top of page

Artificial Intelligence in Epidemiology: Opportunities and Challenges

CEO Caroline Van Cauwelaert enjoyed the opportunity to present at the Flanders AI Academy last week. She spoke about the opportunities and challenges facing Artificial Intelligence (AI) in epidemiology with reference to EPCON’s vital work in countries including Nigeria, South Africa, India, the Philippines and Pakistan.

Van Cauwelaert explained EPCON’s role in helping local authorities take measures, informed by the use of AI, against tuberculosis (TB). The EPCON team, consisting of machine learning architects, bioinformatics and medical experts, as well as public health specialists, combines different sources of data - health records, social-demographic and contextual data - to estimate TB disease burden, predict its evolution and determine the effect of interventions.

TB extracts a huge cost from many developing nations, particularly India, Indonesia, China, Nigeria, Pakistan and South Africa, which together account for 60% of the total TB cases worldwide. Around 40% of the 10 million people who develop TB each year go undiagnosed. With the possibility that one TB patient can infect 10-15 others, case detection is of vital global importance in curbing the spread of this disease. Also, if left untreated, TB is fatal in two-thirds of cases - but it is curable if caught and treated. Therefore, finding missing, or undiagnosed, TB cases is one of the biggest health challenges, especially in high-burden countries.

EPCON provides its AI solution to two types of clients: (1) NGOs whose healthcare workers make use of the Epidemic Control Platform to organize and plan their screening activities, and (2) pharmaceutical companies that need to understand where the disease burden is highest, so they can plan their supply chain and go-to-market strategies.

Utilising AI in epidemiology, as Van Cauwelaert explained at the Flanders AI Academy, is not without its challenges.

These include:

  1. Limited data sets from clients - the larger the data set, the better the training material for the AI algorithm.

  2. Poorly captured data - e.g. no uniform naming conventions for cities or streets, misspellings, columns with missing values.

  3. Data uploading issues - if data is shared inconsistently (not regularly), then changes in the disease dynamics can be missed. Also, detailed and specific data is sometimes lacking.

  4. Data collection - when it comes to screening for TB, various methodologies, from age screening to door-to-door screening, are used, which makes comparisons tricky.

  5. Data mapping - georeferencing (i.e. determining the precise location) of cases is not always clear, and maps of some areas need to be subdivided in artificial ways - for example, polygons, or statistical units, are purposefully created to improve the resolution.

  6. Psychological resistance - healthcare workers can be reluctant to capture data digitally. Also, when AI tech is unknown to these workers, it can be under-appreciated and viewed with some suspicion.

Despite these challenges, Van Cauwelaert emphasised the immense benefits of AI in epidemiology. EPCON’s predictive analytics model has significantly raised the case detection rate, with an increase in identified cases ranging from 50% to 200%.


bottom of page