Evaluation of Waning of SARS-CoV-2 Vaccine–Induced Immunity

Key Points Question How does the effectiveness of COVID-19 vaccines against laboratory-confirmed Omicron infection and symptomatic disease change at different times from last dose administration and number of doses, and how does this compare with previously circulating SARS-CoV-2 variants and subvariants? Findings This systematic review and meta-analysis of secondary data from 40 studies found that the estimated vaccine effectiveness against both laboratory-confirmed Omicron infection and symptomatic disease was lower than 20% at 6 months from the administration of the primary vaccination cycle and less than 30% at 9 months from the administration of a booster dose. Compared with the Delta variant, a more prominent and quicker waning of protection was found. Meaning These findings suggest that the effectiveness of COVID-19 vaccines against Omicron rapidly wanes over time.

Data has been extracted independently by two reviewers. Percentage estimates from the original studies were reported in a spreadsheet. Results were compared and potential discrepancies reassessed and resolved. Estimates of VE against Delta or Omicron infection and/or symptomatic disease for any vaccine product or combination of products, at different times from the administration of the last dose, were extracted from the original studies retrieved from the search to inform a simple statistical model to estimate the progressive waning of immunity. We considered both VE estimates associated with primary vaccination cycle (1 dose for Ad26.COV2.S and 2 doses for the other vaccine products) and primary vaccination cycle followed by a booster dose. Descriptions of the considered endpoints presented in the eligible articles are summarized in eTables 5-6. Data were complete, with no missing information for COVID-19 vaccine products, considered endpoints, or population characteristics. To minimize potential biases led by the initial ramp-up of vaccine-induced protection, we excluded data points associated with VE measured during the first 14 days following the administration of the considered dose. Data points associated with less than 20 infections observed in the vaccinated group were excluded from the analysis.

eAppendix 2. Model Details.
VE is modeled as an exponential decay function of time described as: where: • t is the number of days from maximum protection, which is assumed to occur after 14 days from the administration of any dose; • A is the VE after 14 days from the administration of the last dose; • w is the waning rate associated with the vaccine-induced protection against the considered endpoint. Free model parameters A and w were estimated for each study via a Markov chain Monte Carlo (MCMC) approach with Metropolis-within-Gibbs sampling algorithm applied to the normal likelihood of observing the average values of VE estimated in the original study at different time intervals from vaccination.
The model was informed with mean VE estimates at different time intervals retrieved from the articles included after the systematic review. Such estimates were associated with a specific time interval (expressed in days) and were interpreted as the mean VE(t) in that time interval: for example, if we extracted from one study a mean VE estimate of 70% evaluated between 30 and 60 days after the administration of last dose, we considered it as the mean VE(t) in that interval of time. According to the proposed exponential decay function, the corresponding mean modeled VE in a specific interval of time [ 1 , 2 ] (expressed in days) can be computed as follows: where the time step of the sum is 1 day. Let us consider a VE estimate for a specific interval of time [ 1 , 2 ] extracted from a selected article (with i = 1… n, where n is the number of VE estimates extracted from that article) and denote it with ( 1 , 2 ). We assumed that ( 1 , 2 ) is distributed according to a normal distribution, with mean equal to the modeled estimate for the same time interval computed according to (2) and variance 2 , i.e.
( 1 , 2 ) ∼ ( ̅̅̅̅ ( 1 , 2 ), 2 ) (3) We applied the Gibbs sampling to likelihood (3), using Metropolis-Hastings random walk update for parameters A and w and Gibbs update for 2 . Regarding the prior distributions, we chose ∼ (0,100%) and ∼ (0,1 −1 ) (where ( , ) denotes the uniform distribution between and ). As for the variance, we decided to use the precision = 1/ 2 instead and assumed ∼ Gamma(1,1). This choice was made to exploit the fact that the fully conditioned posterior distribution of is again a Gamma distribution from which is possible to sample to perform the Gibbs update.
Our main analysis focused on providing estimates of VE at the population level. The main analysis relied on studies reporting VE estimates for a sufficiently wide age range (i.e., covering ages for at least 30 years, and including individuals aged 25-60 years). For studies reporting estimates for specific age groups only, a separate fit for each age group was performed and the VE at population level was estimated by combining the posterior distributions associated with each age group in a mixture distribution weighted by the proportion of individuals in each age group included in the original study.

Reference groups
In 38 studies 9-21,24-48 , estimates were obtained by using unvaccinated individuals as a reference group. Individuals who have received a single dose not earlier than 14 days were assumed as proxy for unvaccinated individuals in Fabiani et al, 2022 22 . Similarly, the reference group was defined by subjects who have received a single dose from at least 4 days and not more than 10 days in Fabiani et al, 2022 23 . According to Fabiani et al 22,23 , the rationale for this assumption was that unvaccinated individuals might undergo a higher number of tests and have their social habits altered by restrictions (e.g., EU Digital COVID certificate), thus leading to biased VE estimates when considering them as a reference group.

Endpoints and variants
None of the analyzed studies found a clear temporal waning of VE of a booster dose against Delta 24,33,48 , possibly due to the short follow-up associated with the available records, given that Omicron suppressed Delta circulation soon after the start of boosting campaigns. From the selected papers, we extracted: 1) twenty-one 9-29 studies providing VE estimates for primary vaccination against any SARS-CoV-2 laboratory-confirmed infection (asymptomatic or symptomatic) for Delta and six 13,20,24,[29][30][31] for Omicron; 2) twelve 14,16,32-41 studies providing VE estimates for primary vaccination cycle against symptomatic disease for Delta and nine 33,35,37,[39][40][41][43][44][45] Omicron; 3) three 24,30,46 studies providing VE estimates for primary vaccination cycle followed by a booster dose against any Omicron SARS-CoV-2 laboratory-confirmed infection (asymptomatic or symptomatic); 4) three 33,42,47 studies providing VE estimates for primary vaccination cycle followed by a booster dose against Omicron symptomatic disease.
Age structure eFigure 2. Effectiveness Over Time of Primary Vaccination Cycle and Booster Vaccination Against Omicron Symptomatic Disease. Estimated vaccine effectiveness (VE) over time against symptomatic disease with Omicron across different vaccine products. Lines: mean estimates; shaded areas: 95% CIs; points: original VE estimates from published articles 33,35,37,[42][43][44][45]47 (placed at the midpoint of the time interval for which the estimate was obtained).