We'll discuss recent applications of deep learning to improve disease management by anticipating disease onset in chronic, recurrent diseases. Many chronic diseases such as epilepsy, schizophrenia, or multiple sclerosis are characterized by a biphasic trajectory. In epilepsy, for example, patients only spent about 0.05% of the time in an actual seizure while they are symptom-free the rest of the time. However, the current unpredictability of these symptomatic episodes negatively impacts how these diseases are managed and treated. We'll argue that deep learning along with the increasing availability of large-scale medical datasets offers the unprecedented opportunity to dramatically improve how these diseases are managed. Specifically, we'll provide an example of how deep learning was used to predict the occurrence of epileptic seizures, something that was not considered possible until recently. More generally, this approach exemplifies an important novel use of deep learning in healthcare: to predict the occurrence of symptomatic episodes and thereby drastically improve disease management. We believe that the applicability of this approach will become even more important with more medical datasets becoming available in the future. We'll end by discussing how this approach can be extended to other diseases and may provide an unprecedented opportunity to transform the way we manage chronic, recurring diseases.