|
BACKGROUND: There is insufficient information on the human immunogenic response to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2).
OBJECTIVE: This study aimed to model waning SARs-CoV-2 immunoglobulin G (IgG) antibody detection levels given time since a self-reported positive viral test.
METHODS: A seroprevalence study was conducted within a United States (US) health system located in the Midwest. Participating hospital and clinic employees completing a study survey were eligible to receive a free SARS-CoV-2 IgG antibody test. A generalized linear model was fit regressing IgG detection levels on time since a known prior infection as documented in the study survey. A literature review was conducted to locate and recreate serial SARS-CoV-2 IgG antibody detection level data from external studies. These data were scored based on the study model and used to evaluate the predictive out-of-sample accuracy of the estimate.
RESULTS: Of the 6,009 eligible employees, 2,848 completed the study survey, and 2,118 had antibody testing. Of these employees, 221 reported a prior SARS-CoV-2 infection and date they received the positive test result. These data were used to model IgG detection values given time since a prior infection. Antibody testing for these employees was taken a median of 90 (IQR: 59, 153) days since their reported infection. The study model estimated a multiplicative IgG detection decrease of 0.99 (95% CI: 0.99, 1.00) per day since the prior positive test. Five external studies were found with applicable test information and used to examine out-of-sample model accuracy. Over fifty percent of these external data were recreated and represented 131 patients and greater than 300 individual tests. The crude out-of-sample model error was 0.0 (SE: 0.1) and -0.9 (SE: 0.1) when controlling for patient clusters within studies.
CONCLUSIONS: Given the near-real-time dissemination of pandemic information, flexible modeling strategies and validation processes were explored. The presented modeling approach served to validate the possible utility of the presented IgG prediction model.
|
