Comparing frequency of booster vaccination to prevent severe COVID-19 by risk group in the United States – Nature.com

Study population and data

We defined the study population as persons residing in the United States, age 18 years or older, and fully vaccinated (defined as completion of their primary series and 1 or more monovalent booster doses). The epidemiologic data used in the model reflects the timeframe up until approximately September 2022, coinciding with introduction of bivalent COVID-19 vaccines in the United States. Applying publicly available data from the US Centers for Disease Control and Prevention (CDC) COVID-19 surveillance program, we generated age-specific monthly risk estimates of severe COVID-19 (defined as related hospitalization or death)25,26. Age-specific seroprevalence estimates were obtained from the CDC based on the nucleocapsid antibody, suggesting prior infection, and updated to account for cases since the last survey27 (see Appendix, Prior infection and serosurveillance data).

We developed a stochastic, person-level simulation model (microsimulation) of severe COVID-19 cases in the United States. We created hypothetical cohorts of one million persons in each risk group who were fully vaccinated, defined as having completed their primary series and received at least one monovalent mRNA booster dose. The population size (1 million) for each risk group was chosen to broadly represent the geographic scale of a county in the United States (Table2). We modeled the population in 12 key risk groups defined by: i) age: 1849 years, 5064 years, 6574 years, 75+ years; and ii) immune status: immunocompetent, mild immunocompromised status (e.g., low-dose corticosteroids, mild immunosuppressive medications), and moderate/severe immunocompromised status (e.g., hematologic malignancy with active treatment or poor response to vaccines, solid organ or bone marrow transplant, high-dose corticosteroids or other moderate/severely immunosuppressive medications)16 (see Appendix, Model calibration). Upon entry into the simulation, each person was assigned an age, immune status, vaccine status (1 or 2 monovalent mRNA booster doses)28, and prior infection status27. For the age-specific cohorts and the immunocompromised risk group, prior infection status was informed by estimates of seroprevalence (nucleocapsid antibody consistent with prior infection; see Appendix for full methodologic approach)27,29. Prior infection status was necessary to define whether an individual had protection from hybrid immunity (vaccine and prior documented infection) or vaccination alone, given that hybrid immunity has been suggested to provide more robust and durable protection compared to vaccination alone1 (Supplementary TableS1). Each person was assigned a time since their last COVID-19 vaccine or infection (measured in number of months), to account for waning of protection over time. This timing was determined from sampling of publicly available data on time series data of vaccine administration and COVID-19 cases and then tracked over the simulation period (Supplementary Fig.S4).

We simulated a two-year time horizon, which was chosen to allow adequate time for comparison of vaccine strategies (i.e., one year time horizon would not allow estimation of differences from one-time and annual strategies). We assumed a hypothetical fixed population with no aging or demography. The start of the simulation (time 0) coincided with approximately September 2022, alongside introduction of the bivalent vaccine in the United States.

During the simulation, we applied an individual-specific, time-varying probability of SARS-CoV-2 infection and severe COVID-19 for each month time step, informed by the model calibration using COVID-19 surveillance datasets (see Calibration and Validation section). This probability combined a fixed group-specific force of infection term by age and immune status and an individual, time-varying level of protection against SARS-CoV-2 infection and severe COVID-19. An individuals risk of SARS-CoV-2 infection and severe disease changed over time as protection waned. The primary analysis used a static model of infection, meaning we did not account for indirect effects due to vaccination (i.e., reduced transmission due to vaccine-induced protection), although we did test a dynamic transmission model in an alternative analysis (see Scenario Analysis). Each persons level of protection was based on vaccine status (time since last vaccine) and prior infection history (time since last infection, if applicable). This model explicitly accounted for waning of protection against SARS-CoV-2 infection and severe COVID-19 independently based on timing of last vaccination and prior infection, which was estimated from literature1,2,3,5,20,30 (Supplementary TablesS1-S2). We separately modeled individuals as either having vaccine-induced (without prior infection) or hybrid immunity (defined as vaccination with documented prior infection) since literature suggests far higher and more durable protection for hybrid immunity1,20 (Supplementary Fig.S1).

We simulated severe COVID-19 cases, defined as a composite outcome of COVID-19 related hospitalization or COVID-19 related death. The study focused primarily on severe COVID-19 based on a public health priority to reduce hospitalizations and deaths, although we did simulate non-severe COVID-19 cases and subsequent effects on protection and immunity (Supplementary TableS9). All COVID-19 cases (severe and non-severe) reset the time since last COVID-19 case or vaccine. While acknowledging that a certain fraction of COVID-19 cases will result in long COVID, we did not account for long COVID given limited data to inform these estimates. We assumed no reinfections occurred within 90 days of a SARS-CoV-2 infection. Analysis was conducted in R (version 4.2.1).

We simulated three distinct vaccination strategies with booster vaccines for COVID-19, including: i) one-time booster at the start of the simulation (base case); ii) single booster followed by annual boosters (total of 2 doses); and iii) single booster followed by boosters every 6 months (semiannual; total of 4 doses). In September 2022, the available COVID-19 booster vaccine in the United States was the bivalent vaccine (ancestral strain and Omicron subvariants BA.4/5), followed later by a monovalent formulation against Omicron XBB.1.5. Each round of vaccination was administered in the population over a 3-month period. We calibrated the protection and waning of a mRNA booster dose to published data on vaccine effectiveness over time using data from both monovalent and bivalent COVID-19 booster vaccine literature (Supplementary TableS1)1,2,3,5. We modeled the benefit of a booster dose to restore maximal protection against severe COVID-19 prior to waning (Supplementary Fig.S1). Therefore, the impact of additional vaccination conservatively did not increase the absolute protective effectiveness previously achieved, but only restored the lost protection due to waning. This approach to vaccine modeling resulted in estimates of relative vaccine effectiveness similar to published estimates on the bivalent mRNA booster (Supplementary Fig.S3)3. We estimated the waning protective effectiveness of a booster dose by age group and prior infection status over a 24-month period using a linear mixed effects model. We modeled the outcome of protection against severe COVID-19 and infection as the log of 1 minus protective effectiveness, with predictor variables of the log of months since last vaccine dose or COVID-19 illness (whichever was more recent), age group (1849 years, 5064 years, 65+ years), and prior infection status, based on available literature. We modeled two immunocompromised groups, generating age-specific estimates for a mild immunocompromised group (13% lower protection) and moderate or severe immunocompromised group (25% lower protection, incorporating faster waning)2,16,31,32. We assumed that each repeated booster dose would achieve the same level of effectiveness without immune exhaustion, immune imprinting phenomenon, or reduced vaccine effectiveness due to new variants33,34, although we explored this in sensitivity analyses.

The primary study outcome wassevere COVID-19, measured as the annual absolute risk of severe COVID-19 over a 2-year simulation period in each risk group. Each of the boosting strategies was compared to the base case of a one-time booster at the start of the simulation. For each strategy, we estimated the total number of severe COVID-19 cases, absolute annual risk reduction of severe COVID-19 (cases per 100,000 persons), relative risk reduction, and NNT with a specified vaccination frequency to avert one severe COVID-19 case (calculated per person, not vaccine dose).

We calibrated the model to age-specific estimates of severe COVID-19 risk generated from an average over the 6-month period preceding model initialization (March 2022August 2022). For the two immunocompromised populations, we used literature estimates for their age distribution, assuming the same age-specific risk of infection but 2.8-fold higher risk of severe disease given infection25,29,32 (see Appendix Model calibration; see Table2 for severe COVID-19 risk estimates). This calibration yielded a per month, force of infection coefficient specific to each age and immune status on their risk of severe COVID-19, which was multiplied against 1 minus an individuals current level of protection to obtain individual per month probability of severe COVID-19. The probability of SARS-CoV-2 infection (non-severe) was modeled with an additional multiplier and separate estimates on level of protection (see Appendix, Model calibration). For model validation, we performed a comparison of model-predicted outcomes over the first 3 months of the simulation (September 2022- November 2022).

We repeated the primary analysis under different scenarios for emergence of novel variants with immune evasion (Fig.1A), including one scenario with a variant targeted vaccine. Upon circulation of a novel variant, we modeled two different immune evasion scenarios: i) absolute protection from vaccine or hybrid protection against non-severe and severe COVID-19 is reduced by 10%, due to immune evasion; and ii) absolute protection is reduced by 10%, and rate of waning increases by 5%. We did not simulate variants with higher infectiousness or severity. In the scenario with a variant targeted vaccine, we assumed the vaccine restored the protection lost due to the new variant in vaccine-induced immunity and partially restored protection for hybrid immunity. Novel variants were introduced over a 3-month period. A full description of the analysis is available in the Appendix (see Scenario analysis: Novel variants).

We repeated the primary analysis using a dynamic transmission model, which accounted for the indirect effects of vaccination on transmission. This analysis was designed to test the importance of considering transmission dynamics in the analysis. This model departed from the primary microsimulation model based on the following modifications. First, the force of infection term was formulated to be directly related to the number of SARS-CoV-2 infections in the population in the prior time step (week) with age-specific contact matrices35,36. Second, the simulated population included all age groups and unvaccinated individuals. Third, vaccine strategies were applied with imperfect uptake coverage by age- and immune status to reflect current uptake (Supplementary TableS7). Fourth, the model was only calibrated to match observed severe COVID-19 cases at time 0 (Supplementary TableS8). We compared booster vaccination strategies in the following groups to determine the impact of indirect effects of vaccination: i) 75+ years and moderate/severe immunocompromised; ii) 65+ years and mild and moderate/severe immunocompromised; and iii) all groups 18+ years. In all strategies, we applied one-time booster vaccination as the base case intervention to those 18+ years based on expected uptake. Study outcomes were computed among persons assigned to the booster vaccination strategies (i.e., excluding unvaccinated persons, or those who did not receive additional vaccination), to improve comparability to the primary model. A full description of the model specifications is available in the Appendix (see Scenario analysis: Dynamic transmission model).

We conducted sensitivity analyses on the main microsimulation analysis to evaluate the robustness of our findings. First, we repeated the analysis under optimistic or pessimistic assumptions on level of protection (10% lower or higher) from vaccine-induced and hybrid immunity, as well as differential waning of protection (10% lower or higher) (Supplementary TablesS13S16). Second, we repeated the analysis for a lower (0.5x) and higher (2x) incidence of severe COVID-19(Supplementary Tables S17, S18). Third, we performed analyses under the assumption that additional boosters would have lower vaccine effectiveness (i.e., immune exhaustion)(Supplementary Table S25). Fourth, we performed the analysis with higher or lower seroprevalence and an additional analysis with a population of only previously infected persons (i.e., 100% seroprevalence)(Supplementary Tables S19S21). Fifth, we repeated the analysis assuming higher proportion of sub-clinical infections(Supplementary Table S24). Additional details on sensitivity analyses can be found in the Appendix (see Sensitivity analysis).

We generated uncertainty intervals for the primary analysis based on parameter uncertainty in vaccine effectiveness and waning over time, baseline seroprevalence levels, and non-severe infection multipliers (Supplementary TableS5). This interval is generated by simulating the full range of model inputs at baseline, which define the bounds of the interval; the reported point estimate uses the base case assumption of model inputs, so the bounds are expected to be asymmetric relative to the point estimate. Uncertainty intervals are designed to demonstrate uncertainty within a single vaccine strategy under a range of baseline conditions; vaccine strategies should be compared against one another using the same set of assumed baseline model inputs.

This study was not human subjects research given use of publicly available secondary datasets with aggregated estimates that are not identifiable.

Further information on research design is available in theNature Portfolio Reporting Summary linked to this article.

Read more from the original source:

Comparing frequency of booster vaccination to prevent severe COVID-19 by risk group in the United States - Nature.com