Abstract
BACKGROUND: Belgium experienced two SARS-CoV-2 epidemic waves in 2020, in spring and autumn. Due to limited testing capacity, restrictive case definitions, asymptomatic infections, and incomplete testing compliance, case counts represent only a lower bound of SARS-CoV-2 infection incidence. We estimated this incidence from February 2020 to January 2021 by jointly modelling seroprevalence and surveillance data. METHODS: We developed a hierarchical Bayesian model that jointly fits seroprevalence, hospitalization, and mortality data to a shared latent incidence curve, represented by a spline. The model accounts for time-varying serological test sensitivity (reflecting seroconversion and seroreversion) using informative priors, and simultaneously estimates test specificity, infection-to-event distributions, and time-varying infection hospitalization rates (IHR) and infection fatality rates (IFR). Seroprevalence data comprised 37,235 samples from two repeated cross-sectional studies: residual laboratory samples tested with the EuroImmun IgG ELISA and blood donor samples tested with the Wantai Ab ELISA. Hospitalization and mortality counts were obtained from national COVID-19 surveillance. RESULTS: By early 2021, an estimated 19.0% (95% Credible Interval (CrI) 17.4-20.7), 13.6% (CrI 11.5-15.8) and 10.8% (CrI 8.7-13.2) of the Belgian 18-49, 50-64 and 65-74 year-olds had been infected with SARS-CoV-2. The first wave mostly affected the younger age group, with a peak weekly incidence of 2.0% (CrI 1.7-2.3) late March 2020. The second wave peaked late October 2020 with weekly incidences of 1.6% (CrI 1.2-2.1) among 65-74 year-olds and 2.8% (CrI 2.4-3.3) among 18-49 year-olds. IHR and IFR were considerably higher in older age groups and declined over time. Among 65-74 year-olds IHR declined from 9.9% (CrI 7.3-14.2) to 5.0% (CrI 3.5-7.1) and IFR from 2.8% (CrI 2.0-4.0) to 1.2% (CrI 0.9-1.7). CONCLUSION: An estimated 16.3% (CrI 15.1-17.4) of the Belgian adult population had been infected with SARS-CoV-2 by early 2021. Joint modelling of seroprevalence and surveillance data provides a framework for estimating infection burden.