Supplementary code and data for "Accounting for assay performance when estimating the temporal dynamics in SARS-CoV-2 seroprevalence in the U.S."

This repository includes all the R scripts and data necessary to reproduce the analyses in the manuscript.

Data

• supporting_data.csv: The processed data required to run the model, including the seroprevalence estimates, numbers of reported cases and deaths, excess deaths, numbers of hospitalisations and tests, and what assays were used.

• blood_donors_surveys_processed.csv: The processed version of the blood donors seroprevalence surveys, necessary to make the comparisons against our estimates of proportions infected.

Code

• functions.R: Functions necessary to run the analyses, and model formulae.

• 01_glm_main_reference_model.R: Runs the main reference model (single logistic regression with no waning). This produces output/main_glm.rds.

• 02_glm_waning_models.R: Runs the waning models. This script is intended to run on an HPC due to the large number of models being fit. This produces a large number of files, which in the analyses are assumed to be combined into a single file, which is here provided as output/glm_3-wanings.rds.

• 03_get_overall_uncertainty.R: Estimates uncertainty in our estimates of proportions infected to account for uncertainty in the selection of times to seroreversion for the three assays, and in the lead/lag times between a case being reported and seroconverting. This produces a large number of files, which in the analyses are assumed to be combined into a single file, which is here provided as output/cis_glm_3-wanings.rds.

• 04_glm_with_splines_main_reference_model.R: Runs the main reference model using splines. This produces output/main_glm_splines.rds.

• 05_figures.R: Reproduces all figures in the manuscript (including supporting figures).

Output

• main_glm.rds: GLM object of the main reference logistic regression model, including RMSE and LOO median RMSE. The output is produced by 01_glm_main_reference_model.R.

• glm_3-wanings.rds: A table with summary metrics for all the waning models. The different combinations of times to seroreversion and lead/lag time are arranged by row, with the corresponding AIC, RMSE, and LOO median RMSE for each combination. The output is produced by 02_glm_waning_models.R.

• cis_glm_3-wanings.rds: A table with the lower and upper uncertainty boundaries for both predicted seroprevalences from the models, and the estimated proportions infected. The uncertainty accounts for the 95% confidence interval of each individual waning logistic regression, and the uncertainty due to the selection of the times to seroreversion of the three assays and the lead/lag time (uncertainty across logistic regressions). The output is produced by 03_get_overall_uncertainty.R.

• main_glm_splines.rds: GLM object of the main reference GLM model with splines. The output is produced by 04_glm_with_splines_main_reference_model.R.

• proportions_infected_estimates.csv: Estimates of proportions infected per state and point in time, with lower and upper limits of uncertainty, from the best waning model.