This study was conducted through the Nursing Home Public Health Response Network, a collaborative of eight academic institutions across the United States (see Supporting Information), approximately 50 NHs, a central coordinating center, and a central laboratory that collects and shares real-time data with the Centers for Disease Control and Prevention (CDC) to aid in response to emerging infectious diseases in NHs. From April to November 2023, 24 NH infection preventionists (IPs) performed surveillance per CDC guidance [9] and their facility protocols and notified the academic study team within 24 h of identifying a resident or staff member positive for SARS-CoV-2.

There were two pathways for adult resident and adult staff enrollment, prospectively or after the identification of a positive SARS-CoV-2 test. Residents and staff, regardless of enrollment status, were identified as positive for SARS-CoV-2 by routine testing and public health surveillance, which included exposure testing with both antigen and RT-PCR on Days 1, 3, and 5 per public health recommendations [10]. If prospectively enrolled, roommates of infected residents received more frequent testing on Days 1–7 after exposure. Potential participants were excluded if they had impaired decision-making capacity and no legally authorized representative (LAR) or were unable to communicate in English. Study staff followed enrolled individuals who tested positive for SARS-CoV-2 (“cases”) for 14 days from the first positive test. During each daily testing, two anterior or mid-turbinate nasal swabs were collected using standard methods. One swab was used to conduct on-site antigen testing (CLINITEST or Quidel) per the manufacturer’s instructions. The second swab was shipped to a central laboratory (Marshfield Clinical Research Institute) for further testing (described below). Study staff obtained information on symptoms and antiviral treatment at each testing event via resident interview and/or chart review. Enrolled NH staff self-collected their swabs and directly entered survey responses documenting symptoms, antiviral treatment, and antigen results in a REDCap survey. Participants were allowed to remain in the study if a testing day was skipped.

Study staff abstracted demographic and clinical information (i.e., comorbid conditions, COVID-19 vaccination history, and SARS-CoV-2 infection and treatment history) from resident NH records of enrollees. Information not available from the record was collected through discussion with the resident or their LAR. NH staff enrollees received a questionnaire via email or text to self-complete. Data collection for all sites used standard forms in a centralized REDCap (Research Electronic Data Capture) managed by Abt Global and housed at Vanderbilt University. Participants were compensated for completing milestones of study participation.

All study sites used a common protocol, data collection tools, and standard operating procedures. Abt served as a central Institutional Review Board (IRB) with reliance from each site’s IRB. The Abt IRB determined that the study presented no greater than minimal risk per 45CFR46 and approved the request for a waiver of written informed consent per 45CFR46.117(c)(1)(ii) though some site IRBs required written consent (JHU, UoP) [11]. This study was funded by the US CDC. This study was reviewed and approved by the Abt, Brown University, Case Western Reserve University, Emory University, Johns Hopkins University, Oregon State University, University of Michigan, University of Pittsburgh, and University of Wisconsin IRBs.¹

All swabs sent to Marshfield were tested by qualitative RT-PCR using the TaqPath COVID-19, FluA, FluB Combo Kit on QuantStudio 7 Pro. The qualitative RT-PCR returned cycle threshold (Ct) values, which are inversely correlated with the viral RNA copy number (load) in a sample. Those that tested positive (Ct ≤ 37·0) underwent quantitative RT-PCR testing using the CDC Influenza SARS-CoV-2 (Flu SC2) Multiplex Assay. All Ct values were calculated by Marshfield, which verified a direct association with the log viral load (quantitative result) through a standard curve generated from synthetic SARS-CoV-2 RNA [12]. VC and WGS were performed on a participants’ first test specimen with a Ct < 30; this cutoff was selected based on previous work conducted by Marshfield [12]. WGS was performed by Marshfield. VC was performed by St. Jude Children’s Research Hospital on one qualitative RT-PCR specimen with Ct < 30 collected during each of the following intervals for each participant: Days 1–2, 3–5, 6–8, 9–11, and 12–14. If multiple samples had Ct < 30 in an interval, VC was performed on the specimen with the lowest Ct value. Infectious virus particles were quantified using an immune-staining-based TCID₅₀ in which nasal samples were titrated and incubated for 4–5 days on Vero cells ectopically expressing TMPRSS2 and human ACE2. Infected cell monolayers were stained, and virus-positive cells were identified by two-fold increase in optical density compared to the mean value of negative wells.

Descriptive analyses were conducted to describe enrolled participants and cases. We reported demographic data and WGS on all cases, but restricted statistical analysis to cases with ≥ 10 days of testing of any test type to limit the effect of loss to follow-up. A Chi-squared test was used to compare the proportion of antigen and VC positives during early illness (Days 0–5) compared to later illness (Day ≥ 6) and for Ct < 25 compared to Ct ≥ 25. Resident’s and staff’s duration of shedding was compared using Kaplan–Meier survival analysis on time from first positive to first negative test result for RT-PCR and antigen testing, restricted to patients who had at least 10 days of follow-up after first positive test of given type. Similar analyses were conducted for symptomatic vs. asymptomatic cases with the addition of time to first Ct > 30. Log-rank tests were used to compare survival curves.

To estimate the viral load within an individual over time (i.e., viral trajectory), gamma distribution functional forms were fitted to Ct values [13, 14]. For each participant, we used a likelihood-based approach to calculate optimal gamma distribution parameters (shape, scale, and location) and a scaling coefficient. This produced an estimated viral trajectory for all participants. Participant viral trajectories were aligned by their estimated peak viral load, then the data from individuals were used to fit a new population-level gamma distribution functional form for staff and residents separately. Participants with no measured Ct value before estimated peak viral load were excluded from population-level fitting. From these estimated population trajectories, we calculated proxies for the proliferation period (time from passing the RT-PCR positivity threshold [Ct ≤ 37] to the peak Ct value) and decay period (time from peak Ct to passing the RT-PCR positivity threshold) with non-parametric 95% confidence intervals (CIs). Participants with fewer than five measured Ct values < 43.64 were excluded from individual functional form fitting, as this was deemed insufficient to fit the viral trajectory. Participants treated with antivirals were also excluded from individual functional form fitting because antivirals alter the viral trajectory. See Supporting Information for details.

Descriptive analyses of viral load rebound were conducted for all participants regardless of antiviral treatment. Rebound was defined as a 10-fold increase where the amount of increase was at or above five log copies per milliliter after the completion of SARS-CoV-2 antiviral treatment, or if the participant was untreated, 7 days after the initial positive test [15]. Duration of antiviral treatment was assumed to be 5 days [16] unless a longer treatment course was documented.

CDC staff were involved in the study design, in the writing of the report, and in the decision to submit the paper for publication.

Residents tested positive by both antigen and RT-PCR for longer than staff. However, this result was only statistically significant for RT-PCR results (p = 0.007) (Figure 2). Five days after the first positive test, 29/51 of residents (57%) and 13/21 of staff (62%) remained antigen positive, and 37/42 (88%) and 16/20 (80%), respectively, remained RT-PCR positive. Ten days after the first positive test, 10/51 of residents (20%) and 1/21 of staff (5%) remained antigen positive, and 24/42 (57%) and 4/20 (20%), respectively, remained RT-PCR positive. The median duration of antigen positivity was 6 days in both residents and staff. The median duration of RT-PCR positivity was 11 days for residents and 8 days for staff.

Nine participants (eight residents, one staff) were considered asymptomatic (no symptoms reported during 14-day survey period). Symptomatic participants (i.e., those with at least 1 day of reported symptoms) experienced a longer time to reach their first negative antigen test (median 7 vs. 4 days, p < 0·001) and Ct > 30 (median 7 vs. 4 days, p = 0·013). All asymptomatic participants tested antigen negative by day 6 and reached a Ct > 30 by Day 7 (Figure S2).

Of 218 RT-PCR-positive specimens that underwent VC, 68 (31%) were VC positive. Antigen tests were sensitive to VC positivity (96%) but not specific (28%). Both antigen and VC were more likely to be positive early (Days 0–5) (antigen: 82% [296/361] vs. 23% [113/500] p < 0·001, VC: 44% [54/122] vs. 15% [14/96] p < 0·001) and at lower Ct values (Ct < 25) (antigen: 90% [234/261] vs. 21% [109/528] p < 0·001, VC: 42% [66/157] vs. 3% [2/61] p < 0·001) by a Chi-squared test (Figures 3, 4).

Positive VC that occurred after Day 10 belonged to two participants who were treated with antivirals, were concurrently antigen positive, and experienced laboratory-defined rebound in viral shedding.

Of the 29 residents and 16 staff who met criteria for fitting individual gamma distribution function forms (i.e., ≥ 5 Ct values < 43.64 and no antivirals), only 15 (52%) residents and seven (44%) staff had at least one measured Ct value preceding the model estimated peak Ct value. This implied that the peak Ct value was missed for 48% of residents and 56% of staff.

At the population level, the functional form approximated proliferation period was 4.70 (95% CI 3.94, 7.27) days for residents and 3.35 (95% CI 1.84, 5.43) days for staff. The decay period was 10.83 (95% CI 9.71, 12.14) days for residents and 8.66 (95% CI 7.86, 9.67) days for staff (Figures S11, S12).

Nine residents and three staff in the study met criteria for rebound [15]. Three were treated with molnupiravir, four were treated with nirmatrelvir—ritonavir, four were untreated, and one had no treatment data. Of those who experienced rebound, 11 had specimens that were tested for VC during the rebound period and 55% (6/11) were positive (Figure S4).

Due to potential discomfort and the need to obtain two swabs for both antigen and RT-PCR tests, we used anterior nasal samples rather than nasopharyngeal samples, which may have resulted in lower detection of viral material and could have affected our assessment of viral load or antigen positivity. Our sample size was too small to determine significant differences in terms of clinical comorbidities, vaccination status, or symptoms; however, trends were consistent with previously published data. While nine residents appeared to exhibit rebound viral load, these data cannot be used to compare treatment types due to the low sample size and few patients who received treatment across the cohort. Finally, the study design relied on NH site IPs to notify the study team of new positive cases in the facility. There were instances when this notification process was delayed, leading to a lag in testing and symptom assessment that reduced our ability to track the full viral trajectory of shedding among all enrolled individuals; we attempted to mitigate this limitation by having study staff on site in the facility and ensuring regular communication with the NHs.

This study had several strengths. By prospectively assisting with outbreak investigations and sampling exposed participants prior to symptoms, we were able to obtain a more complete view of viral shedding to visualize the peak of viral load, something that is often missed in studies which enroll participants days after an initial positive test or the development of symptoms. We were also able to identify individuals who tested positive but never developed symptoms and compare kinetics in asymptomatic and symptomatic individuals. Further, because all samples were processed in a central laboratory, we could reliably compare Ct values, VC, and viral loads across the study participants, a limitation of other multi-center studies relying on several laboratories.