BACKGROUND: Most longitudinal studies of COVID-19 incidence have used unlinked samples. The city of Manaus, Brazil, has a blood donation program which allows sample linkage, and was struck by two large COVID-19 epidemic waves between mid-2020 and early 2021. METHODS: We estimated the changing force of infection, i.e. incidence in susceptible individuals. Seroconversion was inferred by a mixture model for serial values from the Abbott Architect SARS-CoV-2 nucleocapsid (N) IgG assay. We estimated the number of suspected COVID-19 hospitalizations arising from each infection over calendar time. RESULTS: Whole blood donations between April 2020 and March 2021 were included from 6734 people, 2747 with two or more donations. The inferred criterion for seroconversion, and thus an incident infection, was a 6.07 fold increase in N IgG reactivity. The overall force of infection was 1.19 per person year (95% confidence interval 1.14-1.24) during the two main waves. The estimated number of suspected hospitalizations per infection, was approximately 4.1 times higher in the second wave than in the first. CONCLUSIONS: Serial values from this assay can be used to infer seroconversion over time, and in Manaus show a higher number of suspected COVID-19 hospitalizations per infection in the second wave relative to the first.