Research by VNS Health has found that generative AI systems can help clinicians better assess and predict patients’ palliative care needs.
The nonprofit recently conducted a case study of three individuals in which staff evaluated methods for using AI to generate a clinical decision support tool. The AI system, an open-source large language model, translated diagnostic and billing codes into a narrative format to identify patients’ needs. VNS Health staff, with lead researcher Rutgers University Assistant Professor Elizabeth Luth, presented their findings at the recent International Nurse Informatics Conference.
”We combined our administrative claims data, our predictive modeling output and basically a template that we could use to prompt AI to pretend as if it was a nurse practitioner and synthesize all the information at the member-specific level,” Carlin Brickner, vice president of data science at VNS Health, told Palliative Care News. “This gives nurses about a three-paragraph summary of medical history, implications for palliative care and what the potential follow ups or interventions might be appropriate.”
VNS Health’s service region spans the New York City area, including four surrounding counties, serving close to 50,000 patients and health plan members. The New York-based operator provides home health, hospice and palliative care, as well as behavioral health and personal care/private duty services.
AI technology has been gaining ground in the health care space, including when it comes to hospice and palliative care decision making. VNS Health’s research points to a potential new application for the technology that could help streamline ongoing patient assessments.
For the research, data scientists at VNS Health developed a 12-month mortality risk algorithm incorporating 3,500 predictors derived from medical and pharmaceutical claims data and clinician-identified triggers to identify seriously ill older adults who may have palliative care needs, according to the case study. They then further applied AI to generate a coherent summary of what the data indicated, similar to clinicians’ notes.
Further research is needed to fully assess the limits and functionality of these systems and methodologies, Kathryn Bowles, vice president and director of the Center for Home Care Policy and Research, VNS Health’s research arm. Bowles is also a professor of nursing at the University of Pennsylvania.
Thus far, the technology shows promise in giving clinicians a “big picture” assessment of patients’ palliative care needs, according to Brickner.
“We have a lot of predictive modeling that goes on throughout our operations where it’s basically we’re trying to predict a future outcome and align it with a workflow and or intervention. When you put [those data] in front of a clinician, that’s often not enough,” Brickner said. “They often ask for a lot of context to what is behind that prediction. What we want to do is to summarize all the factors that are critical for actually enrolling or engaging someone into a palliative care program.”
