Estimating the local burden of disease during the first wave of the COVID-19 epidemic in England, using different data sources from changing surveillance practices

This repository contains the code used to predict the potential trajectories of symptomatic cases and total infections for COVID-19 during the first epidemic wave in England, based on reported COVID-19-related deaths per week and per lower tier local authority (LTLA). This is acheived by estimating a confirmed-case-fatality ratio according to cases detected under the expanded pillar 1 + pillar 2 testing system in place from 18th May 2020, and applying this to rescale smoothed trajectories of deaths from January to June, accounting for case-fatality risk factors of the local population. Data from serological surveys are then used to infer the total number of infections represented by these cases.

The full analysis pipeline can be run with main/run_all.R and consists of the following steps:

0. Import/clean data Import and process population estimates (by age) and covariate values (deprivation index and proportion of ethnic minority population) for each LTLA in England. Import and clean linelist of confirmed cases (incl. pillars 1&2) and COVID-19-related deaths (incl. as underlying cause or mentioned on death certificate). Scripts: 00_1_setup_pop_la.R, 00_2_setup_covariates.R, 00_3_setup_cases.R, 00_4_setup_deaths.R

1. Analysis data setup Aggregate death linelist to week and LTLA, join with covariates/population estimates and calculate age-stratified expected deaths. Aggregate case linelist to week and LTLA. Script: 01_setup_analysis_data.R

2. Descriptive Generate descriptive summaries and figures Script: 02_descriptive.R

3. Model fitting Specify priors and model formulae, then fit and generate posterior samples. Script: 03_run_models.R

4. Model comparison Summarise and compare fit of candidate models Script: 04_compare_models.R

5. Final model Summarise fit of final model Script: 05_plot_final_model.R

6. Predict Generate predictions from selected model at average covariate values, by posterior sampling. Script: 06_predict.R

7. Reconstruct cases Lag and rescale samples to represent cases detectable under the expanded testing system, throughout the whole time period. Scripts: 07_1_reshape_sims.R, 07_2_reconstruct.R

8. Summarise Generate figures and tables to summarise reconstructed case curves. Script: 08_summarise.R

Helper functions are included in /utils.

Dependencies

This analysis depends on the following packages: tidyverse, lubridate, data.table, dtplyr, sf, rgdal, spdep, ggspatial, INLA, patchwork, here, scales. All necessary packages are loaded automatically in utils/setup_env.R.