top of page

Using AI for low-data areas: finding TB hotspots in Bangui

We were delighted that Dr Sumbul Hashmi, Global Public Health Manager at EPCON, presented at the Union World Conference on Lung Health in Paris. The theme of the conference was about transforming evidence into tangible practice. Dr Hashmi spoke about our novel proof of concept approach for predicting TB hotspots at the community level in Bangui, the densely-populated capital city of the Central African Republic. This project was initiated by The Union, and supported by the members of the NTP. The country faces a significant TB burden, with an incidence of 540 per 100,000 inhabitants.

Bangui: limited data

Bangui has population density of of 17,094 inhabitants per square kilometre and 14 TB diagnostic and treatment centre, featuring two referral laboratories:

Locations of TB Diagnostic and Treatment Centres in Bangui, Central African Republic

However, there exists a significant issue of underreporting and underdiagnosis of TB cases, as the available data is limited to hospital or facility-level records. Because there is an influx of health-care seekers from neighbouring regions, this data provides a skewed perspective on TB burden. In consequence, our objective was to identify underserved areas at the highest possible resolution, down to community level, in order to optimise TB services.

To address this challenge, we employed two approaches. These methods involved the collection of various datasets, such as the total number of TB clients registered at each facility, as well as open-source data covering socio-demographics, local context and national incidence estimates. Local contextual data included demographics, literacy, basic vaccination coverage, access to improved water and sanitation, nutritional status in children, access to road networks, and more. Subsequently, we harnessed our Bayesian inference model to estimate missing TB cases within designated population ‘tiles’, defined as 100x100 metre areas.

Discovering hidden hotspots

The result of our approach was the identification of discreet pockets with the potential for high TB risk. These pockets would have remained concealed had we solely examined aggregated data at the facility level. This discovery enables us to implement precise community-based interventions and allocate resources strategically to facilities located in areas with a heightened risk of TB transmission.

Navigating challenges

Undoubtedly, our work on this project was not without its share of challenges, which encompassed:

  • Coping with limited data, necessitating the reliance on assumptions - fortunately, room for improvement exists over time.

  • A lack of confirmatory data to validate outcomes, yet qualitative assessments are currently underway with the input of local experts.

Nevertheless, in the face of these challenges, the project showcased numerous strengths, including:

  • High-granularity outputs, enabling precise and targeted interventions within communities.

  • The potential for real-time model updates, leading to enhanced TB surveillance.

  • The capacity to visualise data on a mapping system.

  • Data-driven decision making rooted in local contextual information.

In conclusion, we were very excited to share our data and insights from this successful project with delegates at this year’s Union World Conference.


bottom of page