Bayesian time series regression methods for estimating national immunization coverage – phases I & II
National estimates of immunization coverage are crucial for monitoring and evaluating coverage levels and trends, as well as immunisation goals and targets, at the national, regional and global levels and are an important indicator for measuring progress towards Sustainable Development Goal (SDG) 3. Countries report estimates of coverage annually to the World Health Organisation (WHO). These are supplemented with survey data and other demographic and contextual data to produce WHO and UNICEF estimates of national immunisation coverage using a deterministic computational logic approach. However, this deterministic approach is incapable of characterising uncertainties related to both the input data sets and estimates of coverage that are produced.
The overarching aim of this project was to develop a novel Bayesian statistical methodology that leverages temporal and spatial variation in immunisation coverage and correlations among vaccines to produce accurate national-scale, annual point and interval estimates of immunization coverage for multiple vaccines and all WHO countries.
In Phase I of the project, a Bayesian hierarchical modelling approach was developed which adopted some of the rules implemented in the WUENIC computational logic approach to process and harmonize the multiple input data and then fitted a Bayesian hierarchical model to the pre-processed data to obtain smoothed estimates of national immunisation coverage and associated uncertainties.
In Phase II, the Bayesian hierarchical model was extended to allow the combination of information from the multiple input data sources within the model, thus eliminating the need to integrate information from the multiple data sources prior to model-fitting.
Outputs from this model are source-free, bias-corrected estimates of national immunisation coverage and associated uncertainties.
We also developed a contributed R package ‘imcover’ implementing the No-U-Turn Sampler (NUTS) in the Stan programming language to enhance the utility of the methodology.
Outputs for this project included:
- Modelled estimates of DTP1, DTP3, MCV1, MCV2 and PCV3 coverage for all WHO countries for the period 2000 – 2020.
- R package ‘imcover’ for implementing the methodology