key: cord-1012993-mcw3ir3a authors: Are, E. B.; Song, Y.; Stockdale, J. E.; Tupper, P.; Colijn, C. title: COVID-19 endgame: from pandemic to endemic? Vaccination, reopening and evolution in a well-vaccinated population date: 2021-12-19 journal: nan DOI: 10.1101/2021.12.18.21268002 sha: 140fb1be3150bb34255c8970ccad39127fccc894 doc_id: 1012993 cord_uid: mcw3ir3a COVID-19 remains a major public health concern, with large resurgences even where there has been widespread uptake of vaccines. Waning immunity and the emergence of new variants will shape the long-term burden and dynamics of COVID-19. We explore the transition to the endemic state, and the endemic incidence, using a combination of modelling approaches. We compare gradual and rapid reopening and reopening at different vaccination levels. We examine how the eventual endemic state depends on the duration of immunity, the rate of importations, the efficacy of vaccines and the transmissibility. These depend on the evolution of the virus, which continues to undergo selection. Slower reopening leads to a lower peak level of incidence and fewer overall infections: as much as a 60% lower peak and a 10% lower total in some illustrative simulations; under realistic parameters, reopening when 70% of the population is vaccinated leads to a large resurgence in cases. The long-term endemic behaviour may stabilize as late as January 2023, with further waves of high incidence occurring depending on the transmissibility of the prevalent variant, duration of immunity, and antigenic drift. We find that long term endemic levels are not necessarily lower than current pandemic levels: in a population of 100,000 with representative parameter settings (Reproduction number 5, 1-year duration of immunity, vaccine efficacy at 80% and importations at 3 cases per 100K per day) there are over 100 daily incident cases in the model. The consequent burden on health care systems depends on the severity of infection in immunized or previously infected individuals. COVID-19 is still spreading rapidly in many countries across the globe. There are indications that the disease will eventually become endemic rather than be eliminated. Natural questions to ask are: how will factors such as vaccination coverage, vaccine efficacy, duration of immunity and disease importation interplay to determine how and when endemic mode will be reached, and how can the transition happen without major resurgence of cases? Despite the widespread use of highly efficacious vaccines globally, vaccines alone have failed to control transmission in many countries. Therefore physical distancing and other non-pharmaceutical interventions (NPIs) are still widely used to control the spread of COVID-19. These restrictions often come at a cost to the economy [1], and individuals' physical and mental well-being [2, 3] . Previous studies that have investigated the impact, on COVID-19 cases, of public health measure relaxation, all agree that some level of restrictions will still be required to keep cases under control [4, 5, 6]. Since then many jurisdictions have lifted NPIs and later re-introduced them when cases surged. But at some point in the near future, it is likely they will wish to implement some level of further reopening once again. Jurisdictions will need to determine the correct level and appropriate speed of reopening to sufficiently prevent negative outcomes such as cases, hospitaliza-Rationale for R NPI : Estimates of the reproductive number for COVID-19 in the absence of social distancing restrictions range from 2 to 4 for non-VOC SARS-CoV-2 [15] . The delta variant has a higher transmission rate than previously-predominant strains, which acts to increase R NPI . It has also been reported to show some potential for immune escape [13] . However, we model the maintenance of symptomatic testing and contact tracing, which can reduce R NPI by 1/3 if done rapidly, optimally, and with the capacity to scale up as cases rise [16] . We therefore explore relaxation from R NPI = 2.2 to R NPI = 2.5, both in the rapid and gradual reopening scenarios. In supplementary analyses, we explore relaxation to a wider range of R NPI values, ranging from 2.4 to 5. Note that in British Columbia, early estimates of R 0 for the original COVID-19 virus were approximately 3 [17] ; with the Alpha and Delta variants both estimated to increase the transmissibility by approximately 100%, this would place Delta (currently predominant) at an R 0 of over 6 [18, 19] . However, a range of measures are in place at the time of writing, including testing, indoor mask use and vaccine mandates, workplace screening and other measures, motivating a lower R NPI . To have a symptomatic case after vaccination, the vaccine has to both fail to prevent disease and symptoms, so efficacy against symptomatic infection is v d = 1 − (1 − v e )(1 − v p ). We take our baseline values from studies on the Pfizer vaccine, giving v d = 95% protective against symptomatic infection [20] and v e = 80% protective against infection [21] , implying a value of v p = 75%. If the vaccine fails to protect against infection in an individual, but does prevent symptoms, in our model the individual is assumed to contract the virus and transmit to others at the same rate as an unvaccinated individual. It is likely that those who are vaccinated but infected anyway are less infectious due to lower viral loads than those who are infected without vaccination, and it is likely that they would not have symptoms. In a framework where testing is driven by symptoms and those with symptoms are encour-5 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint aged to isolate, asymptomatic individuals will not know they are ill and will likely remain circulating and infectious for longer than those who develop symptoms. Thus, longer duration of infectiousness (vaccinated-but-infected individuals may transmit more because they do not have symptoms and therefore do not isolate), and lower per-unit-time transmission (due to a reduced viral load) act in opposing directions. In our model we assume that these effects balance out. Vaccine acceptance: We model age-based vaccine uptake according to data from the vaccine rollout by age in BC [22] . This includes highest uptake in older age categories (98% in those 80+), and lowest uptake in younger age categories (30% in those [12] [13] [14] [15] [16] [17] [18] [19] . Vaccine hesitancy is modelled as around 10% lower in essential worker categories. A full description of the model age-based vaccine uptake is included in Supplementary Table S1 . Vaccine rollout: Following BC vaccine rollout strategy [22] , we vaccinate those 80+ and in long term care (LTC) settings (not modelled explicitly, but hospitalization rates are adjusted to reflect protection in LTC [5]) first, followed by younger age groups 12+. To match the observed age-based trajectory [22] (whilst taking into account that we do not model the two-dose vaccine regimen), we model a rapid uptake in vaccines during May-June 2021 in which 2% of each age group are vaccinated per day up to uptake levels observed during that time (see Supplementary Table S1 ). This is followed by a slower period of 0.2% per day in July-Aug and then 0.22% per day from September onwards, until the overall uptake levels described in in Supplementary Table S1 are reached. Vaccination of those aged 12-19 does not begin until July 3rd 2021. The full age-based rollout is shown in Supplementary Figure S10. We validate the age and contact structured model by matching the model predicted case counts (including vaccine uptake and rollout as detailed above) to reported cases by age in BC from January to November 2021 (Figure 1 ). . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. where N = S + V + E + I + R. The prevalence (I * ) at endemic steady state is obtained analytically and analysed as a 8 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint function of various parameters representing the aforementioned factors that will determine the number of active cases when the disease becomes endemic. For simplicity, we assume that c(t) = 1 at endemic state (no physical distancing measures, but we explore a range of transmission parameters) and that the population is constant over time. The full analytic solution is provided in the Supporting Information. Model validation involves matching COVID-19 reported cases, from February to November 23, 2021, to model outputs. The vaccination rate was set such that the vaccination coverage in the model largely resembles vaccination uptake in BC during that period. The model fit to data is shown in Figure 4 . We explore the severity and risk breakdown of gradually reopening over a 300 day period compared with a near-instantaneous reopening, using the age and contact structured model ( Figure 2 ). We find that under our baseline vaccine efficacy assumptions, even after most of the rollout is complete, we will not be in a position to reopen without seeing a rise in cases. This is consistent with the fact that the current decline in cases (approximately 2% per day) is slow, leaving little leeway for increasing transmission without moving from a decline to a rise. From the date of reopening (Dec 2021) onwards, nearinstantaneous reopening leads to 500K cumulative reported cases and 19K hospitalizations over 900 days while gradual reopening leads to 450K cumulative cases and 17K hospitalizations, in the BC population of 5 million people. Further scenarios are explored in Supplementary Figures S4, S5 and S6, where we consider gradual vs rapid reopening from R NPI = 2.2 to R NPI = 2.4, 2.6 and 3.0, respectively. In all scenarios considered, there are fewer cumulative cases and hospitalizations under gradual reopening than under rapid reopening. 9 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. We considered reopening at two levels of vaccination coverage -BC's 90% coverage and a counterfactual 70% scenario. Figure 3 shows the simulated outcome if we relax measures to R NPI = 2.5 for the two scenarios, according to the age and contact based rollout. Reopening "fully" to R NPI = 2. We compare the age and contact structured model's results to those obtained from a theoretical (SIR) model of herd immunity. Figure S1 illustrates that the age and contact structured model's prediction for the fraction of the population protected at the herd immunity threshold is similar to the theoretical prediction from the simple model. We can therefore estimate whether a given level of immunity obtained through vaccination is sufficient to stop the spread of COVID-19 using the classic relationship in SIR models be- . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint tween minimum herd immunity fraction f m and reproductive number R 0 : f m = 1 − 1/R 0 . Consider this simple theoretical example: in a jurisdiction where 20% of the population declines the vaccine, 10% are not eligible and we have a vaccine that is 80% effective against infection, the fraction of the population that is immune from vaccination alone is The R * 0 to which that fraction confers herd immunity, Accounting for approximately 5% of the population having had COVID-19 but some overlap between past infection and vaccination, f m ≈ 60%, with corresponding R * 0 of 2.5. Higher R 0 values lead to rises in cases. Given the motivation of similar endemic behaviour in the age and contact structured model and the SIR model above, we continue to explore several long-term scenarios using the SVEIRS model, which includes waning immunity. We first consider reopening to various R NPI values, whilst also allowing for importation of infected cases. Our simpler model results suggest that there may be multiple waves of COVID-19 cases before it eventually becomes endemic. The frequency and peaks of the waves will depend on the duration of immunity and whether or not the vaccination campaign will continue to be supplemented with booster doses. When R NPI = 5 or greater, cases rebound very quickly to cause another major wave. In contrast, if R NPI is below 4, reopening will not lead to a major wave before becoming endemic. This is under the assumption that booster doses will be used to maintain relatively high population immunity (Figure 4 A). Furthermore, we study several immunity waning regimes (Figure 4 B) . The endemic state is sensitive to the duration of acquired immunity, even under continual boosting after immunity wanes. Reopening to R NPI = 3.5 where immunity lasts for 1 year will lead to gradual resurgence of cases, with high endemic incidence close to 40 reported cases per 100K per day. The picture becomes more optimistic as immunity lasts longer (Figure 4 B), with endemic incidence closer to 5 reported cases per 100K per day under 2-3 year immunity. However, if booster doses are suspended and immunity wanes, the projections become very pessimistic (See Figure S2 in the Supporting Information). This will be compounded if high transmission and immune escape variants continue to emerge. We compare the impact on COVID-19 dynamics of gradual changes (or small mutations) 12 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. in the virus that make vaccines less effective against them over time, compared to more abrupt mutation(s) that reduce vaccine efficacy more rapidly. Borrowing terminology from influenza viruses, we term these "drift" and "shift", though the biological mechanisms will differ. One rationale for considering lower efficacy is the continued emergence and spread of VOCs that may undermine vaccination as a COVID-19 control strategy. Omicron variant evades immunity is unclear, but it shows reduced antibody neutralization [28, 29]. We model "drift" and "shift" by reducing vaccine efficacy ν e gradually from 80% to 40% over a 500-day period, and a sudden (all-at-once) change, respectively. We 13 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint find that a sudden shift leads to a worse outcome in the model, with a steep rise and fall before the system settles to endemic equilibrium (Figure 4 C) . We explore the impact of four endemicity-determining factors on the endemic incidence: reproduction number, immunity duration, vaccine efficacy and importation rate (Figure (5) A to F). The endemic incidence is sensitive to all of these unknown factors, but is most sensitive to the combination of the underlying transmission R NPI , vaccine efficacy and the duration of immunity. The model's endemic incidence is not always markedly lower than peak incidence levels in the pandemic to date (approximately 20 per 100K per day). High endemic levels occur if immunity wanes rapidly (in under 1.5 years), if R NPI for the combination of virus and long-term measures is above 3 if there are over 6 imported infections per 100K per day, if efficacy is low and for various combinations. We model the true incidence; reported incidence would be lower, and would depend on the surveillance system that is in place and on the extent to which infection caused symptoms and severe disease. We note that when R NPI is relatively low and vaccine coverage is substantially high, our model predicts no incident cases without importations, and in this sense it is an optimistic baseline from which to explore. In practice, heterogeneity in the population, introductions from animal reservoirs, continued viral evolution and other factors not included would likely mean that instead there would be some very low level of endemic incidence at our baseline parameters. Our results suggest that COVID-19 cases will rebound after restrictions are lifted completely under current vaccination rollout plans and vaccine efficacy. In all of our models, children, adults who decline the vaccine, and adults for whom the vaccine did not prevent infection are numerous enough that the pandemic can unfold among them once restrictions are lifted. This occurs even under optimistic assumptions that immunity is 14 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint if testing rates are high. We assumed constant ascertainment over time, but ascertainment rates can change rapidly depending on testing criteria, test-seeking behaviour, and likelihood of symptoms and severe disease. The endemic prevalence of infection will determine the endemic demand for hospital and acute care resources, though both ascertainment and the relationship between infections and hospitalizations may change. Eventually, nearly all infections will occur in those who were either immunized or previously exposed, and with B-cell mediated immunological memory that is long-lasting, it is to be hoped that 100 per 100K incidence (Figure 5 A) will not present a burden to the health care system so strong as to require widespread NPI measures. However, throughout the pandemic in BC to date, reported COVID-19 cases have been hospitalized at a relatively constant rate around 9% [36]. This rate has been largely unaffected by vaccination. Early observations suggest that disease-blocking immunity wanes more slowly than infection-blocking immunity [35, 37] . If this is the case, we can expect the rate at which cases are hospitalized to decrease at endemic state. However, if the endemic incidence is high (over 30 incident infections per 100K per day), even a large reduction in overall severity (such as an 80% reduction) would leave on average just under 30 daily hospitalizations. Current conditions suggest that this would 17 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint place a burden on the health care system, particularly if it were enhanced by seasonal variation, and if capacity were impacted by other seasonal infections such as influenza. Overall, the virus' evolution and the nature of waning immunity will shape the relationships between infections and reported cases, and between infections and hospitalizations/health care burden. If immunity against infection wanes quickly while immunity against disease lasts longer, and testing criteria are largely symptom-based, then reported [1] Atalan A. Is the lockdown important to prevent the COVID-19 pandemic? Effects on psychology, environment and economy-perspective. Annals of medicine and 18 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The prevalence at endemic equilibrium is obtained analytically by solving equation (1) at equilibrium. where: • H = ((1/4(R NPI γ) 2 σ 2 + (Fγ + (µ + σ)(−1/2µ + f ))σ(R NPI γ) + 1/4E)w + 1/4(R NPI γ) 2 µσ 2 + (µ + σ)(−1/2µ + f )(µ + γ)σ(R NPI γ) + 1/4µE) Equation (1) has no disease free equilibrium when f > 0, since cases are continuously introduced into the population. When f = 0, the disease free and endemic equilibrium points exist and their stability depends on R NPI . The reproduction number in the absence of vaccination is R 0 = βσ (µ+γ)(µ+σ) , with vaccination and waning R e = βσ(µ+w) (µ+γ)(µ+σ)(µ+νv e +w) We compare herd immunity threshold calculated from the age and contact structured model and a simple SIR model, and we find good agreement ( Figure S1 ). Also, we explore scenarios where booster doses are not given after 70% of the population are immune to infection due to vaccination. The results are shown in Figure S2. 1 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. decline. This is the so-called "herd immunity" fraction. The theoretical result (blue) is simply 1 − 1/R. The model result is obtained by running a simulation at the given R NPI , as always defined in the absence of vaccination, detecting when infections begin to decline, and obtaining the portion of the population either infected or successfully vaccinated at that time. 2 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. 6 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Figure S6 : Comparison of the impact of gradual vs rapid reopening, from R NPI = 2.2 to R NPI = 3.0, on case and hospitalization rates. Gradual reopening is modelled as occurring over a 300 day win- 7 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint 8 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Table S1 : Specifics of the age-based vaccine rollout in the age and contact structured model. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint 12 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted December 19, 2021. ; https://doi.org/10.1101/2021.12.18.21268002 doi: medRxiv preprint COVID-19 epidemic in the Brazilian state of Amazonas was driven by long-term persistence of endemic SARS-CoV-2 lineages and the recent emergence of the new Variant of Concern P. 1 New SARS-CoV-2 variants-clinical, public health, and vaccine implications Effectiveness of COVID-19 vaccines against the B. 1.617. 2 variant. medRxiv Modelinformed COVID-19 vaccine prioritization strategies by age and serostatus Public health measures and the reproduction number of SARS-CoV-2 Fundamental limitations of contact tracing for COVID-19. FACETS Quantifying the impact of COVID-19 control measures using a Bayesian model of physical distancing The Delta Variant: What Scientists Know. The New York Times SARS-CoV-2 Variant Classifications and Definitions Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine