SARS-CoV-2 Test-to-Stay in Daycare

Analyses were performed under a protocol approved by the institutional review board of the Medical Faculty of the University of Cologne (21-1358).

The screening was implemented under the direction of the Youth Welfare Office of Cologne and accompanied by a team of members of the health authorities of Cologne, the University Hospital of Cologne, and a private diagnostic laboratory (Labor Quade). Samples were collected using the Lolli-Method, with a previously determined sensitivity and specificity of 93% to 100% and 100%, respectively (Fig 1).^17 –19  It consists of sucking a nasopharyngeal swab for 30 seconds and subsequent testing in SARS-CoV-2 Pool-PCR. Samples were collected voluntarily by the children under supervision and support of the daycare staff members or parents. For pool-testing, each child of a daycare group placed a Lolli-swab in a common collection tube. This tube was tested in one PCR-reaction. When a Pool-PCR tested negative, all children of that pool were assumed to be SARS-CoV-2 negative. When a Pool-PCR tested positive, the respective children were retested individually on the next day in Single-PCRs. The identification of an infected child was followed by the isolation of this child. Children exposed to this child were quarantined up to 14 days (median duration 10 days; interquartile range [IQR]: 8–12; Supplemental Fig 6), depending on the specifications of the current Infection Protection Act and the individual situation of the respective contact. During the test-to-stay approach, exposed children were not quarantined but tested in Lolli-Single-PCRs for 5 subsequent days. If tested negative, they could continue to go to daycare.

PCRs were performed at the Institute of Virology, University Hospital Cologne, and at Labor Quade. Labor Quade reported PCR test data to the Institute of Virology, containing the date of sampling, specimen (Pool- or Single-PCR), test result, and daycare ID (Supplemental Fig 7). During weeks 15 to 17 in 2021, daycare facilities were enrolled gradually, and the data collection system was under construction and misclassified the specimen of positive PCRs. These weeks were excluded from all analyses. Pool sizes were reported as weekly average pool size per daycare facility.

The health authorities of Cologne provided data on newly reported SARS-CoV-2 infections (“index cases”) and contact persons recorded with a software developed for the documentation and case management of SARS-CoV-2 in Cologne.²⁰  The data contained the date of the first SARS-CoV-2 detection or the date of the beginning and the end of the quarantine, age, sex, and the zip code of the place of residency (Supplemental Fig 7).

Data on Germany-wide proportions of all reported sequences of variants of concern (VOCs) were collected as part of a national surveillance system and published by the Robert Koch Institute.²¹  The 7-day incidence was extracted from [email protected], an online tool for querying notifiable infectious diseases in Germany.²² 

Aggregated data on the number of inhabitants in 2020 and socioeconomic factors were published online or provided by the Office for Urban Development and Statistics of Cologne.²³ 

We reported counts and fractions of total and positive Pool-and Single-PCRs. Pool sizes were reported as median of the weekly mean pool sizes per daycare facility. Cumulative numbers of PCR analyses were estimated by multiplication of the counts of Pool-PCRs with the weekly median pool size and addition of the Single-PCRs. We calculated Spearman correlation coefficient to quantify the association between the weekly fraction of positive Pool-PCRs and the 7-day incidence.

To account for the variability of the individual parameters, we derived a variability range for the days of avoided quarantine by multiplying the first or third quartile of the pool size with the number of positive Pool-PCRs (equation 1) and the exposed children with the first or third quartile of the duration of quarantine (equation 2). The days of avoided quarantine per eligible child were estimated, dividing the number of avoided quarantine days by the number of inhabitants in Cologne aged 2 to 6 years (n = 51 830), assuming that those were eligible for the screening.

The outcome for the regression discontinuity in time (RDiT) was defined as the weekly count of index cases aged 2 to 6 years in Cologne divided by the weekly count of positive Pool-PCRs (Supplemental Fig 8). We assumed that an infected child in a daycare facility would be detected in the screening and the corresponding positive Pool-PCR would contribute to the denominator of the outcome. We expected that secondary infections would occur in the daycare facility. The infected child and the secondary infections would be reported to the health authorities and contribute to the count of reported index cases aged 2 to 6 years in Cologne, which, in addition to the reported infections in children that were acquired elsewhere than at daycare, defined the numerator of the outcome. We hypothesized that the transition from the quarantine approach to the test-to-stay approach might affect the number of secondary infections and, thus, the outcome.

Analyses were restricted to weeks that did not meet the criteria of low-testing periods as we expected that in those, the denominator of the outcome would be underestimated (eg, during roll-out of the screening or holiday seasons). We defined a low-testing period as a week in which less than 70% of the average weekly count of Pool-PCRs during the entire screening were performed.

The width of the time period (“bandwidth”) drawn around the interruption is an important choice in RDiT.^24 ,25  We determined a mean squared error (MSE)-optimal bandwidth, minimizing the MSE of the regression fit.^28 ,29  We examined variations of the β₂-coefficient across different bandwidth choices (1, 1.25, 1.5, 1.75, and 2 times the MSE-optimal bandwidth).

We extended the model specification given in equation (3) to a second-order polynomial regression instead of a linear regression (Appendix 1).

RDiT requires continuity of baseline covariates across the interruption.^24 ,25  This verifies that changes in the outcome are attributable to the defined interruption and not to coincident changes of covariates. Continuity was assessed with the help of the RDiT framework by applying local linear regressions. Assuming normal distribution, we modeled the median pool size, the fraction of the Delta variant, the 7-day incidence, and the weekly fraction of susceptible children (Appendix 2).

We included the 7-day incidence, the seasons, and the fraction of the respective SARS-CoV-2 variant in Germany in the Gamma regression model (Supplemental Information, Appendix A). We ran stratified analyses for districts with low and middle or high socioeconomic status (SES), as well as age-ranges of the reported index cases (2–3 years and 4–6 years) (Appendix 3).³⁰ 

Analyses were performed using Microsoft Excel for Mac (v.14.7.3.), Prism 9.0 (GraphPad) and R within RStudio (v. 2023.03.0 + 386) and additional R packages.^31 –33 

The screening was conducted from March 2021 to April 2022 in Cologne, Germany (1.1 million inhabitants). All daycare facilities in Cologne and 51 830 children aged 2 to 6 years were eligible to participate twice weekly. The quarantine approach was conducted from week 11 to 36 (2021) and was substituted by the test-to-stay approach from week 37 (2021) to week 14 (2022) (Table 1, Fig 2). During the screening, 714 daycare facilities participated, 4 VOCs emerged (Alpha, Beta, BA.1, and BA.2), and the SARS-CoV-2 7-day incidence in Cologne ranged from 8.67 infections per 100 000 inhabitants (week 25, 2021) to 2573 (week 9, 2022) (Supplemental Fig 10).

219 885 Pool-PCRs and 352 305 Single-PCRs were performed. Median of weekly mean pool sizes per daycare facility was 11.5 (IQR: 8–14.6) swabs per pool (Supplemental Fig 11). Approximately 2 897 437 SARS-CoV-2 analyses were performed in total (Table 1, Fig 2). Of the Pool-PCRs, 6440 (2.93%) tested positive (Table 1, Fig 3A). The fraction of positive Pool-PCRs ranged from 0.0% in weeks 24 to 27 (2021) to 12.44% in week 9 (2022) (Supplemental Fig 12) and correlated strongly with the total SARS-CoV-2 7-day incidence in Cologne (r_s = 0.96, CI: 0.94–0.98; Fig 3B). Of the Single-PCRs, 12 454 tested positive. During the entire screening, the health authorities in Cologne reported 17 208 index cases among children aged 2 to 6 years (Table 1, Fig 3A). Subsequently, our estimation suggests that the screening identified 72.4% of all reported index cases aged 2 to 6 years in Cologne.

During the test-to-stay approach, 6298 Pool-PCRs tested positive (Table 1). The median duration of quarantine after exposure to SARS-CoV-2 of children aged 2 to 6 years in Cologne was 10 days (IQR: 8–12) (Supplemental Fig 6). Consequently, we estimated that 661 290 days (variability range: 352 688–1 027 833) spent in quarantine were avoided during the test-to-stay approach. This translated to 13 days (variability range: 7–20) of avoided quarantine per eligible daycare child during a period of 30 weeks (Fig 4).

Applying RDiT analyses, we aimed to assess the safety of the test-to-stay approach in comparison with the quarantine approach with the underlying causal framework depicted in a directed acyclic graph (Supplemental Fig 13, Supplemental Table 2).²⁵  We first incorporated all observations in the Gamma regression model (“global Gamma regression model”), which yielded an exponentiated β₂-coefficient of 0.74 (95% CI: 0.52–1.06) (Fig 5A). This translated to 26% less index cases per positive Pool-PCR attributable to the change of the test concept and indicated no substantial increase in infections among children. No clear presence of autocorrelated residuals was detectable (Supplemental Fig 14). To address biases arising from potential unobserved confounders that are not in proximity of the interruption, we gradually narrowed the underlying bandwidth of the global Gamma regression model in a sequence of local Gamma regressions (Fig 5B, Supplemental Fig 15). Overall, the main result of the global Gamma regression model was not sensitive to the choice of the bandwidth, minimizing the risk of bias because of time-dependent factors, such as immunization rates. We verified that the result of the global Gamma regression model was similar when using a second-order polynomial specification instead of a linear regression (Supplemental Fig 16).

The weekly median pool size (β₂ = −0.89, CI: −4.12 to 3.23), the share of the Delta variant (β₂ = −0.57, CI: −1.42 to 0.28), and the weekly fraction of susceptible children (β₂ = 0.28, CI: −0.07 to 0.63) showed continuity among the interruption. The 7-day incidence did not meet the continuity assumption (β₂ = −147, CI: −244 to −49) (Supplemental Fig 17).

We accounted for 7-day incidence, seasons, and SARS-CoV-2 variant in separate global models and detected no substantial differences of the corresponding effect estimates or their respective precision (Supplemental Fig 18). Stratification by SES and age-range indicated differences between the statistical significance of the β_2-coefficients of the analyzed strata but there was no indication of an increase in infection frequency (Supplemental Fig 19 and 20).

This study provided evidence suggesting safety of a SARS-CoV-2 Pool-PCR test-to-stay screening in daycare facilities as compared with a post-exposure quarantine concept. We analyzed one of the most comprehensive test-to-stay screenings in daycare facilities worldwide.^34 –37  This screening was one of the earliest attempts to disestablish the use of quarantine of daycare children in Germany. Our estimation of the impact of the test-to-stay concept on quarantine avoidance showed a substantial reduction in the time spent in quarantine. Our RDiT analyses did not indicate evidence of a discernible increase in infections after quarantine measures were discontinued.

Post-exposure quarantine effectively serves as mitigation strategy.^38 ,39  However, its subsequent disruption of childcare has substantial negative implications for children, such as impairment of cognitive development or executive functions and adverse effects on mental or physical health.^40 –42  Furthermore, quarantine was shown to negatively impact parents, increasing negative mood or likelihood of losing temper and punishment.^43 –45  Additionally, high-frequent and unexpected disruption of daycare attendance impacts workforce participation of parents because of the necessity of at-home childcare.^44 ,46  Thus, strategies of reducing duration of quarantine are valuable for children and their families but need thorough safety assessment.

Our study advances current evidence on safety of test-to-stay approaches to the setting of daycare, as previous studies were primarily conducted in schools with children and adolescents older than 6 years, and they incorporated the use of facemasks.^8 –16  In contrast, the screening analyzed in this article was conducted in the absence of the use of facemasks and in an age-group less capable of sticking to hygiene rules. Furthermore, the mentioned studies were conducted during the predominance of one respective SARS-CoV-2 VOC.^8 –16  In contrast, our analyses spanned a period during which 4 distinct SARS-CoV-2 VOCs have emerged (Alpha, Delta, BA.1, and BA.2). These variants show distinct virological features that could affect the effectiveness of infection control measures (eg, incubation period or basic reproduction number).^{27 ,47 –49}  We show that safety of the test-to-stay screening was not affected by these VOCs. Finally, previous studies mainly report on short-term observations with rather small variations in the incidence, whereas our analyses encompass 13 months with low and high incidence periods, providing more generalizable evidence.

One limitation of this study is that the ethical and political circumstances during the COVID-19 pandemic did not justify addressing the research question by a randomized controlled trial, emerging from the initial screening with a quarantine approach. Thus, the quasi-experimental study design differs from the ideal experimental design. Second, as we analyzed aggregated data, we cannot deduce any causal statements on the level of individual children. Further individualization of the data was not possible because of the challenge of matching index cases with their corresponding positive Pool-PCRs. Additionally, matching with contact tracing data was not possible, which does not allow for assessments of changes in infection site or the general risk of infection in daycare, for example by comparing secondary attack rates in daycare with those in households. Third, the risk of bias caused by unmeasured confounders needs to be acknowledged. It might comprise changes in data collection practices of the health authorities or in test indication for SARS-CoV-2 testing beyond the screening. However, we assume that most index cases were constantly reported and recorded, as the screening detected most of all notified index cases among children. Furthermore, we assumed that the risk of confounding or bias, such as changes in immunization, increases with increasing distance from the interruption and applied kernel weights and different bandwidths to address this risk.

The test-to-stay approach substantially reduced the use of quarantine after SARS-CoV-2 exposure. There was no indication of a relevant increase in infections among children with this approach. This highlights it as a safe alternative to quarantine after exposure to SARS-CoV-2 in daycare. Test-to-stay approaches could prove valuable as infection control measures after exposure to emerging respiratory pathogens in daycare during future outbreaks.

The authors thank all children and staff in tested daycare facilities for support and participation; Moritz Lorenz and Paula Lorenz for supporting the development of the Lolli-Method; Sascha Nickel and Anne Fries as well as all staff members of their daycare facilities for their impetus for the development of the Lolli-Method and screening adaptions; all members of the Institute of Virology, University Hospital Cologne; all staff of Labor Quade for conducting the screening and providing data; Stephan Glaremin (Amt für Jugend, Arbeit und Soziales der Stadt Düsseldorf) for administrative support during implementation of the screening in Cologne; Janna Seifried and Sindy Böttcher (Robert Koch Institute) for continuous support and discussions; and Stefan Konigorski and Mayram Ganji (Berlin School of Public Health) for statistical support and discussion.

IQR

interquartile range

MSE

mean squared error

PCR

polymerase chain reaction

RDiT

regression discontinuity in time

RKI

Robert Koch Institute

SES

socioeconomic status