Implementation of rapid and frequent SARS-CoV2 antigen testing and response in congregate homeless shelters

Background People experiencing homelessness who live in congregate shelters are at high risk of SARS-CoV2 transmission and severe COVID-19. Current screening and response protocols using rRT-PCR in homeless shelters are expensive, require specialized staff and have delays in returning results and implementing responses. Methods We piloted a program to offer frequent, rapid antigen-based tests (BinaxNOW) to residents and staff of congregate-living shelters in San Francisco, California, from January 15th to February 19th, 2021. We used the Reach-Effectiveness-Adoption-Implementation-Maintenance (RE-AIM) framework to evaluate the implementation. Results Reach: We offered testing at ten of twelve eligible shelters. Shelter residents and staff had variable participation across shelters; approximately half of eligible individuals tested at least once; few tested consistently during the study. Effectiveness: 2.2% of participants tested positive. We identified three outbreaks, but none exceeded 5 cases. All BinaxNOW-positive participants were isolated or left the shelters. Adoption: We offered testing to all eligible participants within weeks of the project’s initiation. Implementation: Adaptations made to increase reach and improve consistency were promptly implemented. Maintenance: San Francisco Department of Public Health expanded and maintained testing with minimal support after the end of the pilot. Conclusion Rapid and frequent antigen testing for SARS-CoV2 in homeless shelters is a viable alternative to rRT-PCR testing that can lead to immediate isolation of infectious individuals. Using the RE-AIM framework, we evaluated and adapted interventions to enable the expansion and maintenance of protocols.


Study setting and design
The UCSF Benioff Homelessness and Housing Initiative (BHHI), the San Francisco Department of Public Health (SFDPH) and Department of Homelessness and Supportive Housing (SFHSH) developed and implemented the BinaxNOW shelter pilot. Initially, we implemented the pilot at four shelters and added six shelters during the study.
All residents and staff in adult, transitional age youth, and family shelters with congregate dormitory settings in San Francisco were eligible. To detect and isolate individuals as close to their infective period as possible, given resource limitations, test sensitivity, and disease progression (time between exposure, infection, and test positivity), we offered participants twiceweekly testing (Monday/Thursday or Tuesday/Friday) during the study period of January 14 th to February 19 th , 2021.
All participants (or parents or guardians for those under 18) provided verbal informed consent. Consent was documented by non-clinical or clinical staff in a web-based data management platform (PrimaryHealth). We collected data required by state guidelines (contact information, race/ethnicity, gender, and date of birth) in PrimaryHealth. The UCSF Committee on Human Research (UCSF's Institutional Review Board) stated that the project did not meet criteria for human subjects research (and thereby did not require formal Institutional Review Board review) because this study was an evaluation of a public health intervention in the context of "public health surveillance" for SARS-CoV2, rather than a research study.

BinaxNOW shelter pilot implementation
We based our protocol on previous test-and-respond workflows (Table 1), which involved monthly rRT-PCR for shelter residents and encouragement of shelter staff to seek community testing twice monthly. An outbreak team worked with shelter leadership to refer positive individuals to I&Q and conduct a case investigation and contact tracing.
Staffing and sites preparation. A detailed description of staffing and supplies is available online [24]. We recruited "shelter champions," shelter staff members whom we trained and provided with a stipend to advertise testing events, coordinate event setup, and facilitate resident and staff participation. BHHI hired non-clinical staff and laboratory technicians, and recruited community volunteers; SFDPH provided registered nurses (RN) and health workers.
On-site workflow. Initially, we had four testing stations with 8-10 staff (Fig 1): (1) Check-In: three to four non-clinical staff checked in participants, performed a symptom screen [25], provided information about isolation and quarantine, and labeled testing materials. A RN assessed individuals who reported symptoms consistent with COVID-19 to determine if they required isolation; (2) Swabbing: one to two technicians performed anterior nares swab [26]; (3) Testing: one BinaxNOW-trained staff (Tester) started the reaction [26], and one (Recorder) monitored the reaction time, read the result with the Tester, and recorded results on Primary-Health. The RN served as tiebreaker when there was disagreement. (4) Results: A non-clinical staff disclosed negative results to participants [24].
We performed rRT-PCR tests for the first 40 participants for validation [27]. Result disclosure and BinaxNOW-positive participant investigation. The RN, with the shelter champion, located BinaxNOW-positive residents, isolated them, and conducted a case investigation and contact tracing. When staff tested positive, the RN phoned them, gave them isolation guidelines, and elicited if they had close contacts with residents while infectious. We referred close contacts to I&Q.
We provided results within the same shift. PrimaryHealth sent automated emails and text messages to all participants with a link to their results. They sent reports to state and local health departments.
I&Q referral, confirmatory testing and outbreak team integration. We defined outbreaks as at least three COVID-19 cases within a 14-day period in epidemiologically-linked residents and/or staff (i.e., persons with a close contact to a case or a member of risk cohort) [28].
We isolated symptomatic BinaxNOW-negative and all BinaxNOW-positive participants. The RN or shelter champion completed I&Q referrals. Upon arrival to I&Q, asymptomatic BinaxNOW-positive and symptomatic BinaxNOW-negative individuals received confirmatory tests (rRT-PCR or transcription mediated amplification, TMA). Confirmatory test-negative residents returned to the shelter.
Our team notified the shelter of any positive cases, who paused new admissions upon identification of two BinaxNOW-positive cases or one BinaxNOW-positive and one symptomatic BinaxNOW-negative case in a 14-day period.

SYMPTOM SCREENING
• Non-clinical shelter staff screen residents for symptoms daily and on entry to shelter • Screening increases to twice daily when a case/outbreak is identified • Screening guidance remains the same • Clinical and non-clinical BinaxNOW team staff perform additional symptom screening of all participants during each testing event.

COVID-19 EDUCATION & ENGAGEMENT
• Days of outreach and COVID-19 education in advance of every testing event.
• Frequent COVID-19 education and engagement during testing events. Resident participation incentives. We provided monetary incentives to residents who tested regularly (twice in the first two weeks of testing or >75% of tests over the last four weeks) in all but one shelter.

Iteration: Protocol modifications
Team members met daily on weekdays to discuss protocol modifications.
Modifications to staffing and supplies preparation. Starting in week three, to improve the implementation of the pilot, we conducted online training sessions for shelter champions (COVID-19 basics and role overview).
Modifications to on-site workflow. Starting in week five, to increase the number of shelters testing and enable maintenance without external (BHHI) resources, we reduced steps, supplies and staff [24]. We incorporated self-swabbing and reduced teams to five personnel: two Testers who, in addition to their original role, checked-in participants, instructed them on self-swabbing, and supervised self-swabbing; one Recorder; and one RN who, in addition to their original role, disclosed negative results. The shelter staff champion continued in their role. To optimize flow of patients and samples, we arranged the team in one station (Fig 1). The Testers gave self-swabbing instructions at the time of collection. We provided no other prior training to participants. The participants immediately handed the swabs to the Tester after sampling to start the reaction.
To try to increase reach, we modified schedules. Modifications to result disclosure and BinaxNOW-positive participant investigation. In week six, in response to miscommunication about a positive case, we modified the protocol to ensure case identification: (1) the RN reviewed staff responsibilities, and all symptomatic BinaxNOW-negative and BinaxNOW-positive participants at each event; and (2) Primary-Health sent a report with symptomatic BinaxNOW-negative and all BinaxNOW-positive participants to the team every day.
Modifications to I&Q referral, confirmatory testing and outbreak team integration. Starting in week three, because not all asymptomatic BinaxNOW-positive residents received confirmatory testing at the I&Q hotels, the RN completed all I&Q referrals and assessed adherence to confirmatory testing.

Implementation evaluation
Overall approach. We used the RE-AIM framework to guide implementation planning and evaluation (Table 2). We established reach and effectiveness criteria to mirror shelter realities (reach measures). We evaluated case detection and outbreak control (effectiveness). For adoption, we focused on the number of shelters participating, and their ability to offer testing to eligible participants. For implementation, we assessed the consistency of protocols and adaptations made to increase reach and effectiveness. For maintenance, we assessed whether SFDPH was able to sustain the project without outside resources.
Data collection. Reach: SFHSH provided de-identified individual resident census data (including demographics, date of birth, and admission and discharge dates), and staff census data. To assess how many residents were on-site during testing events, the testing team conducted a headcount within the facilities during testing events. To estimate staff who were onsite during daytime testing events, SFHSH provided a count of all staff working day shifts.
Effectiveness: We recorded symptoms and test results on PrimaryHealth. We obtained confirmatory test results from the electronic health record.
We also recorded the timeframe of implementation and number of shelters where we implemented BinaxNOW Testing (adoption), the rationale for changes to protocols, and their outcomes (implementation), and the number of shelters where SFDPH offered testing after the study period (maintenance).
Data analysis. We analyzed data using Matlab [29]. We used census data for the study period corresponding to each shelter. We excluded residents admitted and discharged on the same day. We conducted Mann-Whitney U test to compare age, and chi-squared test to compare gender for the participating and non-participating resident populations. We included shelters with <10% missing gender data in gender comparisons based on the average percent and confidence interval across shelters. We were unable to examine race and ethnicity due to incomplete data.
To determine the period participant residents lived in a shelter, we matched them to the shelter census database using their date of birth. We used the mean number of residents onsite. We assumed participant staff had continuous employment over the study period.
To estimate adherence to twice-weekly testing, we calculated the proportion of tests out of the total number of testing events offered per shelter. We included only shelters with at least six testing events. Because some shelters had high resident turnover, we included only those individuals who remained in the shelter throughout the shelter's testing period. All reported ranges refer to the range of values across shelters.

Reach
Identification and characterization of total eligible participants. We implemented the pilot at ten of 12 eligible shelters during the study period. During the study, 828 unique residents lived in and 435 staff worked at these ten shelters. The mean age among residents and staff was 44 years. Less than one third (28.8%) of residents and nearly half (49.8%) of staff were cis women (Table 3).
Identification and characterization of on-site sample during testing period. On average, the daily census included 645 residents. Over half (57.9% n = 636 in 6 shelters) of residents remained in shelters for the duration of the study (Table 3). Less than half of residents (45.2%, range 10.9% to 81.9%) and one third of staff (31.3%, range 5.4% to 68%) were on-site during testing events ( Table 4).
Completion of twice-weekly testing. Half of eligible residents (47.5%, range 16.5% to 81.4%) and shelter staff (53.3%, range 9.6% to 128%-the latter meaning some staff off their shift tested) participated in testing at least once. On average, one-quarter (24.1%, range 7.3% to 66.7%) of residents and one-fifth (19.6%, range 5.8% to 45.8%) of the staff participated in  (Table 4, S1 File). We found no significant demographic differences between participant and non-participant or eligible populations (S1 Appendix).
Residents completed on average 36.8% and staff 34.1% of the tests offered in the shelters where we tested at least three weeks during the study period (n = 196 for residents, n = 198 for staff). Among these participants, most tested only once; only 4.1% of residents and 3.0% of staff tested twice-weekly during the study (i.e., 100% adherence, Fig 2A, S1 File). Participation was stable over time, although two shelters showed a noticeable increase (shelter 2) or decrease (shelter 5) in participation (Fig 2B).

Effectiveness
Identification of positives and effective isolation. We performed concomitant rRT-PCR on the first 40 participants, all of whom were asymptomatic and BinaxNOW negative. Two (5%) residents tested positive on rRT-PCR. Follow-up tests three days later in I&Q found that the first resident tested negative on BinaxNOW and rRT-PCR, and the second participant, who subsequently developed symptoms, tested positive on both tests. BinaxNOW prevalence was 2.2%: ten residents (2.5%), and four staff (1.8%) in four different shelters tested positive (Fig 2C). Of the 14 BinaxNOW-positive participants, only one resident and two staff were symptomatic. We referred eight of ten residents to I&Q on the same date of the event. Of the two not referred, one resident received isolation guidance and left the shelter. The second resident was not isolated until five days after their positive test due to miscommunication. We identified no resident close contacts during case investigations. The float RN contacted all BinaxNOW-positive staff.
One symptomatic BinaxNOW-negative resident was referred to I&Q. The confirmatory test was negative, and they returned to the shelter.
No residents received a COVID-19 diagnosis outside of BinaxNOW testing during the study; three staff members did.
Identification of incongruent confirmatory tests. Of the eight asymptomatic Binax-NOW-positive residents referred to I&Q, three were confirmed positive; two did not receive a confirmatory test; and three tested negative on confirmatory testing. Two of the three BinaxNOW-positive/confirmatory-negative residents were BinaxNOW-positive in Missing bars indicate no testing event in that date. Shelters 5-10 were added to the BinaxNOW Testing Pilot after the start of the study period. � : Scheduled testing times were modified after indicated event. C) 3 outbreaks and 1 isolated BinaxNOW-positive case were identified. All outbreaks were resolved. No cases were identified in the other 6 shelters during the study period.
Identification of cases and outbreaks. We identified cases in four of the ten shelters. Three shelters met criteria for outbreaks (i.e., � three cases) but no outbreak exceeded five cases (Fig 2C). The three shelters with outbreaks were large, had transient resident populations, or had high participation (Table 4). One shelter had one isolated positive case.

Adoption
Shelter-level offering of BinaxNOW shelter pilot. We implemented testing in ten shelters within weeks of initiating planning. We delayed staff testing for one week for staff to complete release forms.

Implementation
Adaptation and fidelity of implementation. The testing team adopted the new workflow (Fig 1) in all shelters within a week of trying it in one shelter. These changes led to a significant decrease in staffing and supply needs.
We did not see an increase in participation after changes in scheduled testing times due to shelter champion input (Fig 2B).
We replicated the same workflow and protocol modifications in all shelters with few modifications (e.g., incentive structure).
We conducted 1,142 tests before self-swabbing implementation, of which 8 were positive and 4 were false positive (0.70% true positivity). We conducted 562 tests by self-swabbing, of which 4 were positive and 1 was a false positive (0.71% true positivity).
The testing team and shelter champions adopted changes aiming to address deviations to the protocols (e.g., missing confirmatory testing at I&Q or delaying positive case disclosure). We completed all resident confirmatory tests after week three and contacted all BinaxNOWpositive during the event after week six.

Maintenance
Transference of BinaxNOW testing to SFDPH. SFDPH was able to sustain testing in all ten shelters and expanded to all eligible shelters with limited support from BHHI (two nonclinical staff and PrimaryHealth contract) after the end of the study period.

Discussion
In a pilot twice-weekly BinaxNOW antigen testing in congregate homeless shelters in San Francisco, we were able to implement testing, detect SARS-CoV-2 infections, isolate individuals, and identify outbreaks.
The BinaxNOW testing pilot reached approximately one quarter of eligible residents and staff, and one-half of those on-site. Limited reach could have resulted from residents leaving the shelter (residents had to check in every 48 hours to retain their beds). Participation levels varied largely across shelters, and in some shelters over time. This may have been due to shelter culture, the involvement level of shelter champions, or unmeasured characteristics of the residents. In some shelters, testing drew non-shift staff for testing.
Although approximately half of residents and staff tested at least once, there was a low adherence to twice-weekly testing. This may be due to our inability to convey the rationale behind twice weekly testing, discomfort with testing, a recognition of the implications of low community prevalence, or competing priorities. Other incentive structures (e.g., incentives at each event or non-monetary incentives instead of delayed financial rewards) may be more motivating. Twice-weekly cadence may be unrealistic for some shelters. Models that indicate preventive effects depend on twice weekly testing may overestimate the true effectiveness of testing programs.
The effective reach of the BinaxNOW shelter pilot was higher than with prior testing regimens (five shelters once a month with an average uptake of 35% of residents (and no staff) at each event; S. Strieff, personal communication, March 22 nd , 2021). Due to rapid turnaround, we reduced delays in isolation of infectious individuals compared to previous workflows (Table 1). These responses may have contributed to preventing large outbreaks, despite a relatively narrow reach of testing.
Despite mid to low community prevalence (one-week average 30.9 and 7.7 new cases per day per 100,000 during the first and last week of the study, respectively) [30], we detected three outbreaks and one isolated case. The largest had five cases, fewer than prior outbreaks [11,16,31,32]. This may have been a result of our intervention or other changes instituted since early in the pandemic (reduced capacity and improved adherence to mask-wearing) [33]. We detected most cases in asymptomatic participants, highlighting the limitations of symptom screening for isolation of infectious individuals [16]. While BinaxNOW is less sensitive than nucleic acid amplification tests, it appears effective at detecting infectious individuals [27]. Detecting asymptomatic individuals is key to interrupting transmission chains as asymptomatic individuals may account for more than half of all transmission [34].
We detected two of 40 individuals who were BinaxNOW-negative but rRT-PCR-positive. However, one of these was consistent with a prior infection [35], leading to unnecessary isolation, a potential problem with highly sensitive RT-PCR testing. The other individual's test patterns align with early infection, prior to high infectivity. High adherence to twice-weekly testing would be needed to detect these early cases.
We found a relatively high false positive rate, consistent with concerns of using BinaxNOW in low community prevalence and among asymptomatic persons [36]. Three of eight asymptomatic individuals were BinaxNOW-positive/confirmatory-negative, suggesting false Binax-NOW positives. This is costly in terms of transportation and I&Q requirements and may have undermined trust in testing and the healthcare system, raising questions about BinaxNOW testing during periods of low community prevalence. We recommend continuous dialogue and education of participants on BinaxNOW testing strengths and limitations.
As a result of the strong collaboration between academic, public health and homeless service providers, we were able to implement and adapt the pilot, allowing us to minimize errors, increase adoption and reduce resources, which contributed to maintenance. Self-swabbing had no noticeable effect on positivity rates or the ability to detect outbreaks. The SFDPH offered twice-weekly testing in all eligible shelters with little external support after the study, suggesting maintenance of this intervention. We found the web-based test reporting system (PrimaryHealth) to be instrumental to conduct mandatory reporting to local and state officials, and track data. Despite our efforts to streamline staff and resources, BinaxNOW screening testing still requires significant staff and resources. Further implementation studies should evaluate modifications to reduce workflow and staff and consider further task-shifting to nonclinical partners.
The study has several limitations. We focused on implementation of BinaxNOW testing rather than comparing different strategies. Thus, the contribution of twice-weekly testing to preventing and resolving outbreaks is unclear. While infection transmission models suggest that twice-weekly testing is necessary [16][17][18], we tested a lower proportion of eligible participants than recommended by models [18]. However, this may reflect the reality of testing regimens at homeless shelters, where residents do not tend to stay during the days. Those who leave during the day could be at higher risk of contracting SARS-COV-2. We did not study the causes (e.g., low interest, previous COVID-19 infections, testing fatigue) and consequences (e.g., transmission) of low participation. We estimated shelter residents present by headcounts for three weeks; these may not be accurate. We had limited demographic data, which limited our ability to detect differences. In addition, BinaxNOW sensitivity is lower than nucleic acid amplification methods [21], implying that we could have missed infections within the participant population. Furthermore, we conducted our study for six weeks, in one jurisdiction which had implemented aggressive COVID prevention protocols in shelters (including reduced capacity and universal masking) during a period of low to moderate COVID-19 community incidence that preceded vaccine rollouts. Other settings and contexts could have different results. However, the lack of large-scale outbreaks is reassuring.
As the COVID-19 pandemic evolves, the trade-offs of frequent testing in congregate shelters will change. Effectiveness (and cost-effectiveness) of rapid testing for isolation and outbreak prevention should thus be evaluated in the specific setting context. Fluctuations in community transmission affect false positive and negative rates, and determine the probability of infection for individuals. Vaccination campaigns and other interventions may require a change in resource allocation. Vaccinations, as well as previous exposure, can affect the susceptibility of individuals to severe illness and change the population at risk. The rise of new variants has the potential of changing the epidemiology of the disease, as well as the sensitivity of diagnostic tests. While we expect rapid test and response programs to be less cost-effective with high vaccination rates and low community transmission, all these factors should be considered when implementing measures to prevent outbreaks in congregate settings.
Our pilot allowed for detection and isolation of COVID-19 cases among staff and residents. This intervention could be applied to other congregate settings where there is a high turnover of residents, a higher risk of transmission, and challenges to physical distancing [37]. Lessons learned from our rapid test and respond model, and application of an implementation framework matching the cascade of care could be applied to how we approach detection and response for other infectious diseases in this population with additional considerations of how to assign staff and resources to needs. The partnership between an academic group, and representation from city departments that manage public health and the shelters, with additional training and support for a shelter champion could be applied to vaccination or other public health efforts. Low-barrier frequent testing with rapid turnaround is an important public health intervention to improve accessibility and availability of testing and rapid identification and isolation of COVID-19 positive cases in shelters.

S1 Appendix. Demographic description of participant and eligible populations.
(DOCX) S1 File. Census and participant number. Individual shelter participation per event and total census population. This data is shown in Fig 2B. Percentage of tests taken. Individual datapoints shown in Fig 2A. (XLSX) Rachowitz, Scott Walton, Howard Chen, Elisabet Medina, Ronald Thai, Ashley Scarborough, facilitated data availability, coordinated work with multiple stakeholders and gave feedback throughout. Carina Marquez, Genay Pilarowski and Jackie Martinez provided technical advice and support. This project would not have been possible without the help of shelter leadership, especially the shelter champions, and community volunteers.