Applying Machine Learning and Application Development to Lower Back Pain and Genetic Medicine

Abstract

Improvement of the healthcare system is a focal point for academic leaders. In recent years, precision medicine initiatives have gained traction as a solution to improve care by leveraging healthcare analytics and informatic tools to assist clinicians in prescribing individualized treatments based on the patient’s health characteristics. This involves data collection, data management and advanced statistical and machine learning methods, and new tools to deliver the promise of data on the outcomes of health and healthcare. To help clinicians, researchers must leverage electronic health record (EHR) data, however these data are complex as they are made up of multiple modalities with an ever increasing volume. While structured EHR data is a popular modality to use for analysis, clinical notes (i.e. unstructured EHR data), for example, provide more granular information about patients that is useful to clinicians. As a result, there is interest in building cohorts of patients based on unstructured data by using natural language processing (NLP). For analysis, there are recent works that discuss the value of using deep learning to integrate multiple data modalities together to better predict clinical outcomes; however, rigorous testing is needed to fully understand this value. Once data has been collected and analyzed, the final task is understanding how to further patient involvement with this information. In this dissertation, I focus on creating a framework that can build cohorts based on unstructured data, analyze EHR data using the different modalities, and increase patient involvement. The aims are to: 1) compare NLP methods for the classification of lumbar spine imaging findings related to lower back pain, 2) predict decompression surgery by applying machine learning to patients’ structured and unstructured health data, and 3) demonstrate patient delivery and sharing of data in a smartphone app to facilitate family communication of genetic results.