According to a new study, a device known as CHARTwatch may lower the chance of unanticipated mortality among hospitalized patients by acting as an early warning system for quickly declining health.
A clinical pathway is established for high-risk patients using the artificial intelligence (AI)-based system, which monitors real-time data from patients’ electronic medical records to identify individuals who might have an unscheduled admission to the intensive care unit (ICU). By providing real-time notifications to physicians, twice-day emails to nursing teams, and daily emails to the palliative care team, the platform was able to cut mortality in a general internal medicine unit by 26%.
According to lead author Amol Verma, MD, a clinician-scientist at St. Michael’s Hospital and Temerty professor of AI research and education in medicine at the University of Toronto, Toronto, Ontario, Canada, “AI tools hold great promise for helping us improve the quality of healthcare by improving the accuracy and efficiency of diagnosis, assisting with the personalization of treatment decisions for individual patients, enhancing our ability to predict and prevent future health events, and improving the efficiency of healthcare operations,” Medscape Medical News reported.
“It’s essential that these tools be researched and implemented carefully,” he stated. “Like other healthcare interventions, AI tools may have benefits or unintended consequences, and we need to ensure that they are safe and effective.”
Using CHARTwatch
Verma and colleagues spent three years developing and testing CHARTwatch, a suite of tools to notify physicians and other relevant medical professionals when a patient is at high risk of deteriorating, and a recommended care plan for high-risk patients. CHARTwatch also includes a machine learning prediction model. They evaluated the instrument in the general internal medicine unit, which has 70 beds, at St. Michael’s Hospital, an academic health facility located in Toronto’s inner city.
The instrument was put into use from August to October of 2020. Prior to that time, a formal deterioration detection score system had not been established, and the critical care response team would respond to patients based on the discretion of a physician or nurse. By initiating measures earlier and seeking the advice of palliative care professionals when necessary, the main objective of introducing CHARTwatch was to lower the number of nonpalliative hospital deaths.
The research team in this study evaluated the relationship between clinical outcomes and the use of CHARTwatch as an early warning system in a nonrandomized yet controlled manner. They examined mortality among 4023 patients in the postimplementation phase (November 2020 to June 2022) and among 9626 patients in the general internal medicine unit prior to the tool’s use (November 2016 to June 2020). Additionally, the 13,649 patients in the general internal medicine unit were compared to 8470 subspecialty patients in the areas of nephrology, cardiology, and respirology—disciplines in which the instrument was not employed.
When the tool was utilized in 2020–2022, the rate of nonpalliative deaths among patients receiving general internal medicine was, on average, much lower (1.6%) than it was during the previous period (2.1%). For death, the adjusted relative risk (RR) was 0.74. The rate of nonpalliative deaths in the subspecialty cohorts (1.9% vs. 2.1%; adjusted RR, 0.89) did not change appreciably during the intervention period.
The percentage of nonpalliative deaths among high-risk general internal medicine patients with at least one tool-based alert was 7.1% in 2020–2022 compared to 10.3% in 2016–2020, resulting in an adjusted relative risk of 0.69. Nonpalliative deaths during those times did not differ significantly among subspecialty groups (10.4% vs 10.6%; adjusted RR, 0.98).
Additionally, there were no statistically significant differences in total fatalities, palliative deaths or transfers, ICU transfers, or length of hospital stay during those times across analyses and subgroups.
The study team discovered that patients in the intervention group were more likely to get systemic glucocorticoids and antibiotics, as well as to have their vital signs checked more frequently following a high-risk alert. Orders for intravenous liquids, code status orders, or imaging, however, did not seem to have changed.
Putting into Practice in All Hospitals
In order for AI technologies to effectively generate predictions across medical settings, especially among individuals from varied backgrounds, Verma said researchers need to have access to large-scale health databases. This would enable the technology to be utilized more widely.
“It is important for tools that appear to be effective in a single context to be tested in a wider range of settings,” Verma stated. “This is particularly true for AI tools, which perform best in the patient populations that were used for their development.”
The researchers are working with GEMINI, a network of 35 hospitals that shares data, to create a vast, inclusive data collection that will enable this kind of extensive AI-based research and development.
In order to improve patient safety, care quality, and overall efficiency, it is imperative that system upgrades be consistently prioritized in the healthcare industry. Researching and implementing cutting-edge techniques is crucial to overcoming these obstacles and enhancing care delivery as expectations rise and healthcare becomes more complex, according to Rabia Shahid, MD, an associate professor of medicine at the University of Saskatchewan in Saskatoon, Saskatchewan, Canada.
Shahid has studied hospital early warning systems for patient deterioration; he was not engaged in this work. She and her colleagues discovered that although tools can enhance patient outcomes and provider communication, they must be improved and have greater support from stakeholders in order to be successful.
“These tools, particularly those driven by technology such as AI and machine learning, are vital to research and develop because they significantly enhance the quality of care across critical domains, including safety, effectiveness, timeliness, patient-centeredness, and efficiency,” she stated. “By enabling early detection of patient deterioration and supporting clinical decision-making, these tools help ensure that care remains consistent and of high quality, even in the most demanding and high-pressure hospital settings.”
The AMS Healthcare Compassion and AI Fellowship and the Vector Institute Pathfinder Project provided some funding for the study. Verma has received travel assistance from the Alberta Machine Intelligence Institute, is employed part-time by Ontario Health outside of this project, and is sponsored by the Temerty Professorship of AI Research and Education in Medicine. Verma and a number of writers co-invented CHARTwatch, which a startup later purchased. In the future, Verma might be able to purchase a minority stake in the business. Shahid did not disclose any pertinent financial affiliations.
Health and medical journalist Carolyn Crist covers the newest research for Medscape Medical News, MDedge, and WebMD.
