Electronic medical records are a rich source of longitudinal, individualized medical data. The wide use of these resources allowed the development of deep learning methods to better understand disease progression and predict disease onset. While these methods offer impressive accuracy they lack in interpretability, since it is often hard to tell why a specific person is predicted to develop a disease.

At K, we strive to develop data-driven interpretable methods that will help us understand their medical predictions. Here, I will present how we exploit longitudinal patient-specific data and deep learning methodologies to predict the onset of Ischemic Heart disease. I will further focus on how these temporal events can be used to provide medical insights at both the individual and population levels.