Estimating the contribution of subclinical tuberculosis disease to transmission

This repository contains all the data and files necessary to re-run the analyses in:

Emery JC, Dodd PJ et al. Estimating the contribution of subclinical tuberculosis disease to transmission.

Full details of the study can be found in the main paper and supplementary materials.

Repository structure

Contents of the folders in the repository:

1. data: Input data

2. stan: Stan model files

3. R: R files to:

   • 1_odds_ratios: Calculate odds ratios from household contact data

   • 2_run_stan: Perform MCMC runs in stan

   • 3_rel_inf: Estimate relative infectiousness from model results

   • 4_prevalence: Meta-analyse prevalence survey data (e.g. proportion of prevalent TB that is subclinical)

   • 5_prop_trans: Estimate the contribution of subclinical TB to transmission

   • 6_outputs: Contruct plots

4. interim_outputs: Intermediate outputs to be used as inputs for further analysis

5. outputs: Plots to be included in the:

   • main_paper: Main paper

   • sup_mats: Supplementary materials

Note that by default the output folders interim_outputs, outputs/main_paper and outputs/supp_mats contain a placeholder textfile Placeholder.rtf to preserve the otherwise empty folders before any outputs are generated.

Note also that hereafter XXXX denotes either viet (Viet Nam), phil (Philippines), bang (Bangladesh) or act3 (ACT3).

Data sources

Household contact data (Supplementary Table 1): HHC_data_XXXX.csv

Subclinical and clinical disease durations (derived from Supplementary Table 2): duration_data.csv

Prevalence survey data (Supplementary Table 3): prevalence_survey_data.csv

Implementation

To perform analyses as per the main paper or supplementary materials:

Calculate odds ratios from household contact data

1. Household contact data is stored in the folder data and labelled HHC_data_XXXX.csv.

2. Odds ratios by symptoms or smear status are are estimated with odds_symp.R or odds_smear.R, respectively, in the folder R/1_odds_ratios.

3. Results are saved as odds_symp.csv and odds_smear.csv to the folder interim_outputs

Perform MCMC runs in stan

1. Household contact data is stored in the folder data and labelled HHC_data_XXXX.csv.

2. A MCMC run is then executed with run_stan.R in the folder R/2_run_stan using the appropriate household contact data and stan model file from the folder stan.

3. The appropriate data and stan model file are specified in run_stan.R using viet (Viet Nam), phil (Philippines), bang (Bangladesh) or act3 (ACT3). All studies use model.stan except Bangladesh, which uses model_bang.stan since only smear-positive index cases were included.

4. Full details of the model fit are saved as fit_XXXX.rds to the folder interim_outputs.

5. The mean and variance of the (log) posterior of the relative hazards are saved as rel_hazards_XXXX.csv to the folder interim_outputs.

Estimate relative infectiousness from model results

1. Mixed-effect meta-anayses of the relative hazards from subclinical and smear-negative index cases are performed using rel_inf_s.R and rel_inf_n.R in the folder R/3_rel_inf, respectively.

2. In both scripts the mean and variance of the (log) posterior of the relative hazards (rel_hazards_XXXX.csv in the folder interim_outputs) for all studies are imported.

3. In both scripts a mixed-effects meta-analysis is performed with the results for the summary value saved as either rel_hazards_s.csv or rel_hazards_n.csv to the folder interim_outputs.

4. In the script for the relative hazards from subclinical index cases, the duration of subclinical TB versus clinical TB is used to the estimate the relative infectiousness of subclinical TB for each study separately and the summary value. In the smear-negative case the relative hazards are assumed equal to the relative infectiousness.

5. The results for relative infectiousness for each study and the summary value are saved as rel_inf_s.csv and rel_inf_n.csv to the folder interim_outputs.

Meta-analyse prevalence survey data

1. Prevalence survey data is stored in the folder data and labelled prevalence_survey_data.csv.

2. This data is then imported to scripts in the folder R/4_prevalence that perform mixed-effect meta-anayses of:

   1. The proportion of prevalent TB that is subclinical (prop_prev_sub.R)
   2. The proportion of subclincal TB that is smear-positive (prop_sub_pos.R)
   3. The proportion of clinical TB that is smear-positive (prop_clin_pos.R)

3. The results for the summary value are saved as prop_prev_sub_summ.csv, prop_sub_pos_summ.csv and prop_clin_pos_summ.csv to the folder interim_outputs.

4. The results for the individual surveys plus summary value are saved as prop_prev_sub_all.csv, prop_sub_pos_all.csv and prop_clin_pos_all.csv to the folder interim_outputs.

Estimate the contribution of subclinical TB to transmission

1. The proportion of transmission from subclinical TB is estimated for each setting separately using prop_trans_setting.R in the folder R/5_prop_trans.

2. Prevalence survey data (prevalence_survey_data.csv) is imported from the folder data.

3. The summary value for the relative hazards from subclinical and smear-negative index cases (rel_hazards_s.csv and rel_hazards_n.csv) are imported from the folder interim_outputs.

4. The duration of subclinical TB versus clinical TB is again used to the estimate the relative infectiousness of subclinical TB. In the smear-negative case the relative hazards are assumed equal to the relative infectiousness.

5. The proportion of transmission in each setting is then estimated and saved as prop_trans_setting.csv to the folder interim_outputs.

6. A summary value for the proportion of transmission from subclinical TB is then estimated using prop_trans_summ.R in the folder R/5_prop_trans.R.

7. The process is the same as the above except that at step 2) the summary values prop_prev_sub_summ.csv, prop_sub_pos_summ.csv and prop_clin_pos_summ.csv are imported from the folder interim_outputs.

8. The summary value for the proportion of transmission from subclinical TB is then saved as prop_trans_summ.csv to the folder interim_outputs.

Contruct plots

1. Plots for the main paper are constructed using outputs_main_paper.R in the folder R/6_outputs, which imports all relevant results from the folder interim_outputs. All plots are saved to the folder outputs/main_paper with the title of their respective label in the main paper (e.g. fig_3A.png).

2. Plots for the supplementary materials are constructed using outputs_supp_mats.R in the folder R/6_outputs, which imports all relevant results from the folder interim_outputs.

3. The appropriate study is specified in outputs_supp_mats.R using viet (Viet Nam), phil (Philippines), bang (Bangladesh) or act3 (ACT3).

4. All plots are saved to the folder outputs/supp_mats with the type of plot and study name (e.g. trace_viet.png) for the Viet Nam trace plot.

5. Detailed model results (e.g n_eff, Rhat, mean, mcse, sd and sample quantiles) are viewed in a web-browser using shinystan.

Further resources

Stan is used to perform the MCMC runs. The R packages rstan and shinystan are used to interact with stan and analyse the results using R. Meta-analyses are performed using the R package metafor.

rstan documentation: https://mc-stan.org/users/interfaces/rstan

shinystan documentation: https://mc-stan.org/users/interfaces/shinystan

metafor documentation: https://www.metafor-project.org/doku.php