Analysis of the HERO Study

Summary of project

This repository contains an R package which performs the data cleaning, model fitting and plotting for the analysis of the SARS-CoV-2 HERO study performed at Sheffield Teaching Hospital. The mcmc sampling is done in cmdstanr (not rstan).

Installation guide

Quick-install instructions

Clone the repository and use devtools::load_all() to load everything into R. Then go through the vignettes in the /vignettes folder.

Step-by-step installation instructions

• Before cloning, make sure you have installed github lfs on your computer.
• In R, make sure you have installed and loaded devtools.
• Clone this repository onto your computer (e.g. git clone https://github.com/dchodge/hero-study.git)
• Load R and make sure your working directory is /YOUR-PATH/hero-study
• Use devtools::install_dev_deps() to install all dependency packages automatically (alternatively see the DESCRIPTION file for a list of dependencies and their sources). Depending on your operating system several packages may have additional installation requirements (rstan and cmdstanr). See the installation instructions for these packages for details.
• Use cmdstanr::install_cmdstan() to install the cmdstan backened used for model fitting. See the cmdstanr documentation for further details.
• Use devtools::load_all() to install the package.
• Now consult the vignettes in the /vignettes folder to run the code.

Overview of files

There are two Markdown files in the /vignettes folder (study_run_figs.Rmd and study_other):

• /vignettes/study_run_figs.Rmd — runs the three models in the study and plots the figures
• /vignettes/study_other.Rmd — produces everything else needed for the study (tables, point estimations etc)

The guts of the code is in the /R folder. The scripts are as follows:

• /R/data_clean.R — cleans the raw serological data into a R-friendly version
• /R/data_model.R — creates datasets for each of the models
• /R/m1_prev.R — All the code needed to run, process the posterior distributions, and plot the figures for the prevalence model
• /R/m2_abkin.R — All the code needed to run, process the posterior distributions, and plot the figures for the antibody kinetics model
• /R/m3_hetsen.R — All the code needed to run, process the posterior distributions, and plot the figures for the hetergeneous sensitivity and specificity model
• R/supp.R — code to make the supplementary figs

The mcmc samples are in the /data folder:

• /data/datafit.RData Is all the data required to fit the mcmc models. This is saved with the save_all_datafit() function, and loaded into the environment with the load_all_datafit() function.
• /data/fit_mcmc_prev.RData are the raw mcmc samples from the cmdstanr model for the prevalence model shown in the Figure 1.
• /data/fit_mcmc_asymp.RData are the raw mcmc samples from the cmdstanr model for the asymptomatic model shown in the Figure S6.
• /data/fit_mcmc_start.RData are the raw mcmc samples from the cmdstanr model for the starting titre value model shown in the Figure 2.
• /data/fit_mcmc_change.RData are the raw mcmc samples from the cmdstanr model for the change in titre value model shown in the Figure 2.
• /data/fit_mcmc_sens.RData are the raw mcmc samples from the cmdstanr model for the heterogenous sensitivity model shown in Figure 3.
• /data/fit_mcmc_spec.RData are the raw mcmc samples from the cmdstanr model for the heterogenous specificity model shown in Figure S11.

The stan code and the raw mcmc outputs is contained in the include/ folder.

The figures used in the final article are in the outputs/figs/ folder.

Linked publications

This research is still work-in-progress, so the manuscript is not available yet.

Contact details

Please email david.hodgson@lshtm.ac.uk with any queries relating to this code.