Applications of a Deep Learning Model for Clinical Optimization and Population Health Management
Current deep learning applications in health care tend to focus on natural language processing and computer vision using unstructured data. However, building deep learning models on structured data, such as administrative insurance claims, has received far less attention and holds untapped potential. In this talk I will discuss real-life applications of predicting two costly clinical outcomes from health insurance claims data: the likelihood of future back surgery and kidney failure. To accomplish this task with deep learning sequence models, I will cover three methods to improve model performance: i) embedding medical codes for input to the model, ii) transfer learning from a pre-trained general language model to improve model performance on small, context-specific data sets, and iii) using an attention mechanism to make neural networks more transparent. These methods have been implemented to train deep learning models on massive claims datasets and currently used in practice by one of the largest payers in the health insurance industry.
Janos A. Perge Ph.D. is a principal data scientist at CVS/Aetna, where he creates machine learning models for care management programs to reduce healthcare costs and improve health outcomes. Prior to his work in healthcare analytics he developed neural prostheses such as mind-controlled robotic arms for people with paralysis at Brown University. He also developed a research lab to study and modify learning by light impulses (i.e. optogenetics). He earned a Ph.D. in neuroscience from Utrecht University, the Netherlands. Feels passionate about coding strategies of live neural networks and their relevance to problems in artificial intelligence.