Testing and vaccination to reduce the impact of COVID-19 in nursing homes: an agent-based approach

Background Efforts to protect residents in nursing homes involve non-pharmaceutical interventions, testing, and vaccine. We sought to quantify the effect of testing and vaccine strategies on the attack rate, length of the epidemic, and hospitalization. Methods We developed an agent-based model to simulate the dynamics of SARS-CoV-2 transmission among resident and staff agents in a nursing home. Interactions between 172 residents and 170 staff based on data from a nursing home in Los Angeles, CA. Scenarios were simulated assuming different levels of non-pharmaceutical interventions, testing frequencies, and vaccine efficacy to reduce transmission. Results Under the hypothetical scenario of widespread SARS-CoV-2 in the community, 3-day testing frequency minimized the attack rate and the time to eradicate an outbreak. Prioritization of vaccine among staff or staff and residents minimized the cumulative number of infections and hospitalization, particularly in the scenario of high probability of an introduction. Reducing the probability of a viral introduction eased the demand on testing and vaccination rate to decrease infections and hospitalizations. Conclusions Improving frequency of testing from 7-days to 3-days minimized the number of infections and hospitalizations, despite widespread community transmission. Vaccine prioritization of staff provides the best protection strategy when the risk of viral introduction is high. Supplementary Information The online version contains supplementary material available at 10.1186/s12879-022-07385-4.

• Infected agents execute their "Disease progression" sub-model, on which depending on the days since initial exposure, will transition trough the infection states represented S1 according to the rates specified by Table 1 from main manuscript.
• According to the testing schedule, all staff members and one resident per room will be tested for the disease. If an infected staff member is detected, the staff member will be sent back to the community for a period of 15 days and replaced with another staff. If an agent is detected positive, the other residents in the room will be also tested. All detected residents will be sent to one of the isolation rooms until tested negative.
• Agents will move randomly inside the current room, based on proximity and if the agent is infected, the transmission event will follow a binomial distribution P (y = 1|x = 1) = p t , where y i = 1 represents the event that agent i is infected, x j = 1 represents that agent j is infectious, and p t represent the probability of transmission estimated by the equation 1.
• If resident agents transition to hospitalization status, will be moved out of the facility to hospitalization until recovered or death, based on the rates described on Table 1 from main manuscript.
• According to their staff type, staff agents will randomly select a number of residents to contact. If the staff or agent is infected, the transmission event will follow a binomial distribution described by the transmission equation described before.
• Depending on the current simulation hour, the staff change will end their shift and go back to the community, next staff shift will start again. For each staff that returns from the community, the probability of introducing the disease is described by a Bernoulli distribution with p = probability of introduction. Stochasticity in our model plays an important role. Global parameters such as the transmission, introduction, vaccine efficacy and distribution, are randomly sampled from a parameter sample space at the initialization of the simulation. Behaviours of the agents such as movement are completely random, restricted by the nursing home floor plans. Other stochastic behaviours include the order of staff-resident contacts, number of introductions to the nursing facility, testing, and transmission of the disease. Interactions between the agents are defined based on spatial proximity and staff-to-resident contacts, which will influence the diseases spread. An example of a adaptive behaviour in our model is the staff-to-resident interactions, we assumed that all residents will be treated equally and the choice of which resident to contact in any given hour of the staff will be based on the number of previous interactions each resident agent has had during that turn, prioritizing the residents with least amount of contacts. Another adaptive behaviour in our model is the isolation of the infected agents, once detected the isolation of infectious agents will technically limits the disease spread. For every simulation run, individual attributes such as the staff type, staff turn and adherence to PPE use

Dynamic Transmission Model
Resident and staff agents in our proposed model are represented in epidemiological classes of susceptible but not yet exposed to the disease (S), non-infectious exposed individuals incubating the disease whose infection is currently non-detectable by testing (E), infectious individuals with detectable disease who do not yet exhibit clinical symptoms of illness (I a ), infectious individuals exhibiting symptoms of illness (I s ), individuals that have recovered and can no longer infect others (R), symptomatic individuals requiring hospitalization (H), and individuals that succumbed to the disease (D) (Figure 1). Transition parameters are described in Table 1 of main manuscript. Figure S1: Epidemiological classes of the transmission model.