Pakistan
Predictive Modelling for Efficient TB Case Finding with Real-Time Data
Project Context
Pakistan has the fifth highest TB burden globally, and a significant number of cases go undetected each year. Traditional strategies for Active Case Finding (ACF) are often resource-intensive and based on broad assumptions or outdated data. In response, a collaboration between EPCON, Mercy Corps, KIT Royal Tropical Institute, and national stakeholders aimed to modernize TB case-finding using predictive modeling and real-time program data. The project was financed by the Bill & Melinda Gates Foundation and designed to support Pakistan’s National TB Control Program in optimizing the deployment of mobile diagnostic teams and resources.
Project Objectives
-
Predict which communities have the highest likelihood of identifying TB cases
-
Improve cost-effectiveness of mobile screening interventions (e.g., mobile chest X-ray vans)
-
Integrate real-time screening data
EPCON's Approach
EPCON developed and implemented a predictive model that integrated multiple data sources, including:
-
Historical TB case data
-
Contextual and sociodemographic variables (e.g., population density, urbanization, poverty levels, health access)
-
Real-time program data from ongoing ACF campaigns
The model generated weekly risk predictions at subdistrict (union council) level, highlighting areas with high expected case yields. These insights were shared through the Epi-control platform, enabling Mercy Corps and other partners to dynamically adjust their ACF plans.
EPCON also helped automate the integration of data from mobile screening units to continuously improve model accuracy.
Key Outcomes and Impact
-
TB case-finding yield was significantly higher in model-prioritized areas compared to conventional ACF planning
-
Number Needed to Screen (NNS) was reduced, enabling more efficient use of mobile diagnostic resources
-
Results from the pilot were published in BMJ Global Health, demonstrating the model’s accuracy and public health value
-
The approach helped validate a new paradigm for TB control: data-driven microplanning combined with digital surveillance
-
This project laid the groundwork for national scale-up and ongoing digital surveillance innovations in Pakistan
Link to publication: BMJ Global Health article - https://bmjpublichealth.bmj.com/content/3/1/e001424
AI-Guided TB Case Finding in Pakistan: The SPOT-TB Trial
The SPOT-TB trial evaluated whether an AI tool called MATCH-AI could improve tuberculosis detection in Pakistan compared to conventional site selection methods. Conducted across 68 districts in all four provinces, the stepped-wedge trial involved 30 mobile X-ray van teams that gradually transitioned from traditional camp planning — based on local knowledge and historical data — to AI-guided site selection. Over 269,000 individuals were screened between August 2023 and September 2024, making it one of the largest TB trials ever conducted.
When camps were held within 5 km of AI-recommended locations, TB case detection was 32% higher than in conventionally selected sites — a statistically significant finding. The overall (intention-to-treat) analysis showed a 7% improvement, which was not statistically significant, largely because many teams conducted camps outside the recommended zones. Qualitative findings revealed that field staff valued MATCH-AI for reducing bias in site selection, but also highlighted the importance of combining AI recommendations with local expertise.
The trial concluded that a hybrid approach — integrating AI-generated insights with the contextual knowledge of field teams — represents the most practical model for improving TB active case finding, particularly in high-burden, resource-limited settings like Pakistan.
Link to study: Read it now
Partners and Collaborations
