To mark World TB Day, on the 24th March, EPCON presented findings from its South African case study at Tech Care for All’s international online conference, Clinical Updates & Innovation in Tuberculosis for 2022. The conference brought together clinicians, innovators, public health specialists and clinical scientists to share the latest and most important developments in the TB field for 2022.
The EPCON presentation highlighted the company’s work in Port Elizabeth, a city in South Africa, which is among the highest seven TB-burden countries globally. Because 57% of TB positive patients don’t report any symptoms but still contribute to the transmission of infection, it’s imperative to find these ‘missing’ - i.e. undiagnosed - cases. Undiagnosed cases pose various challenges, including:
Assessing whether areas with high notification rates are equivalent to those with high-burden areas
Determining whether it’s worth screening an area not showing high notification rates
Deciding which neighbourhoods are not worth screening
EPCON’s innovative, predictive model provides answers to these kinds of questions and assists with locating local hotspots. Instead of relying on aggregated case notifications at a state or national level, which is the way TB has historically been handled, the EPCON model combines the digital capturing of screening activities with regional contextual variables, such as climate, access to health, demographics and density. By focusing on local information, pockets of TB can be discovered at the granular level rather than focusing on high level trends. In essence, predictive modelling makes use of existing data to reveal previously unknown hotspots.
Constant Incremental Improvements
EPCON’s approach is to combine all incoming data to construct a ‘twin’, or digital representation, of the situation on the ground. Using a Bayesian network model, EPCON creates a dashboard, or geoportal, to show predicted hotspots. As more data is added to the system - e.g. further screening of patients and positive case notifications - the system improves, and as it fine-tunes itself, its predictions become more accurate.
Performance Enhancement through AI
In the first phase of the two year project, the EPCON AI predictive modelling was not yet in place. The number of presumptive TB cases at the mid term evaluation of the first year was 3,808, with 105 positive diagnoses (2.7%) which improved to 6.7% after they changed the strategy to targeted screening of contacts of index patients. After EPCON’s model was implemented, 1,640 positive TB cases were identified from 18,445 presumptive TB cases. In essence, the presumptive positivity rate increased from 2.7% to 8.9% - 4 times higher - following the introduction of EPCON’s AI-based approach.
In South Africa, in general, the cost of finding one positive TB patient and starting treatment is in the range of $9400. Remember, each presumptive TB case needs to undergo expensive further diagnostic tests to determine positivity for TB. However, costs are reduced substantially if presumptive TB cases are found in hotspots, as such people are much more likely to be diagnosed with actual TB. Because of this, EPCON was able to bring down the cost to $498 per positive TB patient diagnosed and starting them on treatment.
To conclude, using AI for case finding can greatly improve yield. It really is about searching for positive TB cases in the right places. Predictive modelling can make use of existing data to reveal hotspots not known before.