Spatiotemporal Components of Adverse Reactions Against BNT162b2 mRNA SARS-CoV-2 Vaccine Reflect Different Immunological Processes

Background: Reactogenicity of vaccines can derive from different immunological processes and is composed of a range of symptoms with varying time courses. This study aimed to decompose complex adverse reactions against the BNT162b2 SARS-CoV-2 vaccine into spatiotemporal patterns and clarify their associations with external information.  Methods: We recruited healthcare workers who were receiving the BNT162b2 mRNA SARS-CoV-2 vaccine in the Chiba University Hospital vaccination program. We collected information on adverse reactions every day after each dose using a smartphone/web-based platform. Information for immediate reactions was collected specifically on the vaccination day. We determined the serum anti-SARS-CoV-2S antibody titers after the 2nd dose. We employed non-negative tensor factorization to decompose the serial reactogenicity information into components. Associations of these tensor components with external information were analyzed using a generalized linear model and a generalized additive model.  Findings: We analyzed 1,774 participants who received two doses of vaccine. Tensor decomposition revealed four components that represented spatiotemporal patterns of adverse reactions. The original reactogenicity data were well explained by these four tensor components with minimal errors. One component, which represented diverse and severe symptoms that occurred early after the 2nd dose, was significantly associated with post-vaccination antibody titers, suggesting the involvement of acquired immunity. On the other hand, the other three components were not significantly associated with antibody titers. These four components were differently associated with immediate reactions and background factors such as demographics, habits, comorbidities, and medications.  Interpretation: Complex adverse reactions against vaccines can be explained by a limited number of spatiotemporal components identified by tensor decomposition. Our data indicate that these components reflect distinct underlying immunological processes.  Funding: This study was supported by a donation to Chiba University Hospital, the Future Medicine Founds at Chiba University, a JST Moonshot R&D Grant, a JST CREST Grant, and Japan AMED Grants. Declaration of Interest: None to declare. Ethical Approval: The study procedures for sample collection and those for analyses were approved by Chiba University Ethics Committee on February 24th, 2021 (No. HS202101-03) and April 21st, 2021 (No. HS202104-01), respectively.