Using Digitized Data To Create Decision Support Tools, Find Life-Saving Signals Is Focus Within Park Lab

Using Digitized Data To Create Decision Support Tools,
Find Life-Saving Signals Is Focus Within Park Lab

There are several ways to recognize a patient’s downturn within the ICU, ranging from clear physical changes to specific monitor alerts. Sometimes, the signal might be hidden within the aggregate of collected bedside data, and finding it early can often have dramatic effects on hospitalized or critically ill patients.

Dr. Soojin Park, Associate Professor of Neurology (in Biomedical Informatics) at Columbia University, is focused on the development and clinical evaluation of innovative decision support tools to find that signal and improve patient outcomes. This research has found a home within the Columbia University Irving Medical Center (CUIMC), where these tools are being developed and evaluated within her lab.

“Using digitized data is a way to extract more complex information from patients,” Park said. “We are working to input that data into a decision support tool to inform patient care, and to help the clinicians, who are incredibly busy and can be reasonably distracted by clinical care, so that they can be able to recognize when something bad is likely happening to a patient.”

Neurology is Park’s specialty — she continues to practice in Milstein Hospital — and while this research was brain-specific at the start, her expertise in digitized and physiological data has enabled her to expand this work into numerous other areas; her lab has ongoing projects including COVID, pediatric ICU, and neurological critical care, and it has expanded into the emergency room.

“This field is considered predictive analytics, so you can understand how this would be useful as a unit-wide tool,” Park said. “We have real need in the units where patients can’t be monitored as closely as they could, or when they need to be in a highly monitored location, but can’t because of a nursing shortage or other reasons, like the pandemic. We want to build tools for those environments as well.”

Both PhDs and MDs collaborate on this research within The Program for Hospital and Intensive Care Informatics (PHICI), a translational research program within the Department of Neurology that provides the framework and pipeline to enable clinicians, scientists, and administrators to maximize use of monitoring device data of the heart, lungs, and brain for predictive analytics.

Using digitized data is a way to extract more complex information from patients. We are working to input that data into a decision support tool to inform patient care, and to help the clinicians, who are incredibly busy and can be reasonably distracted by clinical care, so that they can be able to recognize when something bad is likely happening to a patient.

Soojin Park

“Our trainees work together in a team-science setting to combine data science with deep knowledge of clinical domain to build and design questions that are useful for patient care,” said Park, the director of the program. “We try to find a critical point in patient care where we are driving blind, and develop a hypothesis that might be in data that isn’t being tapped as well as it could. That’s when the multi-disciplinary work starts. You have a very specific, useful clinical question, and then you work on taking that hypothesis and using the data to set up an experiment where the data can try and answer the question of whether the patient is in this category or that category.”

These decision-support tools are not currently being used in critical care, but they have been developed and, using CUIMC resources, are in the silent clinical validation process so they don’t affect patient care. The validation process analyzes, among other questions:

if the signals are being detected earlier than the clinicians found them?
are they accurate?
if possible to know, would this early detection have affected the patient’s outcome?

Her research began while Park was practicing at the University of Pennsylvania, and progressed when she was recruited to Columbia in 2014. Her lab has developed several complex models based on the belief that there was a signal in systemic physiologic markers, as well as ones derived from the brain. The lab continues to test different methods, and has found success combining data-driven and domain-driven approaches.

The informatics focus in this research has been bolstered by Park’s affiliation with DBMI, which became official last year through a joint appointment, but has been an important asset since she joined Columbia.

“DBMI was a very well established informatics group that was very clinically informatics-focused,” Park said. “It was one of the reasons I was attracted to come here.”

Park collaborated with Noémie Elhadad early in her Columbia tenure to receive an NIH K01 Career Development Award in Biomedical Big Data Science, and she recently collaborated with DBMI’s Gamze Gürsoy for the Data Science and Health Initiative (DASHI) grant: “Privacy-preserving Detection of Delayed Cerebral Ischemia using Federated Learning with Differential Privacy.”

She worked with the DBMI Summer Research program during the summer, a program which introduces both undergraduates and high school students to the field of biomedical informatics, and she is excited to continue collaborating with both faculty and trainees in the department.

“There is obvious cross-collaboration that can occur when you share PhD students,” Park said. “I’m looking forward to sharing some of my knowledge with DBMI students who might be interested in the bridging to the clinical side.”