Over the last two years, the COAST project aimed at improving the health scope in Uganda by conducting research using Artificial Intelligence (AI) and Data Science. The focus was to use this information and outputs to aid the Ugandan government and communities to better understand and fight highly infectious diseases like COVID-19 and future pandemics affecting Uganda. This decision was guided by the challenges faced by health officials in conveying accurate health-related information to the public due to limited data, which caused panic and fear across the country. To support this, the team embarked on uncovering the transmission dynamics of the virus in Uganda.
The project was guided by 3 main objectives, namely:
These objectives were further broken down into four workstreams:
Figure 1. The four work streams under the COAST Uganda project
Working with AI and Data Science to uncover the transmission dynamics of COVID-19 in Uganda was a unique experience for the six teams involved and who were responsible for various components of the project, including the chatbot team, radio data team, modeling team, and air quality team. For the chatbot team, constructing a chatbot, especially one that utilizes a digitally under-resourced language, was a pioneer effort and has involved continuous learning on-the-go. Although the primary tools, such as the chatbot support platform (RASA) and the language model frameworks already existed, their coherence, assembly and alignment into functional chatbots within the project’s context was largely novel and required many improvisations.
Photo 1. Grace Kebirungi from the Chatbot team illustrates how the platform works. (Photo credit: Hilda Mirembe, COAST Project Uganda)
The timing of the project also presented challenges that required quickly adapting our approaches and making decisions with limited information about COVID-19. The radio data team faced a considerable challenges related to time-sensitivity as new COVID-19 variants came out and lockdowns and vaccination drives were implemented. As a result, we had to process and analyze much more data within a shorter time frame to keep up with the pace of COVID-19 events as they occurred in the country. However, through collaboration and teamwork, we managed to take the pulse of such events.
The modeling team used both mathematical models (deterministic compartmental models) and statistical models (dynamic regression models) to predict the transmission dynamics of COVID-19 in Uganda. These predictions were used to inform public health authorities to support decision making with regards to employing mitigation strategies including community or country borderline-based measures aimed at stopping the spread of the disease. Relevant data from other work-streams was also aggregated to be used in statistical modeling and attaining parameters for use in mathematical modeling.
The air quality team used multivariate analysis and linear models to determine the association between air quality and human mobility. The team also fitted Generalized Additive Models (GAMs) to capture non-linearity in the data. We compared the model fit for the linear models and the GAMs using the ANOVA test. The results of ANOVA tests comparing linear models with GAMs provided statistical evidence to suggest that incorporating nonlinear relationships of the mobility variables did not improve the model. Our findings suggest that air quality data closely mirrors human mobility data and could thus be used as a proxy to human movement patterns in cities and countries. The infectious disease control programs could thus leverage air quality data to study transmission patterns for infectious diseases and in the process implement control measures without necessarily compromising on the privacy of individuals plus other limitations and functions associated with the use of GPS data from mobile phones.
One of the successes included the development of mathematical and statistical modeling frameworks that can be adapted to other infectious diseases and the aggregation of data-sets to be used for modeling purposes. With this framework, we were able to incorporate underlying epidemiological data concerning the natural history of COVID-19, such as the incubation and infectious periods, transmission by individuals not showing symptoms, and the relationships between the different modeling compartments. The radio and data teams managed to get great insights on community perceptions as they worked directly with different communities and genders which was an exciting experience for everyone involved.
"Over the past two years the COAST project has built multidisciplinary research teams from computer sciences, statistics, health, and engineering that have come together to leverage the potential of AI and data systems for COVID-19 management and response. We now have tangible products including an AI for COVID chatbot system, language technology for radio broadcast data analysis, machine learning-based diagnostics for lung ultrasound images, decision support systems, modeling and forecasting tools, and air quality risk analysis for COVID-19. Together these form a strong foundation of AI for health challenges" Engineer Bainomugisha, Principle Investigator, COAST Project Uganda.
Photo 2. A visual representation of the flow of intents and how they are interlinked in Botpress, a chatbot development framework. (Photo credit: Hilda Mirembe, COAST Project Uganda)
As part of the coast project, multiple outputs have been produced. The air quality team published a journal article titled, “Air Quality Data as a Measure for Human Mobility: Evidence from Two Ugandan Cities during the Covid-19 Mobility Restrictions”, which was published by the Environmental Science and Pollution Research [ESPR] Journal on 15th December 2022.
The modeling team also published a manuscript in the PLOS Global Public Health journal titled, “Community Purchase of Antimicrobials during the Covid-19 Pandemic in Uganda”, which underscored the extent of self-medication and potential antimicrobial resistance in Uganda during the COVID-19 pandemic.
The chatbot team has produced a separate English and Luganda COVID-19 Chatbot and also put in place a gender sensitive and equitable Uganda-specific COVID-19 information database.
The COAST project is being implemented with grant support from Canada’s International Development Research Centre (IDRC) and the Swedish International Development Cooperation Agency (Sida) as part of the Global South AI4COVID Program. To learn more about our work follow us on Twitter and visit our website.Tags: COVID-19, AI4COVID, AI