Analyzing COVID-19 Serology Data Using Sparse Longitudinal Methods

Abstract

The Arizona Healthcare, Emergency Response, and Other Essential Workers Study (AZ HEROES) is a study funded by the US Centers for Disease Control and Prevention(CDC) that aims to examine the epidemiology and immunity of SARS-CoV-2 infection and COVID-19 illness among adults with high occupational exposure. This study follows a cohort of essential workers from Arizona with frequent exposure to COVID-19.As part of the surveillance in the study, participants regularly contribute blood draws that are tested for antibodies that are protective against COVID-19. One of the primary study objectives for AZ HEROES is examining post-vaccine immunologic response, and a common method for researching this is correlates of protection. However, due to participants missing blood draws or collecting blood at timepoints outside of the time window defined by the protocol, the serology data can be described as sparse, which can cause potential analytic issues.This thesis aims to investigate how immunologic response after vaccination differs for individuals in the AZ HEROES study using two different analytical approaches that account for the sparseness of the data. First, we use a Bayesian multi-level model to test for differences in peak antibody levels and rate of antibody decline over time between different populations defined by age, sex, and comorbidities. Then, we use the PACE (Principal Analysis by Conditional Estimation) method to evaluate overall and individual-level antibody trends over time. The PACE method is useful for sparse longitudinal datasets where the number and spacing of data points is irregular, which is the case for the AZ HEROES serology data.