Abstract

Understanding patients' journeys in healthcare is an important task for a wide range of downstream applications such as prediction of surgery, prediction of ER admission, prediction of readmission, recommendation of the next best steps etc. This task aims to learn a dense representation that is able to encode hidden dependencies among medical events as well as the time gaps between them. A patient journey is a sequence of medical events over time that can be represented at three levels; patient, visits, and medical codes (diagnosis, procedure, drug). One of the main challenges of patient journey understanding is to design an encoding mechanism that is able to effectively capture the multi-level structure as well as the temporal information. Inspired by BERT, we propose a deep neural network architecture that can simultaneously capture the contextual and temporal information of patient journeys.

About the Speaker

Ebad Ahmadzadeh is an FIT graduate (2018). His PhD dissertation title is Data Mining Algorithms for Decision Support Based on User Activities. His current research focus (@ Cohere Health) is on NLP and Machine Learning for healthcare. He participates in project design, technical direction, research and development of different AI solutions mainly related to the problem of prior authorization, a process in which health insurance companies evaluate and authorize service requests in advance which has been a highly manual process traditionally. Beside the time saving for the insurance companies, the AI solutions help patients across the USA get faster access to services they need, and providers and medical doctors have a more comprehensive and summarized understanding of their patients so they spend more time with the patients rather than looking for patient information in documents.