An analytical framework to estimate the global public health impact of 50 years of the Expanded Programme on Immunization (EPI). This framework estimates deaths averted, years of live saved, and years of full health gained for 14 pathogens within the EPI portfolio. Exclusions include SARS-CoV-2/COVID-19 and HPV. The analysis is performed at country level for 194 UN member states for the timeframe mid-1974 to mid-2024, with results summarised at the global and WHO-region level. This analysis builds upon infectious disease modelling estimates produced by the Vaccine Impact Modelling Consortium (VIMC) and the Global Burden of Disease study (GBD).
The code in this repository fully reproduces all analyses and results presented in the following peer-reviewed manuscript. This manuscript has been accepted for publication in The Lancet and is currently in press.
Shattock AJ, Johnson HC, Sim SY, Carter A, Lambach P, Hutubessy RC, ... & Bar-Zeev N. Vaccine contributions to reductions in childhood mortality: quantifying 50 years of the Expanded Programme on Immunization. The Lancet. 2024.
See our frequently asked questions guide to help aid interpretation of this study and its findings.
This open-source repository is primarily written in R, and is stable for R versions 4.3.0 and 4.3.2. It may be possible to run alternative versions of R, but these have not been tested and verified. Configuration files are primarily written in YAML markup. The use of YAML files requires no additional software beyond the R packages upon which this repository depends.
We invite any potential collaborators interested in expanding upon this analysis to fork this repository. Please submit a pull request if you feel the parent repository would benefit from any changes to code or configuration files.
All R package dependencies will be automatically installed (if necessary) and loaded the first time the pipeline is run. If many packages need to be installed, the installation process could take up to one hour. See dependencies.R in the main directory for a full list of all R package dependencies, all of which are available on CRAN. All of the data used to generate this analysis is publicly available. In some cases, this data is stored in the /input/ directory, whilst in other cases data is pulled from the internet. As such, a live internet connection is required to run this analysis.
No special cluster computing resources are required to run this analysis. Local parallelisation is used throughout the pipeline to improve runtime, however this parallelisation is only available on UNIX operating systems. Running this analysis on Windows operating systems is still possible, yet will be considerably slower. Expect the full analysis (excluding any package installations) to take around one hour on a UNIX machine and two to four hours on a Windows machine, dependant on machine specifications. R console output will keep the user updated with runtime progress.
All code required to run the pipeline sits within the main repository directory (called epi50 by default). Initially, three sub-directories exist within this main directory:
- A
/config/directory containing a series of configuration files written in YAML. - An
/input/directory containing the input data required to run the pipeline. - An
/extern/directory containing processed results from the measles and polio models used in this analysis.
A fourth directory, /output/, is automatically created when the pipeline is first launched. Intermediary and final results are stored in this output folder. The figures presented in the peer-reviewed publication are stored within the /output/figures/ sub-directory.
The files contained in the /config/ directory, all written in YAML mark up, configure the analysis to be run. These configuration files set key options such as diseases and vaccines to be modelled, countries to be run, and covariates data to be used for regression modelling. These configuration files also set general model assumptions, general algorithm options, and define data dictionaries for converting between various data sources and EPI variable naming conventions. Configured data sets include the WHO Immunization Information System (WIISE), World Population Prospects (WPP), Global Burden of Disease (GBD), and GapMinder.
As standard, the configuration files in this repository will fully reproduce the analysis presented in the aforementioned publication.
The analysis pipeline consists of eight modules, identified by numbers 1 to 8, each described below. In general, these modules should be run consecutively from 1 to 8. The module/s to be run are defined by the run_module variable in line 20 of launch.R. Use a single value to run/re-run a specific module, or use a vector to run/re-run a subset of modules or all modules.
# All of the following are valid syntax for defining modules to be run (line 20, launch.R)
run_modules = 1
run_modules = c(1, 2, 4)
run_modules = 2 : 5
run_modules = 1 : 8
Modules are launched by running launch.R. Preferred usage is to 'source' this file (without 'echo' is ideal). When sourced, the current working directory is automatically reset the EPI50 repository. Alternative UNIX command line usage is to cd to the EPI50 repository then call sh launch.sh. By default all 8 modules will be run, in the process producing all outputs presented in the corresponding publication.
The run_prepare module loads and formats modelling estimates from VIMC and GBD, vaccine coverage data from WHO (WIISE and SIA databases), demography data from WPP, and pubic health covariate data from GapMinder. Further, this module interprets all configuration files, using data dictionaries to convert variables to EPI50 naming conventions where appropriate.
The run_external module loads and formats outcomes from transmission models simulated outside of VIMC scope. Both measles and polio are considered 'externally modelled' for the purpose of this analysis. Only minor formatting jobs are applied to externally modelled diseases, as this analysis attempts to use the outcomes in their purest form.
Note that it is possible (although not necessary) to (re)simulate the DynaMICE measles model directly from within this module. However this requires the user to clone another github repository and also have access to a computing cluster that uses the SLURM queueing system. If the user does not explicitly run the model, previously simulated output (stored in /extern/) is used.
The run_static module produces vaccine impact estimates for diseases outside of VIMC scope and which haven't been 'externally modelled'. In this analysis, static modelling is used for diphtheria, tetanus, pertussis, and tuberculosis. Broadly, vaccine impact is calculated using a combination of GBD burden estimates, vaccine efficacy profiles, and vaccine coverage. A full description of the static modelling approach used in this analysis is provided in the Supplementary Material of the publication.
The run_regression("impute") module uses time-series regression models to produce vaccine impact estimates for countries outside of VIMC scope, for diseases within VIMC scope. A series of models that vary in terms of predictor covariates are evaluated for each vaccine in each country, with the most parsimonious model selected. For countries without impact estimates, such estimates are imputed using the most commonly selected model in each WHO region.
The run_impact module seeks to determine the relationship between vaccine coverage and vaccine impact for each vaccine in each country. This is achieved by fitting four pre-defined statistical functions (straight line, exponential growth, logarithmic growth, sigmoidal growth) in cumulative space, and selecting the most parsimonious function.
The run_history module compiles the final vaccine impact results by evaluating impact functions for each vaccine in each country using all coverage values. Temporal extrapolation takes effect at this stage, where vaccine coverage values that do not have corresponding impact estimates are evaluated.
The run_regression("infer") module reapplies the regression models used for geographical imputation, but for all vaccines (including statically modelled and externally modelled diseases). The purpose of this additional regression step is to infer the predictors of vaccine impact for all diseases. Note that outcomes from this module were not presented in the final publication.
The run_results module produces all result and diagnostic figures presented in the publication.
- Andrew J. Shattock ([email protected], [email protected])
- Helen C. Johnson
- Austin Carter
- So Yoon Sim