Artificial intelligence can help identify easy to miss patients who might be good candidates for a palliative or hospice care referral, a recent pilot at Mass General Brigham (MGB) revealed.
The results of the findings were presented Friday at the Value-Based Payment Summit.
Timely end-of-life care benefits patients. Patients and their families may also be more open to a conversation about goals of care during a hospital stay, MGB said in presentation slides shown to Fierce Healthcare.
A pre-launch analysis found the health system’s Newton-Wellesley Hospital was slightly below the nationwide median hospice referral rate, at 3.7%. Boosting the rate to 4.2% would equal 15 additional referrals to hospice per year, at a savings opportunity of $411,825 annually for the health system’s Medicare Shared Savings Program Accountable Care Organization (MSSP ACO).
The six-month pilot began at the hospital in 2023 and relied on an AI-driven tool from tech company Radial called Smart Hospice. Over the course of the pilot, 9% of 1,522 MSSP ACO patients were recommended for a consult by the tool, with 40 receiving inpatient palliative consults. An estimated 17 of those would have been missed without the Radial tool.
If 13 identified patients enroll in hospice, the hospital will gain $850,000 in Total Medical Expenditure savings, equal to $2 million annualized, while improving those patients’ quality of life, the organizations said.
Identifying those patients furthers the health system’s population health goals of facilitating home-first care, improving inpatient capacity and ensuring limited skilled nursing facility (SNF) beds for those who need them most.
MGB is also hoping to decrease total medical expenditure by avoiding unnecessary SNF stays and reducing acute inpatient utilization. MGB data showed that more than 10% of MSSP ACO members were dying within 30 days after discharge from a SNF.
“Unfortunately, in the U.S., we have this culture and practice of providing intense and high-cost medical care at the end of a patient’s life,” Amy Baughman, M.D., medical director of care continuum at MGB, told Fierce Healthcare. “It's also very challenging because it can be hard to identify patients, and that’s the beauty of this tool.”
The Smart Hospice tool leverages data from claims, EHRs and health information exchanges to look at comprehensive healthcare utilization, identifying non-obvious, “on the border” patients at risk of unnecessary interventions who might benefit from palliative care.
“By being able to see the total trajectory … the technology is able to take a much longer term view and understand there’s a broader context to this,” Thaddeus Fulford-Jones, Ph.D., co-founder and CEO of Radial, told Fierce Healthcare.
Clinicians might be focused only on a given point in time on a patient journey—on what is directly in front of them, he added. “It’s really hard to figure out where this fits within a broader context, so that’s what Smart Hospice does automatically.”
AI augments, 'never' replaces humans in end-of-life decision-making
Providers' calls to incorporate a secondary tool into end-of-life decisions has previously drawn scrutiny, particularly when those decisions have financial implications.
At least one major health system, HCA Healthcare, allegedly encouraged staff to transition more patients to palliative and end-of-life care, thereby increasing churn and boosting hospital quality scores. Some sources from an NBC News investigation outlined an algorithm used at certain HCA hospitals that quantifies the risk of mortality and identifies patients who score high and become candidates for transfer to palliative care.
When asked about the ethics of using an AI-driven tool, Baughman was quick to emphasize Radial's tool is only intended to recommend patients to clinicians that might otherwise be missed. From there, it is up to several care teams working in concert to decide on the best next steps.
Newton-Wellesley Hospital's transition care management team was tasked with using the tool daily and leveraging Epic to review recommended patients and recommending them as needed to the palliative care team on the first or second day of hospitalization. That team would then consider the right placement and work with the transition care team to engage with a patient and their family on next steps.
“The beauty of this tool is that the purpose has never been to replace human decision-making,” Baughman noted. “This is a tool to augment and support workflows that are already in place to help you identify blind spots.”
Users were prompted to rate in the tool every patient that was recommended by the platform, though not all patients were scored for agreement. Users agreed that 92% of recommended patients could be suitable candidates to screen for palliative or hospice care.
An analysis of internal pre- and post-pilot surveys found that a quarter more respondents felt the level of inpatient palliative care utilization was “just right,” and a quarter more felt they had the right tools to justify a palliative consult.
“It’s nice to have AI behind you sometimes, to give that boost of confidence because these are loaded conversations,” Baughman said.
Overall, the hospital team is better educated and more confident identifying patients for various palliative care referrals, according to Baughman. The pilot also identified topics of interest for future palliative care education for teams and demonstrated how a hospital stay can be effectively leveraged to accelerate discussions around goals of care.
“The default is to have a lot of medical care. If we don't do anything different patients will have a lot of aggressive interventions,” Baughman said.