Algorithm-Based Referrals Increase Palliative Care for Advanced Cancer Patients

This study presented at the 2024 American Society of Clinical Oncology (ASCO) Annual Meeting investigated the use of an automated system to improve access to palliative care for patients with advanced cancer. Palliative care focuses on improving quality of life for patients with serious illnesses, and early intervention can lead to better outcomes.

The study involved randomizing clinics within a large cancer network. In some clinics, doctors received automated prompts to refer high-risk patients (identified by a scoring system) to palliative care specialists. In other clinics, referrals remained at the doctor's discretion.

The results were promising. Compared to usual care, clinics using the automated system saw a significant increase in patients completing palliative care visits (over 46% vs. 11%). Additionally, these clinics observed a decrease in patients receiving end-of-life chemotherapy, potentially indicating a shift towards more comfortable and appropriate care approaches.

Importantly, doctors participating in the study found the automated system acceptable. While some concerns existed about staffing limitations and patient suitability, the overall feedback was positive.

This study suggests that using algorithms to prompt referrals can be an effective way to increase access to specialist palliative care for patients with advanced cancer. This can lead to improved quality of life and potentially less aggressive end-of-life care.

Here, Ravi Parikh, MD, MPP, assistant professor of medicine and assistant professor of medical ethics and health policy at the Perelman School of Medicine, University of Pennsylvania, discusses the abstract.

Transcription:

0:05 | This was a pragmatic trial. It was done in a cluster randomized framework among 15 clinics as part of a large community oncology network in Tennessee. So a very real world setting where palliative care access is resource-constrained. And so what we deployed in the context of this cluster randomized trial was an EHR-embedded intervention that flagged patients that were high risk, and then offered those upper default referrals to clinicians. What we found was that clinician opt-out rates were generally very low, which was a positive finding, because it signals that most clinicians agreed with what the algorithm says. Opt-out rates were less than 10%. We also found that rates of completed palliative care consultation increased by nearly 4x in our intervention arm compared to our control arm, from 11% to over 46%. Concurrent with those increased palliative care referrals, we found a 2x decrease in end-of-life chemotherapy among individuals who died in our trial. Chemotherapy near the end of life decreased from around 16% to around 6% in the intervention arm. And so those are promising signals of effectiveness in a real-world setting of a default palliative care referral intervention. I should note that there were some findings that we did not observe; for example, we did not observe patient quality-of-life benefits. That had to do I think more so with the fact that in the real world, it's really difficult to get quality-of-life survey completion. And so we need more robust way to captures patient quality of life and other patient reported assessments in the real world.

1:31 | The couple of big findings are 1, default palliative care that doesn't rely on the clinician referral is a really promising avenue when we can introduce palliative care and standardized formats without explicit approval from someone's primary oncologist. We can increase patient engagement and interest in palliative care, and we can get them to palliative care faster and more often. So I would encourage practices that are really interested in investing in increasing palliative care referrals to think about defaults. The second thing is that algorithms, they don't need to be AI, they can be relatively simple algorithms are out there. And they will be used, because one of the biggest clinician complaints that I have is well, how do I know who to refer? And these algorithms are out there we are; we have made ours in the paper that's forthcoming publicly available. It can be translated into very easily any EHR format. It's a simple rules-based algorithm. And so whether you're going to use something proprietary, where they're going to use something that's a little more simple, rules-based. And these can be integrated with robustness, and should be used, I think, to help tailor how patients receive palliative care.

Transcription created with AI and edited for clarity.