Utilizing Group-Based Models to Identify Adverse Event Patterns After an Intervention

Wei Wang¹Sara Abbaspour²Kimberly G Blumenthal²Dean M Hashimoto²Gregory K Robbins²Elizabeth B Klerman^2*

• ¹Brigham and Women's Hospital, Boston, United States
• ²Massachusetts General Hospital, Harvard Medical School, Boston, United States

Background: Standard adverse event (AE) monitoring only records whether an event has occurred after the intervention, and not whether these events vary over time.Objective: To test whether there were statistically distinct time-varying trajectories of AE (e.g., "side effects") after an intervention and identify characteristics of individuals associated with these patterns.Design: Group-based trajectory models applied to an observational study of individuals who received one or two doses of a mRNA COVID-19 vaccine (i.e., the intervention).Participants: 50,484 healthcare personnel who received their vaccinations within the Mass General Brigham (MGB) healthcare system.Main Measures: Allergic and non-allergic AE for 1-3 days after each of two COVID-19 vaccinations.Key Results: Trajectories models identified distinct groups with different trajectories after intervention: two groups after the first vaccination and five groups after the second vaccination. These groups differed by demographics, age, prior prescription for epinephrine auto-injectors, prior COVID-19 history, time-ofday of vaccination, and vaccine manufacturer.Conclusions: Several different time-based trajectories after the intervention (e.g., first two COVID-19 vaccinations) were noted; individuals in these groups varied by demographic and clinical criteria. These time-based methods may be able to identify groups at higher risk of future adverse reactions, provide a basis for future studies of the physiology underlying these risk differentials, and improve counseling surrounding interventions associated with AEs. We suggest that trajectory-based methods be added to post-intervention surveillance.