CARE-Africa

The challenge

Diarrhoeal diseases remain a common cause of ill-health and death in sub-Saharan Africa, particularly among children. In addition to their immediate health impacts, diarrhoeal diseases have multiple long-term consequences, affecting growth, development, schooling and economic prospects in adulthood. 

Management of diarrhoeal diseases is a major challenge. A multitude of different viruses, bacteria and parasites can all cause diarrhoea, and identifying the underlying cause of symptoms can be difficult. Outcomes can vary widely, too – some infections rapidly self-resolve while others may progress to severe disease with a risk of death. Again, it is difficult for clinicians to judge which children are most at risk.

The project

The CARE-AFRICA project is developing and evaluating an AI-based tool to guide clinical decision-making, based on analysis of patient data entered by a clinician and automated retrieval of regional and national data on other factors likely to affect the risk of diarrhoeal diseases, such as environmental, climate, socioeconomic and demographic data. The tool is designed to integrate these data sources and provide a clinician with a likely diagnosis and a recommended treatment. 

In addition to presenting the probabilities of infection and co-infection with a range of different pathogens, the tool also calculates the likelihood that the infection will be resistant to particular antibiotics, which informs its treatment recommendations. It also provides a clinician with an explanation of its results and recommendations. Results will be generated within 5 minutes in healthcare environments, and the system can retain information offline until an internet connection is established (for use in more disadvantaged areas). 

To develop the tool, the CARE-AFRICA team will train AI agents on multi-scale and multi-modal data sources, so they learn the associations between a wide range of variables and different causes of diarrhoeal disease and types of drug resistance. To underpin this training, the project is collating historical data from Ethiopia and Uganda and collecting new data from a set of sentinel sites in contrasting locations. The tool’s interface and functionality are being developed in close collaboration with end-users. It is being designed to run on a tablet device.

Once trained and validated, the tool will be initially tested in a pilot study in eight facilities in Ethiopia and Uganda. A revised version will then be tested more extensively across 40 facilities, with results obtained for approximately 4,700 children.

In a second strand of the project, the team will develop models based on the training data to provide dynamic predictions of diarrhoeal disease risk. This will give policymakers a tool for exploring how case numbers might be affected across different scenarios, for generating risk maps to target interventions, and for exploring potential impacts. Community consultations will be organised to gather data on community-level risk factors for diarrhoeal disease and barriers to care, which will also be incorporated into the model. 

Impact

The CARE-AFRICA project has the potential to radically change the assessment of diarrhoeal disease cases in children in sub-Saharan Africa. It will:

• Create a sophisticated clinical decision-making support tool that requires minimal input from clinicians yet draws on a wealth of data sources to make assessments of the likelihood of different disease aetiologies.
• Provide an immediate assessment of the likelihood of antibiotic resistance, without specific susceptibility testing, which typically takes days to deliver results.
• Reduce antibiotics’ misuse by matching antibiotic use to the likely susceptibility profile of an infection.

If validated, the new tool could ensure that children are more likely to receive an appropriate treatment for diarrhoeal disease – speeding up recovery times, reducing unnecessary use of antibiotics and thereby relieving pressures driving antimicrobial resistance, and optimising use of health system resources.