We’re thrilled to announce our recent breakthrough, as detailed in our article published in Tropical Medicine and Infectious Disease.
Our innovative AI-driven hotspot mapping has revolutionised community-based Active Case Finding (ACF) in South-Western Nigeria, achieving remarkable success over the past three years, from 2020 to 2023.
In the fight against TB, a staggering 40% of cases traditionally remain undiagnosed or unreported, posing a significant challenge in controlling the spread of this disease. With Nigeria being one of the top five countries burdened by TB, the need for more effective strategies is urgent and clear. That’s why we are overjoyed that our Bayesian predictive model has successfully identified a significantly higher number of cases in high-risk areas. Read on to find out how we’re making huge strides towards a healthier future in low-resource areas.
How do we detect TB?
To find undiagnosed TB cases, community-based interventions and facility-based screenings are usually carried out. However, these methods require significant resources. There is a pressing need for more innovative and efficient approaches. In addition, any interventions also need to take regional conditions into account.
Keeping this in mind, we used data and technology for evidence-based decision-making to improve TB control. Our article explains how we developed a TB risk predictive model that works at the community level. We also assess how this model affects the success of TB Active Case Finding (ACF) interventions and compare the effectiveness of strategies in areas predicted to be TB hotspots versus those that are not.
Collaborating closely with regional teams
To secure local buy-in, we collaborated closely with the team in Nigeria. We added value by:
Assisting with geo locations for community screenings These details are crucial for targeting interventions accurately using predictive analytics. In the beginning, the regional teams added these coordinates using Google Maps. However, this approach is not ideal. Locations can be difficult to pinpoint accurately, and errors are possible. We, therefore, developed a method of integrating Telegram functionality into the data entry process, allowing field workers to input data directly by sharing their location automatically.
Helping to secure monthly validated dataSince data was manually recorded on paper, updates often faced delays. Such delays and gaps in data are not ideal when building predictive models that rely on timely and accurate information. For the model we used monthly validated data.
High success rate
To combat TB more effectively in Nigeria, we divided Wards – the smallest administrative units – into smaller population clusters, with a population of approximately 10 000 people, using a machine learning algorithm. We then linked these clusters with TB screening results from community screenings. By merging this data with high-resolution contextual information, we trained a Bayesian inference model to predict TB hotspots within these communities. These predictions were then displayed on our Epi-control platform , enabling regional teams to identify high-risk areas and plan more targeted health interventions.
To evaluate our model, we contrasted the yield – defined as the ratio of individuals with TB to those screened – between areas predicted as hotspots by the model and those deemed non-hotspots. Our findings confirmed the model’s high accuracy; TB detection rates in predicted high-risk areas were 1.75 times higher than in other locations across all four states in South-Western Nigeria.
This success indicates that our Bayesian predictive model can significantly improve health interventions by accurately identifying where undiagnosed TB cases are likely to occur. A targeted approach of this kind leads to better resource allocation within the community, enhancing overall public health efforts. Please take a look at our paper for further details.