 to 2 weeks before or after the actual birth date, we assumed that most celebrations would be likely to occur close to the birthday week. Finding similar changes in COVID-19 diagnosis rates 4 or 8 weeks before a birthday (compared with the 2 weeks after a birthday) would suggest that our results are spurious and likely not owing to birthday-related social gatherings. We conducted this analysis by estimating multivariable linear regressions similar to our main regression but interacted weekly COVID-19 cases per 10 000 decile indicators with an indicator for whether or not a household had a future birthday in the following 4th week (or 8th week). To account for potential biases occurring from increased testing surrounding birthday-related social gatherings, which could lead to detection of asymptomatic cases, we also used COVID-19–related hospitalizations in the 4 weeks after a household birthday as a dependent variable (eMethods in the Supplement).

    Third, as a falsification test,16 we assigned random birthdays to each household. Finding an association between COVID-19 infection rates and randomly assigned birthdays would also suggest that the primary results were spurious and not owing to birthdays. We interacted weekly COVID-19 cases per 10 000 deciles with indicators for randomly assigned birthdays. Stata, version 16 (StataCorp LLC) was used for all statistical analyses. All P values were from 2-sided tests and results were deemed statistically significant at P < .05.

    Results
    Study Population

    The study population included 6 535 987 individuals in 2 926 530 households followed up from January 1 to November 8, 2020. The study population was similar in sex, age, and geographic distribution to the broader commercially insured population nationally (eTable 1 in the Supplement). The share of households with a birthday in a given week was uniform across weeks of the year (eFigure 1 in the Supplement) and the distribution of individual and household characteristics did not vary according to the quarter of the year in which an individual’s birthday fell, both consistent with the hypothesis that the timing of birthdays across the year was plausibly not associated with individual or household risk factors for developing COVID-19. For example, across each birth quarter, the age, sex, family size and composition, week day of birthday, geographical region (US Census region), and industry of the primary insurance subscriber were similar (Table).

    Association Between Household Birthdays and COVID-19 Infection

    The presence of a birthday within a household was associated with greater rates of COVID-19 diagnosis, with the absolute adjusted rate difference between households with and without a birthday increasing with greater COVID-19 county prevalence (Figure 1). For example, in the top decile of counties in COVID-19 prevalence, households with a birthday in the 2 weeks prior had 8.6 more diagnoses per 10 000 individuals (95% CI, 6.6-10.7 per 10 000 individuals) compared with households without a birthday in the 2 weeks prior, vs 0.9 more diagnoses per 10 000 individuals (95% CI, 0.6-1.3 per 10 000 individuals) in the fifth decile of counties in COVID-19 prevalence (P < .001 for interaction). Relative to the baseline rate of COVID-19 infections among households in the top decile of COVID-19 prevalence (27.8 per 10 000 persons), the estimated increase of 8.6 more COVID-19 diagnoses per 10 000 individuals corresponded to a 31% relative increase in the likelihood of household COVID-19 infections in households with a birthday in the 2 weeks prior. We observed similar results when using quintiles of COVID-19 prevalence (eFigure 2 in the Supplement) and in models that included household fixed effects and clustering of SEs at the household level (eFigure 3 in the Supplement). No differences were found when using randomly assigned birthdays (eFigure 4 in the Supplement) or when examining diagnoses in weeks prior to birthdays (eFigure 5 in the Supplement), suggesting that these findings were not spurious. Similar patterns were observed when using COVID-19–related hospitalizations as the outcome of interest, suggesting that increased COVID-19 infection rates were not owing to increased testing (eFigure 6 in the Supplement).

    Subgroup Analyses

    The regression coefficient of the association between household birthday and COVID-19 diagnosis was larger in absolute magnitude in instances where the birthday in a household was for a child vs an adult (Figure 2). For example, among counties in the fifth decile of COVID-19 prevalence, households with a child’s birthday had an increase in the likelihood of COVID-19 diagnosis of 1.0 per 10 000 persons (95% CI, 0.3-1.7 per 10 000 persons), whereas households in the tenth decile of COVID-19 prevalence had an increase in COVID-19 diagnoses of 15.8 per 10 000 persons (95% CI, 11.7-19.9 per 10 000 persons) (P < .001 in a test of interactions). In contrast, the increase in COVID-19 diagnoses associated with an adult’s birthday in a household was lower in magnitude than the increase in COVID-19 diagnoses associated with children’s birthdays. For example, among households with an adult’s birthday, rates of COVID-19 diagnosis among households in the tenth decile of COVID-19 prevalence increased by 5.8 per 10 000 persons (95% CI, 3.7-7.9 per 10 000 persons) compared with 15.8 per 10 000 persons (95% CI, 11.7-19.9 per 10 000 persons) for households with a child’s birthday (P < .001 in a test of interactions). For each decile of COVID-19 prevalence, the change in household COVID-19 infection rates was larger in magnitude for households with a child’s birthday than for households with an adult birthday (in a test of interactions, the increase in COVID-19 diagnosis rate among households with a child’s birthday was statistically greater than the change observed for households with an adult’s birthday for the eighth and greater deciles of COVID-19 prevalence).

    The association between birthdays and household COVID-19 diagnosis did not vary according to whether a household had a milestone birthday, majority-Trump vs majority-Clinton voter share in the household’s county, precipitation in the county during the week of the birthday, or the presence of county-level shelter-in-place policies during the week of the birthday (Figure 3; eTable 2 in the Supplement).

    Discussion

    Although SARS-CoV-2 is known to spread primarily through person-to-person contact, estimating the risk entailed with small gatherings, and therefore the potential individual outcomes of policies that limit such gatherings, is challenging because of data and methodologic limitations. Not only must large-scale data be linked between individuals’ social gathering practices and subsequent COVID-19 diagnoses, but an empirical approach is needed that can address the confounding that may occur if those who are more likely to gather at any given time are at higher risk for COVID-19 owing to other unobserved behaviors.

    We relied on the timing of birthdays being plausibly not associated with COVID-19 risk to assess whether COVID-19 rates were higher in households with a specific reason to gather, rather than those observed to gather, which is similar to an intent-to-treat analysis. Using administrative health care data on 2.9 million households, with information on birthdays and COVID-19 diagnoses for each household member, we found that the presence of a birthday within a household was associated with significantly greater COVID-19 diagnosis rates in those households in the 2 weeks after birthdays. The absolute rate difference in COVID-19 diagnosis between households with and without a birthday increased with the prevalence of COVID-19 in a household’s county, consistent with the likelihood of SARS-CoV-2 transmission at any given gathering being greater when community transmission is higher. The estimated increase in COVID-19 infections in households with birthdays was larger in magnitude in households that had a recent child’s (vs adult’s) birthday, suggesting an increased likelihood of gathering around children’s’ birthdays, a larger number of participants, relaxed masking and social distancing behavior, or a combination of these. Households may also be more likely to have a social gathering for a child’s birthday than for an adult’s birthday.

    The increase in COVID-19 infections in households with birthdays was not greater for milestone birthdays, suggesting that these birthdays did not offer a special reason for people to gather or change the way they gather compared with non-milestone birthdays. The increase in COVID-19 infections in households with birthdays was also not greater in counties where President Trump had a larger voter share than presidential candidate Hillary Clinton in the 2016 national election, suggesting that individuals’ decisions on whether and how to gather for birthdays were similar between these areas, despite differences in state policies and political views around social distancing and masking.17 The increase in COVID-19 infections in households with birthdays was no greater during birthday weeks with precipitation, suggesting that the decision to gather was not associated with this specific element of weather. Finally, there was not a decrease in COVID-19 infections in households with birthdays in counties with an active shelter-in-place order, suggesting that compliance with these policies for these particular events may be low. Other studies have used cellular telephone–based measures of mobility and found that physical mobility responses to shelter-in-place policies are small and dissipate quickly.18-20 It is also possible that shelter-in-place orders, which are variable in their reach, may focus on formal gathering places such as restaurants, stores, gyms, and other venues. Our findings suggest that policy interventions designed to limit disease transmission should also focus on informal gatherings as well.

    Our findings are informative as to the increased risk of COVID-19 to individuals associated with greater social gathering. Among households in the top decile of COVID-19 prevalence, for example, the increase in COVID-19 diagnoses associated with birthdays corresponded to a 31% relative increase for those households. However, this meaningful increase in risk was unlikely to be evenly distributed across households, because many households may have complied with the Centers for Disease Control and Prevention’s suggested guidelines on social distancing behavior, even for birthday celebrations.1 Thus, the 31% relative increase is likely to be disproportionately borne by a smaller number of households, implying that the increased relative risk associated with a single additional gathering could be considerably larger.

    Limitations

    Our study had several limitations. First, the study was observational and, despite relying on the quasi-random timing of birthdays to study the association between small social gatherings and COVID-19 risk, residual confounding is possible. More important, no associations between birthdays and household COVID-19 diagnosis were found in 2 separate falsification analyses: when using randomly assigned, rather than actual, birthdays or when examining diagnoses in weeks prior to birthdays. Second, our analysis relied on administrative health care data, rather than laboratory-confirmed data, to identify COVID-19 diagnoses. However, in a recent study, administrative diagnosis codes for COVID-19 had reported sensitivity of 98.0% and specificity of 99.0%,11 and any bias in administrative diagnosis codes should not be associated with the timing of exposures (birthdays). Third, we relied on COVID-19 diagnoses that appeared in medical claims data and thus increased to the level of severity that required medical care. We were thus unlikely to capture asymptomatic cases that did not require medical attention, which could lead us to underestimate total COVID-19 transmission after household birthdays. Fourth, if households with birthdays were more likely to seek COVID-19 testing simply because they hosted a birthday party, an ascertainment bias in testing could occur because cases that would otherwise have gone unidentified would be identified in households that hosted birthdays. However, we found a similar association between birthdays and COVID-19 hospitalizations, an outcome that should not be susceptible to ascertainment bias. Fifth, we focused on a specific event for which people may gather, birthdays, because this event was identifiable in administrative data. Although this event may not be generalizable to other types of gatherings, studying changes in COVID-19 diagnosis rates in the weeks after birthdays is arguably more relevant to understanding risks of SARS-CoV-2 transmission when gatherings are likely to be small and local, compared with gatherings that may occur around holidays (where long-distance travel may also have occurred) or in public settings (eg, bars, restaurants, or large public events).1,21,22 Sixth, our study used medical claims data from a large population with employer-sponsored private insurance that, while broadly representative of this population, did not include patients with public insurance (eg, Medicare and Medicaid) or patients without insurance. These patient populations are more vulnerable to COVID-19 infection.23 If similar gatherings around birthdays occur for these populations, then our study results may understate the association of birthdays with COVID-19 transmission. Seventh, it is possible that the households may be less likely to host gatherings for birthdays when county COVID-19 prevalence is high. If true, this precautionary behavior would lead us to underestimate the association between birthdays and subsequent COVID-19 infections in counties with a high prevalence.

    Conclusions

    We quantified the likelihood of COVID-19 infection within households associated with small social gatherings by studying changes in household COVID-19 infection rates after the presence of a birthday in a household. Using administrative health care data on 2.9 million households, with information on birthdays and COVID-19 diagnoses for each household member, we found that the presence of a birthday within a household was associated with significantly greater COVID-19 diagnosis rates in those households in the 2 weeks after birthdays.

    Back to top
    Article Information

    Accepted for Publication: May 1, 2021.

    Published Online: June 21, 2021. doi:10.1001/jamainternmed.2021.2915

    Corresponding Author: Anupam B. Jena, MD, PhD, Department of Health Care Policy, Harvard Medical School, 180 Longwood Ave, Boston, MA 02115 (jena@hcp.med.harvard.edu).

    Author Contributions: Drs Whaley and Jena had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Whaley, Cantor, Jena.

    Acquisition, analysis, or interpretation of data: All authors.

    Drafting of the manuscript: Whaley, Jena.

    Critical revision of the manuscript for important intellectual content: All authors.

    Statistical analysis: Whaley, Cantor, Jena.

    Obtained funding: Whaley.

    Administrative, technical, or material support: Whaley, Pera, Jena.

    Supervision: Jena.

    Conflict of Interest Disclosures: Dr Whaley reported receiving consulting fees from Doximity outside the submitted work. Dr Jena reported receiving personal fees from Pfizer, Bioverativ, Bristol Myers Squibb, Merck, Janssen, Edwards Life Sciences, Novartis, Amgen, Eli Lilly, Vertex, AstraZeneca, Celgene, Tesaro, Sanofi Aventis, Precision Health Economics, and Analysis Group outside the submitted work. No other disclosures were reported.

    Funding/Support: Support was provided by grant 1K01AG061274 from the National Institutes on Aging (Dr Whaley).

    Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    References
    1.
    Hobbs  CV, Martin  LM, Kim  SS,  et al; CDC COVID-19 Response Team.  Factors associated with positive SARS-CoV-2 test results in outpatient health facilities and emergency departments among children and adolescents aged <18 years—Mississippi, September-November 2020.   MMWR Morb Mortal Wkly Rep. 2020;69(50):1925-1929. doi:10.15585/mmwr.mm6950e3 PubMedGoogle ScholarCrossref
    2.
    Nishi  A, Dewey  G, Endo  A,  et al.  Network interventions for managing the COVID-19 pandemic and sustaining economy.   Proc Natl Acad Sci U S A. 2020;117(48):30285-30294. doi:10.1073/pnas.2014297117 PubMedGoogle ScholarCrossref
    3.
    Gao  S, Rao  J, Kang  Y,  et al.  Association of mobile phone location data indications of travel and stay-at-home mandates with COVID-19 infection rates in the US.   JAMA Netw Open. 2020;3(9):e2020485. doi:10.1001/jamanetworkopen.2020.20485PubMedGoogle Scholar
    4.
    Courtemanche  C, Garuccio  J, Le  A, Pinkston  J, Yelowitz  A.  Strong social distancing measures in the United States reduced the COVID-19 growth rate.   Health Aff (Millwood). 2020;39(7):1237-1246. doi:10.1377/hlthaff.2020.00608 PubMedGoogle ScholarCrossref
    5.
    Meyerowitz  EA, Richterman  A, Gandhi  RT, Sax  PE.  Transmission of SARS-CoV-2: a review of viral, host, and environmental factors.   Ann Intern Med. 2021;174(1):69-79. doi:10.7326/M20-5008PubMedGoogle ScholarCrossref
    6.
    Chen  YH, Glymour  MM, Catalano  R,  et al.  Excess mortality in California during the coronavirus disease 2019 pandemic, March to August 2020.   JAMA Intern Med. 2021;181(5):705-707. doi:10.1001/jamainternmed.2020.7578PubMedGoogle ScholarCrossref
    7.
    Dave  D, Friedson  AI, Matsuzawa  K, Sabia  JJ.  When do shelter-in-place orders fight COVID-19 best? policy heterogeneity across states and adoption time.   Econ Inq. 2020. doi:10.3386/w27091PubMedGoogle Scholar
    8.
    Whaley  CM, Pera  MF, Cantor  J,  et al.  Changes in health services use among commercially insured US populations during the COVID-19 pandemic.   JAMA Netw Open. 2020;3(11):e2024984. doi:10.1001/jamanetworkopen.2020.24984 PubMedGoogle Scholar
    9.
    Whaley  C, Schneider Chafen  J, Pinkard  S,  et al.  Association between availability of health service prices and payments for these services.   JAMA. 2014;312(16):1670-1676. doi:10.1001/jama.2014.13373 PubMedGoogle ScholarCrossref
    10.
    Cantor  JH, Sood  N, Bravata  D, Pera  M, Whaley  CM. The impact of the COVID-19 pandemic and policy response on health care utilization: evidence from county-level medical claims and cellphone data. National Bureau of Economic Research. 2020. Accessed April 29, 2021. https://www.nber.org/papers/w28131
    11.
    Kadri  SS, Gundrum  J, Warner  S,  et al.  Uptake and accuracy of the diagnosis code for COVID-19 among US hospitalizations.   JAMA. 2020;324(24):2553-2554. doi:10.1001/jama.2020.20323 PubMedGoogle ScholarCrossref
    12.
    USAFacts. Our nation, in numbers. Accessed April 29, 2021. https://usafacts.org/
    13.
    Wilson  DJ. Weather, mobility, and COVID-19: a panel local projections estimator for understanding and forecasting infectious disease spread. Federal Reserve Bank of San Francisco. February 2021. Accessed April 29, 2021. https://www.frbsf.org/economic-research/files/wp2020-23.pdf
    14.
    Barrios  JM, Benmelech  E, Hochberg  YV, Sapienza  P, Zingales  L.  Civic capital and social distancing during the Covid-19 pandemic.   J Public Econ. 2021;193:104310. doi:10.1016/j.jpubeco.2020.104310 PubMedGoogle Scholar
    15.
    Cook  J, Newberger  N, Smalling  S.  The spread of social distancing.   Econ Lett. 2020;196:109511. doi:10.1016/j.econlet.2020.109511 PubMedGoogle Scholar
    16.
    Prasad  V, Jena  AB.  Prespecified falsification end points: can they validate true observational associations?   JAMA. 2013;309(3):241-242. doi:10.1001/jama.2012.96867 PubMedGoogle ScholarCrossref
    17.
    Leventhal  AM, Dai  H, Barrington-Trimis  JL,  et al.  Association of political party affiliation with physical distancing among young adults during the COVID-19 pandemic.   JAMA Intern Med. 2021;181(3):399-403. doi:10.1001/jamainternmed.2020.6898PubMedGoogle ScholarCrossref
    18.
    Berry  CR, Fowler  A, Glazer  T, Handel-Meyer  S, MacMillen  A.  Evaluating the effects of shelter-in-place policies during the COVID-19 pandemic.   Proc Natl Acad Sci U S A. 2021;118(15):118. doi:10.1073/pnas.2019706118 PubMedGoogle ScholarCrossref
    19.
    Goolsbee  A, Syverson  C.  Fear, lockdown, and diversion: Comparing drivers of pandemic economic decline 2020.   J Public Econ. 2021;193:104311. doi:10.1016/j.jpubeco.2020.104311 PubMedGoogle Scholar
    20.
    Gupta  S, Nguyen  TD, Rojas  FL,  et al Tracking public and private responses to the COVID-19 epidemic: evidence from state and local government actions. National Bureau of Economic Research. April 2020. Accessed April 29, 2021. https://www.nber.org/papers/w27027
    21.
    Harris  JE. Geospatial analysis of the September 2020 coronavirus outbreak at the University of Wisconsin–Madison: did a cluster of local bars play a critical role? National Bureau of Economic Research. November 2020. Accessed April 29, 2021. https://www.nber.org/papers/w28132
    22.
    Dave  D, McNichols  D, Sabia  JJ.  The contagion externality of a superspreading event: the Sturgis motorcycle rally and COVID-19.   South Econ J. 2020. doi:10.3386/w27813 PubMedGoogle Scholar
    23.
    Rosenthal  N, Cao  Z, Gundrum  J, Sianis  J, Safo  S.  Risk factors associated with in-hospital mortality in a US national sample of patients with COVID-19.   JAMA Netw Open. 2020;3(12):e2029058. doi:10.1001/jamanetworkopen.2020.29058PubMedGoogle Scholar
    ×