Geostatistics is a short-hand for the collection of statistical methods relevant to the analysis of geolocated data. The aim is to study geographical variation throughout a region of interest, but using data that are limited to observations from a finite number of sampled locations. The term model-based geostatistics (MBG) was coined to mean the embedding of geostatistics within the general frameworks of statistical modelling and likelihood-based inference, as applied to geostatistical problems.
MBG uses a model-based approach, whereas the current method uses a design-based approach, to estimate trachoma prevalence. The current design-based approach uses prevalence data and the population census from a single evaluation unit (EU), and provides a point prevalence estimate and confidence intervals. A model-based approach also uses prevalence data, but can use data from multiple EUs. It allows the inclusion of relevant factors such as population density and environmental and sociodemographic covariates, whilst accounting for spatial variation exhibited in the data. In addition to the prevalence estimates, MBG provides a prediction of how likely an area is to have reached elimination (Probability of being Below Threshold, PBT) as well as a point prevalence.
The PBT (formerly referred to as the predictive probability of elimination, or PPE) is the probability of having achieved the elimination target: a prevalence of less than 5% for trachomatous inflammation—follicular (TF), or less than 0.2% for trachomatous trichiasis (TT). As PBT values get closer to 100% or 0%, we can be very confident that the elimination has or has not been achieved. A PPE of 50% would indicate a lot of uncertainty in the achievement of elimination.
Setting the PBT threshold is a decision that is both political and technical, as it is a question about how much uncertainty the global trachoma community is willing to tolerate. A special advisory group (SAG) is in the process of being created, led by Lancaster University as a WHO Collaborating Centre. The SAG will include members from research, programmatic, policy and NGO backgrounds, in order to get insight from as wide a range of perspectives as possible.
No, surveys will still be required in most cases as some data are always needed to run the models. The more data that are available, the better the predictive ability of the model.
To date, MBG has been used to create a geostatistical survey design to quantify the likelihood that TT prevalence at EU-level is below the TT elimination threshold. Click here to view the publication.
Permission to use country-specific GTMP- and/or Tropical Data-supported prevalence data, with access provided to the individual-level data contained in all relevant datasets. Access to the relevant shapefiles if these are not already publicly available. Other criteria will also need to be fulfilled in order to determine whether MBG can be used, such as: having data to feed into the model; the data showing spatial correlation; having shapefiles to demarcate the EU boundaries; having cluster-level GPS coordinates; and having the necessary covariate data.
Past research in both trachoma and other neglected tropical diseases (NTDs), such as soil-transmitted helminthiases, lymphatic filariasis, has shown that MBG requires fewer clusters compared to the standard survey design. This is because MBG targets certain areas for sampling based on expected prevalence of TF/TT, and provides more information than random sampling.
MBG has not yet been used to estimate prevalence in inaccessible EUs, however, it is theoretically possible to do this when there are enough existing data in nearby EUs to estimate trachoma prevalence in the area of interest. There would need to be strong spatial correlation present in the available data in order for this application to work, and several assumptions made that correlations that exist within the available data also apply to the population in the inaccessible area.
When there is large population movement within an EU, one of the challenges for the models is trying to incorporate more assumptions and to account for more variation than normal, and this leads to a lot of uncertainty. It may be possible to account for population movement if there was information on how people move, however this would require more specialised models, which in turn would require a lot more data. To date, this has not been done as there are very few reliable data on population movement. For reference, the current, standard data analysis method makes an assumption that there is no population movement.
This would depend entirely on the spatial relationship of trachoma amongst the different islands. If there was strong spatial correlation between islands it may be possible to use MBG, however this would need to be reviewed on a case-by-case basis.
To date, MBG has been implemented to estimate trachoma prevalence for a specific area (usually an individual EU) by using data from a wider area within the same country. However, we do not use data from neighbouring countries, because previous research has shown that the relationship between trachoma prevalence and the covariates (environmental, demographic, etc.) tends to be highly specific to each country and so using data from another country would not necessarily be reliable. While it is possible to use MBG to reanalyse historical data within a country, there must be a specific and justifiable reason to do so, as the purpose of MBG is to enhance TD methods, not replace existing survey results.
Funding is currently provided by the Task Force for Global Health and USAID, up until May 2024.
Currently, the MBG models require intensive computations and the larger the datasets that have to be processed, the longer it takes - some take up to a week to generate results! It is therefore not yet possible for Tropical Data or Lancaster University to support health ministries to implement MBG themselves. There is an app being developed that may in the future allow health ministries to run MBG themselves after appropriate training, however this is not yet ready.
In terms of support for health ministries to engage with MBG methods, we will be creating training materials and guides on when MBG can be used, how the models work, and how to interpret the outputs. We hope there will be funding available in the future to hold workshops on utilising MBG as part of the Tropical Data platform.
MBG was officially confirmed as an acceptable method to demonstrate that TT prevalence was below the elimination threshold at the 4th Global Scientific Meeting on Trachoma in 2018 (https://www.who.int/publications/i/item/who-htm-ntd-pct-2019.03). In relation to incorporating MBG as a routine option for data analysis and/or survey design within the Tropical Data platform, we are still in the research stage of the process, and there are a number of steps that need to be taken before we can start implementing it on a regular basis. If there is interest in using MBG before it is a routine offering, contact the Lancaster University (email@example.com) and/or Tropical Data (firstname.lastname@example.org) teams directly.
All outputs from the model will be owned by the countries whose data were used to generate them.
Data will be shared with Lancaster University via the health ministry. Tropical Data is available to support this process by preparing the required dataset(s) and a secure link for the health ministry to share these data with Lancaster University. Lancaster University has a memorandum of understanding that they will not share the data they receive with any third parties, and will only use it for the purpose agreed upon by health ministries.
This is a technical decision and a political one. We will seek advice from the SAG to help make recommendations to health ministries for such scenarios.