School of Nursing Students and Faculty Aim To Improve Patient Care With Artificial Intelligence

(March 7, 2025) — A novel interdisciplinary research project seeks to streamline hospital admittance to psychiatry inpatient units by creating an artificial intelligence (AI) model to inform decisions concerning staffing levels and bed availability.

The research project is a collaboration among faculty at Georgetown’s School of Nursing, Graduate School of Arts and Sciences and the MedStar Health Research Institute. The work is supported by Georgetown’s The Red House, the university’s education research and development unit.

According to the research team, psychiatric practice is unique in not having objective clinical measures, such as laboratory or imaging data, to help determine diagnosis and how much staff support a potential patient requires. Instead, multiple factors — including patient reports, mental status and other behaviors — are considered to estimate risk for violence as well as inpatient treatment needs.

“Acuity is a very important measure to determine whether the unit can manage a new patient,” said Karan Kverno, PhD, PMHNP-BC, FAANP, FAAN, professor of nursing at the School of Nursing. “Inpatient units are often receiving requests from multiple emergency departments (ED) where, if not transferred, patients with possible acute psychiatric conditions may have to stay in the ED overnight or longer, which is not ideal for patients or the staff.”

Acuity is defined by the amount of staff required to effectively treat a patient as well as the staff required to maintain a level of care across the psychiatric unit. “Acuity of the unit often changes in the same nursing shift as staffing levels and patient needs change,” said Kverno.

The research team collected patient acuity and unit acuity survey records completed each shift by MedStar inpatient psychiatry nurse managers from two MedStar hospitals to develop a large database to use with a facilitated machine learning, a subtype of AI. The goals of machine learning will be to refine the current algorithms used to estimate psychiatric acuity in order to improve accuracy and decision-making in real time.

By taking into account both the needs of the patient and the demands on the hospital staff, the research group behind building the AI model hopes to maximize the availability of care and improve patient outcomes.

An Interdisciplinary Approach to Patient Care

Shortly after Kverno arrived at Georgetown in September 2023, she overheard Qiwei “Britt” He, PhD, associate professor in the data science and analytics program and director of Georgetown’s AI-Measurement and Data Science lab, speaking about AI research while attending a CNDLS event. Kverno and Mihriye Mete, PhD, associate research professor in the School of Medicine and director of behavioral health research at the MedStar Health Research Institute, worked on a pilot examination of acuity measurement in psychiatry and had previously talked about a potential role for AI. Kverno suggested a meeting between all three.

At the meeting, the group discussed collaborating on a project using AI to better measure acuity on psychiatric units.

“The three of us brought our different training and perspectives to the project,” said Kverno. Britt He had a lab and graduate students, who were able to help build the model by inputting the large data sets supplied by staff nurses at MedStar Health facilities.

Data science graduate students brought their technical skills to the project by conducting quantitative analysis and applying AI models, while nursing graduate students provided essential contextual validation of the AI-processed data. The collaboration brought together technological capabilities and real-world clinical knowledge.

As a retired Navy nurse with 31 years of experience, Stacia Fridley (G’24) understood the necessity of better calibrating unit acuity when she joined the project as a Doctor of Nursing Practice (DNP) student at Georgetown. “Nurses are taught to anticipate the needs of our patients by constantly reassessing and preparing for the next shift,” said Fridley.

Nurse managers filled out patient acuity survey records, answering questions on a scale involving factors such as suicidal ideation and risk to falls, which require a higher level of care. The intake forms also asked the nurses to assess the current staffing needs of the unit. Graduate nursing students like Fridley were brought into the study to interpret notes on the intake forms recorded by the nurse managers and validate the data from the AI model to make sure what was suggested made sense in practice.

Building an AI Model for Acuity

The team performed advanced linguistic analyses that would have been difficult without AI tools. Data science graduate student Leqi Ying (G’25) estimates inputting over 500 surveys into the model so far.

“When Dr. He first mentioned the project to me, I wanted to work on it because I believe we can use AI to help people,” said Ying. “It was also helpful to work with the nursing students because I would input the data and then go to them to have them explain what it meant in terms of what was being scored.”

“The data team also asked us to interpret what we refer to as ‘nurse speak,’ common shorthand phrasing used by nurses on units that wouldn’t be recognized by a computer or AI model initially,” said Fridley. “For example, in one of the data sheets, a nurse drew a symbol conveying that two nurses from the psychiatry unit were sent to the emergency department to help staff that unit.” Fridley explained this would be important information to include in the model, which tries to consider workload and staffing levels.

“What we’ve learned from nurses, especially during COVID, is that when acuity on the unit is too high, meaning when the burden of patient care is too much, they leave the profession, which can be devastating, especially the loss of specialty nurses,” said Kverno. “But monitoring the level of acuity on the unit ensures patients get great care and nurses feel they can safely handle the burden of care.”

The research group hopes the study will produce an AI model that could eventually be deployed not only in MedStar Health psychiatric units but also throughout different units in the hospital.

“We need more studies like this,” said Fridley. “Nurses’ time with patients is precious and should be protected, and if we can use AI to standardize and automate some administrative tasks, what a gift that would be to the profession.”

Heather Wilpone-Welborn
GUMC Communications