Predictive AI model maps high-risk TB areas in CAR's Bangui at community-level precision

Despite being a high TB burden country, with an estimated incidence rate of 540 per 100,000 in 2020, the CAR continues to rely heavily on passive case detection and notification from limited health facilities. In Bangui, only 14 TB clinics serve nearly one million residents, leaving critical gaps in coverage, particularly in the city’s north and eastern zones.

A new artificial intelligence-powered predictive mapping model has uncovered previously undetected tuberculosis (TB) transmission hotspots in Bangui, Central African Republic (CAR), offering a high-resolution spatial blueprint that could transform how public health interventions are deployed in the country’s capital.

The research, published in Tropical Medicine and Infectious Disease with the title "Leveraging Artificial Intelligence to Predict Potential TB Hotspots at the Community Level in Bangui, Republic of Central Africa," was conducted by a team of international and local researchers from The Union, EPCON, and the CAR National Tuberculosis Program (NTP). Using Bayesian modeling and geospatial data, the team identified high-risk neighborhoods underserved by current surveillance systems, raising urgent questions about resource distribution, notification discrepancies, and data-driven interventions.

How can AI improve TB surveillance in a fragile health system?

Despite being a high TB burden country, with an estimated incidence rate of 540 per 100,000 in 2020, the CAR continues to rely heavily on passive case detection and notification from limited health facilities. In Bangui, only 14 TB clinics serve nearly one million residents, leaving critical gaps in coverage, particularly in the city’s north and eastern zones.

The study's AI model attempted to close this gap by creating a 100x100-meter resolution spatial grid of TB risk, using a combination of population data, clinic locations, health service accessibility, and social determinants of health such as poverty, sanitation access, literacy, and vaccination coverage. Researchers calculated catchment areas for each clinic based on 20-minute travel radii, then layered in clinic-level notification data and residential diagnostic data for the years 2021 and 2022.

From this, a digital epidemiological map emerged, pinpointing neighborhoods with high TB positivity rates, especially in regions that are remote, crowded, or poorly connected to clinics. This granular approach revealed major mismatches between officially notified cases and predicted TB burden, a sign that passive surveillance may be missing critical transmission clusters in underreported or inaccessible communities.

Where are the true TB hotspots, and are clinics reaching them?

Of the 14 clinics included, only two were situated within what the model defined as high-risk zones. Seven were misaligned, operating in areas where either predicted or notified TB positivity sharply diverged. Notably, the Hospital Communautaire was located in an area flagged by the model as low-risk, yet reported a high number of TB-positive notifications, suggesting possible inflow of patients from distant areas or inconsistencies in diagnostic coverage. In contrast, several facilities such as Obrou Fidel Camp and Amis Afrique ONG, operated in zones where AI-predicted TB rates exceeded actual notifications, hinting at underdiagnosis or missed outreach opportunities.

Across Bangui, high-risk areas were characterized by greater population density and more intense night-time light emissions, a proxy for urban activity. These indicators showed moderately strong positive correlations with both notified and predicted TB rates. However, the model also identified statistically significant negative correlations with HIV prevalence, child undernutrition, and stunting - factors generally expected to increase TB risk.

This paradox, the researchers say, points to structural underreporting in areas marked by severe poverty and poor access to healthcare. Residents in such zones may not seek care due to distance, stigma, or inadequate services. As a result, high-risk regions may appear deceptively low-risk under conventional notification-based systems, reinforcing the need for active case-finding and targeted outreach informed by predictive analytics.

What are the implications for public health and TB strategy in CAR?

The model’s greatest utility, according to the researchers, lies in its potential to refine CAR’s 2024–2028 TB control strategy. That plan aims to increase case detection in low-notification districts from 45% to 90%, expand molecular diagnostics, and raise the notification rate of multidrug-resistant TB (MDR-TB) from 19% to 90% by 2028. By integrating high-resolution risk maps into planning, health officials can prioritize active case-finding in overlooked districts, allocate diagnostic tools more effectively, and identify clinics in need of reinforcement or relocation.

The authors caution, however, that these predictions must be validated through field studies and time-series comparisons with actual case trends. They acknowledge that assumptions such as uniform city-wide TB incidence and reliance on outdated population grids, introduce uncertainty. Conflict-driven migration, informal settlements, and changes in health-seeking behavior may skew spatial estimates and reduce model precision in some areas.

Even so, the Bayesian model provides a robust starting point for what the authors call “adaptive TB surveillance”—a data-driven system that can respond in real time to new information and shifting transmission patterns. The model is also designed for expansion, capable of incorporating genomic data on drug-resistant TB strains and evolving through continuous integration of new programmatic data.

Further enhancements may include Latent Class Analysis to identify vulnerable subgroups, improved spatial stratification to capture micro-level variability, and hybrid evaluation techniques that mix traditional AI metrics with geospatial performance indicators.

While AI is not a replacement for robust surveillance infrastructure, the researchers argue it is an indispensable tool for countries like CAR, where health systems are overburdened and epidemiological data are scarce. Predictive mapping can inform policy, drive equity in service delivery, and guide the efficient use of limited resources.