Code and data for:
Dodd PJ, Mafirakureva N, Seddon JA, McQuaid CF. The global impact of household contact management for children on multidrug-resistant and rifampicin-resistant tuberculosis cases, deaths, and health-system costs in 2019: a modelling study. The Lancet Global Health. 2022 May 18 https://doi.org/10.1016/S2214-109X(22)00113-9
The analysis takes the form of a probabilistic sensitivity analysis (PSA) and is split into a number of analyses which are designed to be run in a particular order, usually indicated by the two digits at the start of their filename.
The directory structure is
.
├── data
├── indata
├── output
└── R
└── utils
NB the folder data/ is excluded from the repo - it must be created by hand in the right relative location for scripts to write large temporary data to. All necessary input data to run the analysis is in indata/.
This analysis was run using R version 4.1.0 (2021-05-18).
The following R packages must be available:
- utility: here, glue, stringr
- data manipulation: data.table, dplyr,
- plotting: ggplot2, ggthemes, scales, ggrepel, ggpubr
- calculations: discly, HEdtree (via devtools::install_github('petedodd/packagename'))
- spatial strucutre (optional, since output also included): tmap, rworldmap, spdep
- additional for cost analysis (optional, since output also included): tidyverse, viridis, readxl, httr
The estimates of household size by index TB age/sex is based on previously published work:
- https://www.thelancet.com/journals/langlo/article/PIIS2214-109X(18)30401-7/fulltext
- https://github.com/petedodd/PINT
With a full-size PSA, a reasonable amount of RAM is needed (it was run on a machine with 36Gb; it probably needs at least 10Gb to not cause trouble).
Commenting out the rep limitation in file 04 around line 14 (ie using 100 replicates) should allow use on a machine with (eg) 8Gb RAM.
This file generates the estimates of FQR resistance in RR. It is optional to run this analysis as the output is included in indata/. Within this analysis, there are two parts: part 1 generates the nearest neighbour structure and is also optional (output in indata/); part 2 uses the neighbour structure to generate the FQR estimates.
This file generates the country-level unit cost estimates. It calls utils/TB_care_costs.R to download some data from the web if not already present, and do some initial processing. Further collated input from indata/other_public_costs.csv is used. The outputs are saved in outdata/country_unit_costs.R. It is optional to run this script as the output data is included in the repo.
This file uses notification data and previous estimates of household contacts to construct numbers at risk as the starting point for the PSA object.
This file calculatates the RR case detection ratios as RRest.csv for later use, but is otherwise self-contained.
This file builds the base PSA object for onward analyses.
This file performs the calculations. It contains functions defining the interventions, and calculating the costs from unit costs. After running the two scripts below in utils/, the outcome data for all scenarios are computed and saved.
It sources:
utils/
- maketree.R (builds decision tree)
- ModelFuns.R (functions for constructing necessary variables)
This file constructs the decision tree and makes the functions to compute outcomes.
This contains most of the non-structural modelling logic. It loads the parameter distributions (with test output if uncommented), specifies the dependencies of transition probabilities on attributes, and includes other functions to generate a PSA initial dataset.
This file computes tabular and graphical summary output from the PSA data.
It sources utils/makeLYs.R which will be a bit slow first time around. This script computes mean discounted life-years for each country to associate with deaths.
These LY computations uses the R package https://github.com/petedodd/discly whose underlying data is described in the paper and repo:
The sensitivity analysis to apply additional restriction factors for pulmonary TB are run by changing the input indata/pulmonary.sensitivity.analysis.txt to an integer >0 and then running scripts 01 through to 05. The R/utils/uploadresults.R (authors only) will load the tables to a separate part of the document.