Seroprevalence of anti-SARS-CoV-2 antibodies 6 months into the vaccination campaign in Geneva, Switzerland, 1 June to 7 July 2021

Background Up-to-date seroprevalence estimates are critical to describe the SARS-CoV-2 immune landscape and to guide public health decisions. Aim We estimate seroprevalence of anti-SARS-CoV-2 antibodies 15 months into the COVID-19 pandemic and 6 months into the vaccination campaign. Methods We conducted a population-based cross-sectional serosurvey between 1 June and 7 July 2021, recruiting participants from age- and sex-stratified random samples of the general population. We tested participants for anti-SARS-CoV-2 antibodies targeting the spike (S) or nucleocapsid (N) proteins using the Roche Elecsys immunoassays. We estimated the anti-SARS-CoV-2 antibodies seroprevalence following vaccination and/or infection (anti-S antibodies), or infection only (anti-N antibodies). Results Among 3,355 individuals (54.1% women; 20.8% aged < 18 years and 13.4% aged ≥ 65 years), 2,161 (64.4%) had anti-S antibodies and 906 (27.0%) had anti-N antibodies. The total seroprevalence was 66.1% (95% credible interval (CrI): 64.1–68.0). We estimated that 29.9% (95% Crl: 28.0–31.9) of the population developed antibodies after infection; the rest having developed antibodies via vaccination. Seroprevalence estimates differed markedly across age groups, being lowest among children aged 0–5 years (20.8%; 95% Crl: 15.5–26.7) and highest among older adults aged ≥ 75 years (93.1%; 95% Crl: 89.6–96.0). Seroprevalence of antibodies developed via infection and/or vaccination was higher among participants with higher educational level. Conclusion Most of the population has developed anti-SARS-CoV-2 antibodies, despite most teenagers and children remaining vulnerable to infection. As the SARS-CoV-2 Delta variant spreads and vaccination rates stagnate, efforts are needed to address vaccine hesitancy, particularly among younger individuals and to minimise spread among children.

Since the beginning of the pandemic, have you had a confirmed COVID-19 diagnosis, meaning having a positive COVID-19 from a nasal/throat swab, either RT-PCR or rapid antigenic test?

Yes / No
If yes, date (day / month / year

Educational level
What is the highest level diploma/certification that you have obtained?
-None -Primary school and/or orientation cycle -Secondary education -Maturité/High School -Professional training -Certified apprenticeship (CFC) -Professional training -Non-certified apprenticeship -Professional training -Higher professional degrees (post-CFC) -Tertiary education -Bachelor/Master degree -Tertiary education -Doctorate/PhD degree -Other

S1. Overview of Statistical Framework
Our aim was to infer the proportion of the population having any antibody against SARS-CoV-2, as well as the proportion of those who acquired antibodies through natural infection as opposed to vaccination. We do so by modelling jointly the antibody response measured by the Roche-N and Roche-S immunoassays together with participants' responses to a vaccination questionnaire. We disentangle natural infection from vaccination antibody responses using the fact that the only available vaccines in Switzerland to date-the mRNA-1273 from Moderna/US NIAID, [5] and the mRNA-BNT162b2/Comirnaty from Pfizer/BioNTech [6]-both elicit a response exclusively to the S protein of SARS-CoV-2, as opposed to natural infections which typically elicit a response to both the N and S virus proteins. We expand previous Bayesian modelling frameworks used for seroprevalence estimates that account for demographic parameters (sex and age), test performance and household infection clustering. [1,2] The main additions to the previous models are that we now model jointly the response to both tests, and that we account for vaccination-induced antibody response. The underlying probability of antibody status accounts both for the probability of natural infection ⋋ and vaccination status, ∈ {0,1}. Following previous modeling frameworks, [1,2] we model the probability of natural infection as a function of sex and age category, accounting for household infection clustering through a random effect, ℎ :

S1.1 Multinomial response model
where is the matrix of covariates, and the vector of regression coefficients. The probabilities of antibody status are then given by: where ++ , −+ , +− are the conditional probability of having + + , − + , + − responses, respectively, upon natural infection, is the probability of having a vaccine-induced + response as a function of the conditional probability of antibody response upon infection , = × .

S1.2 Vaccination
To obtain population-level seroprevalence estimates, we also model the proportion of vaccinated individuals in each sex/age class following the approach used for natural infection: Given vaccination policy recommendations in the state of Geneva, previously infected individuals were discouraged from being vaccinated in the early phase of the vaccination campaign, thus making the probability of vaccination dependent on the infection status of the individual. We account for this dependence by modelling separately the probability of vaccination given the infection status and marginalizing out the infection status: where is the vector of all model parameters, ,~ indicates infection and non-infection, respectively, and is the vector of regression coefficients giving the difference in probability of vaccination between infected and non-infected individuals. When estimating the population-level seroprevalence, we account for the conditional probability of vaccination given non-infection, |~, in the probability of a negative S and N response accounting for household vaccination clustering, −− , as:

S1.3 Diagnostic test performance
The individual performance of both N and S tests is incorporated hierarchically following Gelman & Carpenter [7]. The sensitivity, + , is determined using + RT-PCR positive controls from a laboratory validation study [8], of which + tested positive. The specificity, − , is determined using − prepandemic negative controls, of which − tested positive. For the Roche N test, these values are modulated by data in Ainsworth et al. [9]. For the Roche S test, the laboratory study data are modulated by those available on the Roche website (last accessed July 19, 2021).

S1.4 Priors
We follow a similar setting of the priors on the tests' sensitivity and specificity as Gelman & Carpenter [7]. For study j, the specificity − and sensitivity + are drawn from normal distributions on the log odds scale:
We used standard normal (0,1) priors for the logistic regression coefficients for infection . For coefficients of vaccination and for coefficients of the difference in probability of vaccination between infected and non-infected individuals , we also used standard normal except for the youngest age group (ages 0-5 years and 6-11 years). For these two age groups, ~ (−10, 0.01) to reflect the fact that there was almost no vaccination in these youngest age groups in Geneva at the time of the study (NB vaccination registration for those aged 12-15 years opened on June 16, 2021: https://www.ge.ch/en/getting-vaccinated-against-covid-19/covid-19-vaccination-campaign-geneva, last accessed July 20, 2021). Correspondingly, ~ (0, 0.01) for these two age groups.
The priors for the means of the household random effects ℎ and ℎ, , followed standard normal, and for standard deviations of the household random effects were positive half-normals, ℎ ~ + (0, 2) and ~ + (0, 2). We use a Dirichlet prior on the conditional probability of having + + , − + , + − responses upon natural infection, ++ , −+ , +− , ~Dir(10, 1, 1), to highly favour production of both anti-S and anti-N antibodies upon infection. Finally, we put a strong prior on the conditional probability of antibody response after vaccination ~ Beta(10, 0.1).

S1.5 Implementation
The model was coded in the probabilistic programming language Stan [10] using the Rstan package [11]. R [12] version 4.1 was used for data analysis. Four chains were run with 1500 iterations each, 250 of which were warmup, to give a total of 5000 posterior samples. Convergence was assessed by checking that R ̂≈ 1, that the effective sample size was reasonable for all parameters, and visually using shinystan [13] diagnostics checks.

Supplementary Figure S2. Comparison of age and sex composition of study sample (bars) and the Geneva population (dots)
Dark yellow represents males; blue represents females.

Supplementary Figure S4. Antibodies response category and vaccination status
Number of participants in the four possible categories of the S and N tests. + indicates antibodies detected; -indicates antibodies not detected. .0 (53.6-58.4) a Self-reported having received at least one dose of any COVID-19 vaccine, more than 14 days before blood drawing. b Estimated vaccinated proportion in population, reported as % and 95% credible interval, adjusted for test performance of both immunoassays and post-stratified to account for age distribution in the Geneva general population and for household clustering of infection and vaccination. c Self-reported education level among participants aged ≥18 years (N = 2520). Percent increase calculated as: ((Jun-Jul seroprevalence / Nov-Dec seroprevalence) -1) x 100.

Supplementary Table S4. Comparison of seroprevalence of anti-SARS-CoV-2 antibodies a naturally developed through infection by November-December 2020 and
Absolute increase calculated as: Jun-Jul seroprevalence -Nov-Dec seroprevalence.
Absolute increase N calculated as: absolute increase % x Geneva population Seroprevalence estimates for November-December 2020 from previous seroprevalence study [2].