On January 21st we were excited to host Noémie Elhadad, Associate Professor of Bioinformatics at Columbia University. Her research focuses on ways in which large, unstructured clinical data sets (e.g. patient records) and health consumer data (e.g. online health communities) can be processed automatically to enhance access to relevant information for clinicians, patients, and health researchers alike.
One of the major challenges facing the medical system today is the way information about patients is stored and processed. Physicians have access to all of their patients’ records through Electronic Health Records (EHRs), which store a wealth of information about a patient’s entire interaction with the medical system. For the most part, all of this information is hardly digestible unstructured text organized in chronological order and results in information overload for anyone trying to access it.
Previous attempts to build tools and dashboards which summarize EHRs for clinical use have always either focused on a specific disease (making it hard to apply to patients with multiple conditions), on a specific setting (like the ICU), or chose to completely ignore the text portions of the EHRs (since doing NLP was too hard).
Elhadad’s team is working towards the optimal way to extracting relevant information from the EHRs. The team built a tool for the New York Presbyterian Hospital that synthesises the information found in EHRs into an easily accessible format by using domain knowledge, NLP, and data science.
Elhadad approached this project by shadowing physicians to see how they interact with the EHR. One key finding was that physicians look at a patient’s history longitudinally, meaning that they focus on the patient’s current condition through history as opposed to reviewing a patient’s health record chronologically. The tool the team built, HARVEST, mimics this behavior. It extracts content from EHRs, aggregates information across multiple care settings, and visualizes the key parts in an interactive, intuitive, and most importantly, longitudinal way.
Going forward, Elhadad hopes to add a prediction model of disease progression. By incorporating EHRs, clinical trials, and other sources, she wants to be able to predict which patients are more at risk of progressing faster in their disease. If this or any other parts of Elhadad’s research spark your interest, you’re in luck! She’s hiring right now, so feel free to reach out to her!
If you’re interested, you can download the slides from the presentation here. Thank you to About.com for hosting us and providing food and drinks!