Using Deep Learning with Canadian Primary Care Data for Disease Diagnosis

The majority of Canadian primary care systems record patient data in the form of Electronic Medical Records (EMR). EMRs hold structured, semi-structured and unstructured demographic and health care data about patients. The value of EMR data for research, health surveillance and quality improvement continues to be explored. Data analytics such as Machine Learning (ML) and statistical modeling techniques have been applied to de-identified EMR data repositories to advance our understanding of different health conditions and patient care. More recently, the application of Deep Learning (DL) approaches to structured, semi-structured and unstructured data of the EMRs is being investigated as an avenue for improved identification of health conditions. Supervised ML methods have dominated disease classification for more prevalent diseases. A large cohort of labeled data is required to train ML models using supervised learning methods. For less common diseases, the amount of available labeled data is often insufficient, and a variety of strategies are being explored to deal with inadequate, noisy and missing data. This chapter describes the benefits of using DL models with EMR data for research to improve provisioning of health care in primary care settings. A few prominent DL models such as Multi-Layered Perceptron (MLP), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) are discussed with example scenarios that demonstrate application of some of these predictive analytics models to both structured and unstructured EMR data using regular and weak supervision methods for diagnosing both prevalent and non-prevalent diseases.