Although chronic diseases such as coronary artery disease are typically seen as characteristic of more affluent and developed nations, the disadvantaged and marginalized residents of these countries share the highest rates of disease incidence. Coronary heart disease, one of the greatest killers in the United States, costs taxpayers over $3 billion a year in healthcare costs, despite the high possible for disease prevention. The goal of our program was to use machine learning to recognize high-risk areas within each state that could benefit, both via reduced economic strain and improved patient outcomes, from targeted medical outreach. Our application's goal was to better inform medical providers and healthcare policy makers, but also to allow patients to feel empowered to take their health into their own hands.
What it does
Our web app uses machine learning to predict the likelihood of coronary artery disease occurring in any county following input of demographic information such as median income, educational attainment levels, or percent of residents living in poverty. This output is provided to the user as a probability statistic. This information can be used by hospitals to initiate outreach initiatives, such as nutritional health and wellness courses, based on the necessity of the service within a specific county. Additionally, this information can be used by local and state-level policymakers to determine resource allocation necessary for different regions. The wealth of healthcare data publicly available is not functionally useful without a coherent output, and DataWell is able to generate meaningful insights into recommended healthcare actions.
How we built it
Using publicly available data from the U.S. Census, Behavioral Risk Factors Surveys, and other medical data sets, we created a compilation of demographic data relevant to likelihood of coronary heart disease diagnosis. We then used Python and libraries like Numpy, SKLearn, and TensorFlow to create a neural net and machine learning program where an input of demographic data predicts the likelihood of disease probability. We created a mockup of an app in Swift to visualize what we hope to achieve with this data, which is to encourage data-driven work for social good.
Accomplishments that we're proud of
Incorporating machine learning, a topic none of us were originally familiar with, proved to be the most challenging part of our project. After considering different ways to use artificial intelligence, especially considering the platforms we had access to, incorporating machine learning and building a neural network (with TensorFlow) was our proudest accomplishment due to the actionable results it produced.
What's next for DataWell
Although our current interactive results screen only features counties in Illinois, our predictive data analytics are capable of generating actionable insight into risk of heart disease and other fatal chronic illnesses throughout the country using the same neural network technology. By expanding the scope of DataWell o incorporate more variables that influence health outcomes, such as crime rates, population density, and access to nutritious food, DataWell can become an even better predictor of coronary artery disease. The predictions for likelihood of disease generated by the neural network can also be expanded to include historical economic consequences of different actions taken by policymakers to generate a cost-benefit report that varies across county by disease likelihood or severity.