AI-driven algorithm yields ‘practical, scalable framework’ to increase palliative care use
Click Here to Manage Email Alerts
Key takeaways:
- EHR referral resulted in higher rates of completed palliative care visits.
- Palliative care referral decreased end-of-life chemotherapy use among those who died during the study.
CHICAGO — An algorithm-based default referrals system led to a fourfold increase in specialty palliative care use when implemented in a large community oncology network, results of a randomized study presented at ASCO Annual Meeting showed.
Researchers also noted less end-of-life chemotherapy use among patients who received the default palliative care referrals and died during the study.
“We’ve done some prior work where we’ve used machine learning-based algorithms embedded in the electronic health record to prompt oncologists to have earlier conversations about symptom management and end-of-life care, and we’ve had pretty good effectiveness,” Ravi Bharat Parikh, MD, MPP, FACP, assistant professor of medicine and of medical ethics and health policy at the Perelman School of Medicine at University of Pennsylvania, told Healio. “The challenge is oncologists are busy, so using that paradigm, that framework and applying it to palliative care referrals has been a strong interest area of ours. In this study, we sort of did an extension of that.”
Background and methodology
Early specialist palliative care intervention can improve outcomes in individuals with advanced solid malignancies. These patients typically experience poor quality of life and aggressive end-of-life care treatment approaches.
However, those with advanced cancer do not always receive palliative care referral before death. Both clinician inertia and difficulty identifying which patients are considered high risk are noticeable barriers that can prevent patients from receiving such a referral, according to researchers.
The multiarmed BE-a-PAL trial included 562 adults (mean age, 68.5 years; 79.5% white; 48.8% women), with more than three-quarters (77%) of participants having a diagnosis of stage III or IV lung or noncolorectal gastrointestinal cancer.
The study utilized an automated EHR algorithm adapted from NCCN Palliative Care prognostic or psychosocial risk factors to assign patients a risk score.
Researchers included 15 clinics in a large community oncology network; the intervention arm included 296 patients, whereas the control arm included 266.
Through the intervention arm, oncologists received weekly default EHR notifications promoting specialty palliative care referral for high-risk patients, with patients whose oncologists did not opt out of the referral being introduced to specialty palliative care via a standard script with an offer to schedule a visit.
In the control arm, oncologists referred patients for palliative care at their own discretion.
Results
Researchers noted a similar mean risk score for intervention and control patients (3 vs. 3.2) through the study.
In the intervention arm, researchers observed 89% of clinicians allow for palliative care referrals, leading to 79% of patients agreeing to such visits. Compared with the control arm, the intervention arm saw a significantly higher rate of completed palliative care visits (46.6% vs. 11.3%; adjusted OR = 5.4; 95% CI, 3.2–9.2).
The 179 patients who died in the intervention arm less often received end-of-life chemotherapy compared with controls (6.5% vs. 16.1%).
Researchers noted no difference in quality of life or length of hospice stay among those who died during the study.
Through interviews, researchers noted that clinicians viewed algorithm criteria as appropriate, with a nurse coordinator introducing palliative care to patients.
Study investigators identified staffing-related issues and inappropriate use of palliative care due to low symptom burden or stable disease as study limitations.
Next steps
Parikh and colleagues reported a fourfold increase in specialty palliative care using the algorithm-based default referral framework while cutting end-of-life systemic therapy in half. These results show the potential for a “practical, scalable framework to increase palliative care access with automated risk prediction,” researchers wrote.
However, additional factors need to be considered moving forward, according to Parikh.
“There’s probably two or three areas that are a bit unanswered for us moving forward,” Parikh told Healio. “First, how do you design your algorithms in ways where you curate a higher-risk population? Our algorithms were still inaccurate some percentage of the time.”
A second topic for future study involves improving the structure of the palliative care intervention, Parikh said.
“In our case, the only intervention that we routed to was an initial palliative care consultation because we wanted it to be pragmatic. But any palliative care specialist will tell you that if you want to enjoy the benefits of palliative care most, it’s not just a single consultation that you need,” he added. “And lastly, what happens if you try to deploy this where most patients receive their care, which are community-based oncology practices all around the country? It is a great question because it’s less about the technical capability and more about the preexisting motivations from these practices to be involved.”
Collapse
Parikh RB, et al. Abstract 12002. Presented at: ASCO Annual Meeting; May 30–June 3, 2024; Chicago.
Click Here to Manage Email Alerts