Clinical Outcomes, Costs, and Cost-effectiveness of Strategies for Adults Experiencing Sheltered Homelessness During the COVID-19 Pandemic

Key Points Question What are the projected clinical outcomes and costs associated with strategies for reducing severe acute respiratory syndrome coronavirus 2 infections among people experiencing sheltered homelessness? Findings In this decision analytic model, daily symptom screening with polymerase chain reaction (PCR) testing of individuals who had positive symptom screening paired with nonhospital care site management of people with mild to moderate coronavirus disease 2019 (COVID-19) was associated with a substantial decrease in infections and lowered costs over 4 months compared with no intervention across a wide range of epidemic scenarios. In a surging epidemic, adding periodic universal PCR testing to symptom screening and nonhospital care site management was associated with improved clinical outcomes at modestly increased costs. Meaning In this study, daily symptom screening with PCR testing of individuals who had positive symptom screening and use of alternative care sites for COVID-19 management among individuals experiencing sheltered homelessness were associated with substantially reduced new cases and costs compared with other strategies.


INTRODUCTION
In this supplemental appendix, we provide additional model inputs, results from one-way and multi-way sensitivity analyses, and scaled results for a cohort of 1,000 adults experiencing sheltered homelessness. In addition, we provide details on the Clinical and Economic Analysis of COVID-19 interventions (CEACOV) model and management strategies for people experiencing sheltered homelessness.

SUPPLEMENTAL METHODS
Health States CEACOV simulates individuals transitioning between the states of susceptibility to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), infection with SARS-CoV-2 and coronavirus disease 2019  illness, recovery, or death. Susceptible individuals face a daily probability of acquiring infection. After acquiring SARS-CoV-2 infection, individuals may progress through the following health states (eFigure1): • Pre-infectious latency • Asymptomatic infection • Mild/moderate illness: symptomatic • Severe illness: dyspnea and/or hypoxemia ideally managed in a hospital with standard supplemental oxygen but not requiring intensive care unit (ICU) • Critical illness: ideally managed in an ICU with high-flow supplemental oxygen, non-invasive positive pressure ventilation, or invasive mechanical ventilation • Recuperation: for those recuperating from critical illness and improving while remaining in the hospital or other health care facility • Recovered Individuals in the asymptomatic infection, mild/moderate illness, and severe illness states can transition directly to the recovered state. Individuals in the critical illness state can eventually die, or transition to the recuperation state and then to the recovered state. The recovered state is an absorbing state, and recovered individuals are assumed to have immunity to SARS-CoV-2 over the model time horizon.

Natural History Paths
After being infected with SARS-CoV-2, a susceptible individual first transitions to the pre-infectious latency stage. Then, the individual has an age-dependent probability of progressing along one of four "paths," culminating in either asymptomatic infection, mild/moderate illness, severe illness, or critical illness. Before reaching a more advanced illness state, individuals first transition through intermediate states (e.g., those destined for severe illness must first pass through the asymptomatic infection state and the mild/moderate illness state, eFigure1).

Transmission
The basic reproduction number (R0) is defined as the daily rate at which an infected individual contacts susceptible individuals and infects them in a fully susceptible cohort, multiplied by the duration of infectivity: The Nominal Transmission Rate is defined as R0 in a fully susceptible cohort divided by the average duration of infectivity (D). Equivalently, it can be defined as a function of the average number of contacts per day an infected individual has with susceptible people in a fully susceptible cohort (K) multiplied by the probability of transmission per contact between an infectious individual and a susceptible person (b), therefore the contact rates and probability of transmission per contact are implicitly included in the nominal transmission rate. R0 estimates the expected number of secondary cases produced by an infected individual in a fully susceptible cohort. Once the epidemic is underway and a subset of the population is infected, the effective reproduction number (Re) measures the number of secondary cases produced by an infected individual in a progressed epidemic. This Nominal Transmission Rate captures the ratio (rather than the magnitude) of daily infectivity across different infection states (e.g., asymptomatic to mild/moderate, or mild/moderate to severe). Infected individuals do not transmit while they are in the latent state or in the recovered state. Patients in other infected states can transmit SARS-CoV-2 to susceptible individuals. The effective magnitude of the transmission rate changes over time as social interventions alter the number of contacts (K) and infectivity (b), and subsequently, the effective reproduction number (Re); thus, the magnitude is adjusted using the transmission multiplier.Transmission multipliers are setting-specific, time-dependent adjusting factors, roughly accounting for population density and interventions that can alter the number of contacts (e.g., social distancing) and the infectivity per contact (e.g., masking) in the setting being modeled. In this analysis, the transmission multipliers were calibrated to represent a surging, a growing, and a slowing epidemic. Given the timedependent uptake of such policies, time-varying transmission multipliers may be used to alter the effective transmission rate (Reff) over time. Reff determines the effective transmission rate of infected individuals on each day of simulation and is equivalent to the effective reproduction number (Re) divided by the duration of infectivity. Therefore, the effective transmission rate (Reff) in CEACOV and the effective reproduction number (Re) are directly related.
In this analysis, we assumed that all susceptible persons have an equal probability of contacting infected individuals and acquiring the virus (i.e., homogenous mixing). As the epidemic grows, the number of susceptible persons declines. Thus, not all of the daily contacts of infected individuals will be with susceptible persons. The daily infection rate for a susceptible person is equal to the sum of transmission rates from all infected persons across all infection states divided by the cohort size. This leads to an expected daily number of infections equal to the number of susceptible people multiplied by the infection rate on that day.
It is worth noting that the number of secondary cases each infected individual produces is defined by the effective reproduction number (Re). While the assumption on mixing (homogenous vs. heterogenous) affects 'who' an infected individual transmits to, the average number of new infections caused by a given infected individual is determined by the effective reproduction number.
Under the homogenous mixing assumption, an infected individual in Shelter A will uniformly infect susceptibles in other shelters; similarly, an infected individual in Shelter B will uniformly infect susceptibles in other shelters, including Shelter A. However, under the heterogeneous mixing assumption, an infected individual in Shelter A would most likely infect susceptibles in the same shelter, and an infected individual in Shelter B would similarly infect susceptibles in Shelter B. We assume that all shelters are roughly similar and that the assumption of homogenous mixing for this population is adequate.

Calibration and Validation of CEACOV
We initiated the model with a cohort of 1 million simulated persons who are meant to represent the 6.9 million population of Massachusetts (MA) in 2020. Each person's age category was drawn at model start based on MA age distribution data. 1 Prevalence of COVID-19 at model start was set to 0.14% to represent the approximate prevalence of COVID-19 in mid-March in MA. 2 We tracked the number of people in each health state over a 45-day horizon.
We calibrated the transmission multiplier (see above) to the COVID-19 epidemic in MA from mid-March to mid-April (first 30 days) and used the data from the remaining 15 days to validate the model. We assumed the reported number of COVID-19-attributable deaths would be close to the actual number of deaths. Hence, the number of reported COVID-19-attributable deaths was our main calibration target. We removed 59% of deaths to account for the deaths occurring in long term care facilities (LTCF) and not in the community. To ensure good model fit, we checked the mean absolute percentage error (MAPE)

7
*To obtain the initial illness distribution, we seeded the model with 0.01% of the starting cohort infected with the SARS-CoV-2 virus. We simulated the model for 10 days until the 8 prevalence of SARS-CoV-2 in the cohort reached 2.2%. We used the illness distribution on Day 10 from this initialization run to inform our initial illness distribution in all model runs.