top of page
Epcon_research child.jpg

Tracking down TB

A Nigerian Case Study

Epcon’s influential intervention in Nigeria has helped transform the West African nation’s approach towards locating missing cases of Tuberculosis (TB). Nigeria is one of the 14 high burden countries for TB, HIV-associated TB and drug-resistant TB. This means that the illness is given the highest priority at a global level. Nigeria is ranked seventh among 30 high burden countries and second in Africa; the problem of TB is exacerbated by HIV/AIDS and drug-resistant TB, and by undiagnosed cases. In fact, Nigeria contributed 4.4% of the global “missing” - i.e., undiagnosed - TB cases in 2019.

Locating undiagnosed TB cases in Nigeria, as well as Drug-Resistant TB, is imperative. We aim to find the maximum number of unwell patients, some of whom don’t even know they are carrying active TB, so that they can be connected to care. 


The EPCON model revolves around creating a digital representation or a digital ‘twin’ of the disease situation in the country. By combining real world contextual data and evidence from programmatic data, we generate an ‘Epidemiological Twin Model’ that provides insights about the present situation of the disease and its progression in time and space in a real-time, continuously evolving world. The model is based on highly sophisticated and powerful Bayesian network. Epcon’s model currently covers four key Nigerian states: Lagos, Oyo, Osun and Ogun.

EPCON_Nigeria screened versus prediction.png
Discovering new hotspots

Routine health surveillance data, from health facilities and community workers in the field, is combined with other open source data of known risk predictors for TB in order to build the AI based TB twin model.

We are then in a position to predict and map TB risk. The model predictions are communicated in the form of digital dashboards and geoportals, where one can see mapped data. It’s much easier to visualise the predictions in this way than just seeing tables and figures. And, importantly, EPCON identifies new hotspots, even in far-flung, hard-to-reach areas.

 

Granularity

One of the biggest advantages of the TB twin model is that it allows for high resolution predictions. Because the prevalence of TB is hard to calculate, the relevant health authorities usually only have access to national or state level estimates, which are generally calculated via surveys. Beyond that estimate, you won’t know more. With the predictive model, we can add granularity or high resolution - we can really extrapolate figures in very small areas or population clusters. With more zoomed-out data,  TB cases could be missed, so this obviously helps with prioritising intervention.

EPCON_Nigeria TB case
Real-time prediction capacity

The disease risk modelling in Nigeria is not without its challenges. EPCON’s model is capable of providing predictions and updates in real time. However, in some regions of Nigeria, either the partners don’t have the digital data capturing tools or are not being utilized as commonly (though Epcon can provide these tools) to capture real-time program data. Thus, there is frequently a delay in the update of new data into the system - up to two months on occasion. But even making predictions with delayed data is meaningful, because TB is a disease that extends over a period of months - sometimes without any symptoms for long periods of time. This means there is still some time to locate potential cases.

The predictive analytics model developed by EPCON is highly scalable. This scalability extends both geographically - potentially to other at-risk areas, such as South Africa or South East Asia - and to other diseases, including malaria or dengue.

Epcon’s mission - to use technological innovation to improve health for all - has led to the development of the Epidemiological Twin Model, a powerful data processing and analytic platform for finding hotspot locations. Ultimately, this targeted intervention will fuel Nigeria’s efforts in its fight against the pervasive enemy of under-diagnosed TB.

bottom of page