top of page
Search

Finding TB Cases in Pakistan: Using AI for Active Case Finding


Abdullah Latif, public health surveillance and data expert working for Mercy Corps in Pakistan, recently gave a fascinating virtual presentation to a selected number of public health experts about the cooperation between KIT - Royal Tropical Institute Amsterdam, the National TB Control Program, Mercy Corps and EPCON.


EPCON's AI model

Pakistan with a population of 231 million people, spread out over a vast geographical area, has one of the highest burdens of TB in the world. With a case notification rate of approx. 50% the challenges of allocating resources for greatest impact remain high.


The evidence indicates that it is beneficial to use active case finding - or ACF - to locate those with TB in their communities rather than waiting for people to seek health care. Digital data capturing, including location, together with predictive modelling, can lead to reduced transmission and better health outcomes. The challenge is to determine where and among which population groups to look.


In his presentation, Abdullah Latif explained about the implemented solution used by the NTP and its partners to target areas of interest and hotspots. This solution includes two major elements:

  • The electronic Case-Based Surveillance tool that generates real-time and location aware mobile chest camp data.

  • The MATCH AI Bayesian inference model analysing the programmatic data in combination with environmental context to make “hotspot” predictions.


To implement such a solution requires the digitisation of paper-based data from chest camp screenings. Having near real-time data contributes to the dynamic aspect of the AI driven predictive capacities. EPCON’s contribution is focussed on the data collection framework and methodologies, the core AI inference engine and predictive modelling, the creation of digital dashboards for data monitoring and intervention planning. KIT has brought in strong epidemiological expertise and helps coordinate and steer the project, the in-country interventions, validation and capacity building.


The real-time digital platform is able to:

  • Visualise areas of interest (vulnerability, TB hotspots, low resource settings, etc)

  • Visualise chest camp performance

  • Monitor data quality

  • Provide context and insights to help with chest camp recommendations and planning


Previously, an average of three cases were found per chest camp. This new predictive capacity brings a significant improvement: yields are in the region of just over four cases per camp now, with the potential to reach eight cases per camp - this is clearly a significant improvement.


Major challenges in implementing this project included the limited availability of programmatic TB data to inform the model, mixed digital literacy in users of the electronic surveillance tools, and patchy internet and data upload facilities in many areas.


EPCON and KIT were able to make a game-changing contribution to overcoming these challenges by re-designing the data collection tool for data to be captured in a hybrid manner: partly offline (mobile) in chest camp settings, but online (web mode), with relevant information uploaded to a server, when internet connectivity returned. EPCON also ensured that the data collection tool was designed to adapt to the Mercy Corps workflow with the understanding that user friendliness increases implementation. Chest camp data was then triangulated with other relevant contextual data to build the AI predictive model.


Capacity building has also been integral to the project, involving vital knowledge and technology transfer from KIT/EPCON to Mercy Corps. This ensures that the data collection tool, as well as the steering dashboards and geoportal, can be maintained in a sustainable manner. Such maintenance will serve to support the AI model’s capacity to continue improving case detection and program efficiency of TB in Pakistan.


bottom of page