This study uses a previously published dynamical model (Additional file
1: Fig. S1) [
17‐
19] designed to examine COVID-19 burden in Rhode Island (RI), Massachusetts, Connecticut, and Pennsylvania. The model incorporates key aspects of SARS-CoV-2 dynamics, including asymptomatic transmission, vaccination, and age-specific (10-year age bands) risks of infection and severe outcomes. The model also accounts for underreporting of disease burden through a reporting rate, allowing the infected classes to represent reported symptomatic cases. Eleven daily data streams were used for the Bayesian model fit, which showed consistency in its inference of clinical and epidemiological transition parameters across four different states and at different phases of data collection [
17,
18]. We adapted the previously published model, and based on its previous fit for RI [
17,
18], we calibrated the transmission rate for the US and produced simulations for the present study.
Data sources
Daily data for cumulative cases, hospitalizations, and deaths are publicly available from the RI Department of Health (DOH) [
20]. From these, we calculated daily incident cases, hospital admissions, and deaths. To apply results from a model calibrated to RI to the entire US, we collected weekly incident cases and deaths for the US, which are publicly available from the US Centers for Disease Control and Prevention (CDC) [
1], and weekly hospitalization data, which are publicly available from the Department of Health and Human Services (HHS) [
21]. These data are available by state and were summed to represent national burden. For both RI and the US, we collected data between March 1, 2020, and November 30, 2022.
COVID-19 vaccination data for RI are publicly available from RI DOH and CDC. Weekly counts of people who completed the doses of a primary series (two doses of Pfizer or Moderna vaccines or single dose of Johnson and Johnson vaccine) from RI are available from RI DOH [
20] from December 13, 2020, to July 30, 2022. The data afterwards are available from CDC [
22] through November 30, 2022. Both data sources contain age-specific totals of individuals receiving a COVID-19 vaccine (either an initial series or booster). The age groups from CDC were as follows: 0–4, 5–12, 13–17, 18–64, 65 + , and these were adjusted to our 10-year age bands assuming a uniform distribution of coverage within each CDC age band.
Monthly influenza vaccination coverage from the 2010–2011 season to the 2020–2021 season was collected from the CDC [
23], provided by the National Immunization Survey-Flu (NIS-Flu) and the Behavioral Risk Factor Surveillance System (BRFSS). Coverage was estimated as the proportion of the US population that received an influenza vaccine based on telephone surveys conducted from October to May in each season. Coverage among children (6 months–17 years old) is reported by NIS-Flu, and coverage among adults is reported by BRFSS.
The national and jurisdictional cumulative counts of delivered and administered therapeutics including nirmatrelvir/ritonavir (Paxlovid), molnupiravir (Lagevrio), and tixagevimab/cilgavimab (Evusheld) are available from HHS [
24]. We defined treatment coverage in the current season as the coverage of Paxlovid because it is the most commonly used COVID-19 therapeutic in the US. Current treatment coverage is 13.7%, calculated from total administered doses of Paxlovid (6,279,116) and total COVID-19 cases (45,666,906) between January 1, 2022, and November 30, 2022, in the US.
Model adaptation
We introduced one vaccination class to the previously developed model. The vaccine provides protection against infection: after receiving a vaccine, individuals are moved to this class (“Vac” in Additional file
1: Fig. S1). The vaccine only protects individuals who are not infected, including individuals who recently recovered or were discharged from hospital, and susceptible individuals. While we allow individuals who are infected and not yet symptomatic to get vaccinated, including individuals who are presymptomatic or asymptomatic, they will continue a normal course of infection. We assumed that mRNA vaccines have around 95% efficacy against infection for age > 18 based on previous studies [
25,
26]. Thus, we assume 95% vaccine efficacy against infection for all ages. From December 13, 2020, to November 30, 2022, the vaccination rate was determined by the weekly number of people who completed a primary series (two doses from Pfizer or Moderna, or one dose from Johnson & Johnson) reported by RI DOH or CDC.
To account for treatment of symptomatic individuals, additional transitions were added between classes. During the 6-day infection period, symptomatic individuals who receive treatment will do so during the first 3 days of infection. As a rebounding effect (the relapse of symptoms or viral load within a short time after treatment) of the therapeutics has been reported [
27‐
32], we consider three treatment outcomes: (
i) successful treatment, where patients recover completely after treatment and move to the recovered class after day 3 of infection; (
ii) treatment failure, which includes rebound, where patients are still symptomatic and infectious after failed treatment; and (
iii) hospitalization, where patients are hospitalized after failed treatment. Treatment efficacy is modeled by two parameters: probability of treatment failure and hospitalization fraction given treatment failure. We set the probability of treatment failure to be 5.9% based on population studies on the rebounding effect of Paxlovid and Molnupiravir [
32,
33] and the probability of hospitalization after failed treatment to an 88% risk reduction of hospitalization when compared to no treatment [
12,
34]. Other values for these parameters are discussed in the “
Potential risks of vaccine or treatment failure” section.
We introduced waning of infection-induced and vaccine-induced immunity starting at the beginning of the Delta period (June 2021). A number of studies have been published comparing times to reinfection among individuals who have or have not been previously vaccinated or infected [
35‐
42] many of which align with an estimated 40% to 70% becoming susceptible to reinfection after 1 year (Table S
1). Comparison among studies and inference on absolute measures of protection are challenging as the studies (
i) used different controls groups and (
ii) used cohorts that were exposed to infection during different periods of the epidemic. Nevertheless, the rates of reinfection in these studies appear to be consistent with average rates of immune waning between 365 days (1 year) and 730 days (2 years), so we considered both durations as well as 548 days (1.5 years) in analyses. Exponentially distributed waning periods with mean durations of 365 days, 548 days, and 730 days imply that 63%, 49%, and 39% of individuals lose immunity within 1 year of infection, respectively.
Since we project disease burden for a long duration (8 years), we also incorporated natural births, deaths, and aging in the model. The national natural birth and death rates were based on recent CDC reports [
43,
44].
Parameters pertaining to clinical progression, such as probability of symptomatic infection, hospitalization, and death, were previously fit for RI using data through June 6, 2021 [
18]; we used these values in the model for this time period. These, as well as parameters for vaccine and treatment efficacies [
45], were updated based on published literature for the Delta period (June 7, 2021–December 20, 2021) and Omicron period (December 21, 2021 onward) [
46‐
51]. Details are presented in Table S
2.
Between June 7, 2021, and November 30, 2022, we calibrated the time-varying transmission rate to fit the daily hospitalized cases in the US such that at least 85% of the days within this period produced modeled hospitalization values within 10% of the observed hospitalization incidence. Across the three transmission assumptions, the weeks falling outside this window occurred in June, July, and August of 2021 and May, June, and November of 2022. We chose to fit hospitalization data because it is less affected by underreporting. We then applied the average transmission rate parameter in the Omicron period (December, 21, 2021, to November, 30, 2022) to December 1, 2022, through February 28, 2023, multiplied by a factor such that the observed deaths between March 1, 2022, and February 28, 2023, would be approximately 170,000, reflecting the probable number of deaths during this period (162,136 between March 2, 2022, and January 31, 2023, as of February 7, 2023) [
1]. To apply results from a RI-calibrated model to the entire US, we scaled model outputs using the ratio of the total reported symptomatic cases between the US and RI during each of the four variant periods: wildtype (March 1, 2020–March 30, 2021), Alpha (March 31, 2021–June 6, 2021), Delta (June 7, 2021–December 20, 2021), Omicron (December 21, 2021 onward). We used this method rather than scaling by population size because RI contributes a substantially higher number of cases compared to population. The US population is 330 times larger than the RI population, and US case numbers were between 220 and 240 times higher than RI case numbers; this is likely due to higher population density and higher-than-average reporting in RI. This method is also equivalent to comparing population-based incidence for RI and the US. The scaled-up results match the trends observed in the US (Additional file
1: Fig. S2).
Using current coverage of vaccination and treatment, we established a “status quo” scenario, representing COVID-19 projections if vaccination and treatment use remain unchanged. We set the status quo treatment coverage at 13.7% and the initial vaccination coverage to 49% based on the allocation of Paxlovid and the administered number of primary series and boosters during the third year of the epidemic (2022–2023) (detailed assumption of vaccine coverage are in Additional file
1: S1 Text, Table S3, and Fig. S3). Current statewide levels of treatment and vaccination are available in Additional file
1: Fig. S4 and Additional file
1: Fig. S5.
Projecting future disease burden
We assumed future transmission would be equal to the average over the observed Omicron period (December 21, 2021–November 30, 2022) and adjusted the transmission levels in the model for future projections based on influenza dynamics in the US. Seasonal wintertime forcing in transmissibility was introduced starting in the 2023–2024 season, where transmission increases by up to 20% in a sinusoidal curve between October and February, estimated from influenza transmission during winter in the northeastern US [
52]. We assumed monthly COVID-19 vaccination patterns would resemble those of influenza vaccinations. Age-specific proportions of influenza vaccines administered during each month were linearly interpolated to generate age-specific proportions of administered COVID-19 vaccines administered each week in a year (Details in Additional file
1: S1 Text). The annual age-specific trend of vaccination is shown in Additional file
1: Fig. S6.
From the status quo, we generated model projections from March 1, 2023, to February 28, 2033, and recorded cumulative reported cases, hospital admissions, and deaths. We then applied changes in vaccine coverage and treatment coverage to the model to estimate burden under different strategies of vaccination and/or treatment. Because the epidemic showed transient dynamics over the first two seasons (2023, 2024) and settled down to stationary behavior starting in 2025, we used averaged annual cases, hospital admissions, and deaths over 2025–2033 as the primary measure for comparison across strategies. We repeated analyses for two additional transmissibility scenarios, optimistic and pessimistic, referring to transmissibility equal to half and double, respectively, of that observed during the 2022–2023 season.