To more quickly and efficiently screen patients for admission into the Sequoia SN1 Patient Care Unit in Redwood City, Calif., Stanford Health Care has implemented its first home-built LLM-based application embedded in the care workflow. Nerissa Ambers, Director for Health Informatics Transformation, and Hamed Barahimi, Director for Inpatient Systems and Services, describe the project and the benefits it will offer patients.
LT: What is the screening tool all about?
Nerissa Ambers: The screening tool is designed to quickly determine if patients qualify for the Sequoia SN1 Patient Care Unit. This helps save time and improve efficiency for our clinicians. The tool identifies patient eligibility and allows for real-time screening updates as the patients’ conditions change.
Hamed Barahimi: Fully utilizing the Patient Care Unit is very important. The 24 beds in SN1 serve as a release valve for some of our capacity challenges at 300P and 500P. Identifying patients that are appropriate to be seen at SN1 is challenging as the criteria is nuanced, and it often requires manual chart review of hundreds of patients by our Case Management team.
NA: The establishment of SN1 patient care unit at Sequoia is Stanford’s first “hospital within a hospital,” which provides a great care environment for a particular set of patients. The transfer of those patients requires extensive chart review and coordination between numerous departments. The tool created automates part of a complex workflow and uses AI to scan patients’ notes and orders for relevant clinical information to help identify patients with certain conditions like inflammatory bowel disease and diabetic foot ulcers.
It can take from five to 20 minutes of multiple clinicians’ time to manually review the hundreds of potential patient charts each day to determine eligibility. This tool provides near real-time information on eligibility for each patient in the emergency department and across the inpatient units. - Nerissa Ambers
LT: How was the process of taking it live?
NA: The first step was to get clinical consensus on the patient criteria and map the data elements for the patient population. That exercise uncovered some limitations in identifying patient conditions using discrete rules. Enter AI. The tool leverages the first SHC-built, LLM-based application embedded in a care workflow. HB: We started by developing rule-based tools within Epic to help operations more easily identify patients that may be appropriate to transfer to the Patient Care Unit at SN1. To further enhance this capability, we’ve been partnering with our data science team to integrate capabilities Click the image for a larger view. from the chatEHR tool they developed which allows us to evaluate all data, including nondiscrete data such as provider notes, within a given patient’s chart.
LT: What sort of challenges did you face?
NA: The electronic health record, especially for our complex patients, can be huge. The data extraction, transformation, load and integration is a non-trivial exercise. The TDS teams came together with open minds and collaborative spirits to bring this new idea to fruition. The path ahead is complex, but the commitment to improve patient care is steady.
LT: How does it relate to Stanford’s mission of research, education, and clinical care?
NA: The origin story of the AI capability which powers this tool stems from a student-led research project developed for the clinical care environment and guided by our TDS data science team. The project was presented at the Data Science Extended Team meeting which includes a vast interdisciplinary team of researchers, clinicians, and students. This team is the quintessential convergence of our mission statement.
LT: What are some ways that TDS has contributed to the success of this project?
HB: A multidisciplinary TDS team has been working to help with this – Epic ASAP, Epic Inpatient, Data Science, and Nursing Informatics.
NA: “Applications + Data Science + Integration = Success.” I want to place particular emphasis on the strong work of Sree Ram Akula, Clancy Dennis, Krishna Jasti, Anna Kalinsky, Nikesh Kotecha, Pranav Masariya, Duncan McElfresh, Jack McKeown, Satchi Mouniswamy, Abby Pandya, Aditya Sharma, and AJ Wessels.