As artificial intelligence (AI) tools are handling ever more of the paperwork that underpins palliative care, ongoing trials are expanding their role in clinical decision making.
Providers have turned to bots for the narrow, repetitive tasks that take patient-facing time away from administrators and clinicians. AI is also adept at tasks that involve gathering and summarizing large volumes of data from multiple sources.
Responses to Hospice News’ 2025 Market Intelligence survey reflect this division of labor, with documentation software cited as the most common use of AI by 53% of respondents, followed by real-time audit tools (22%) and predictive clinical alerts (19%).
But with little research into their effectiveness, knowing where to invest time and money in these emerging technologies becomes trial and error. The appeal of AI for documentation is partly driven by its demonstrable return on investment, said Tiffany Hughes, chief clinical officer at PalliCare, a Texas-based provider offering in-home palliative care.
PalliCare uses AI embedded in its practice management system, athenaOne, for charting and routing faxed patient records, as well as transcribing the audio from patient conversations, known as ambient listening. Hughes said that integrating AI through an existing platform subscription can help hedge against obsolescence in a fast-moving technology.
PalliCare has also joined early testing of a clinical decision support feature on the platform, which summarizes patient data pulled in from ambient listening, test results, state-run data registries and prescriptions, among other sources.
“Six months ago, even a year ago, you’re like, ‘Hmm, I don’t know about that. I don’t know how helpful that would be.’ Whereas now it is extremely helpful and very accurate. You can just see it by the day getting better and better,” Hughes told Palliative Care News.
Testing of the clinical decision support feature, known as Patient Summary, began with small groups of users in summer 2025 said Chad Dodd, vice president of product development at athenahealth, which makes the athenaOne platform. Tests of a virtual assistant that allows users to query patient summaries followed a few months later. Both are set to launch this summer.
The summaries highlight changes since a patient’s last visit, and identifies conditions that it has clinically inferred from a patient’s reported or detected symptoms, without any clinician providing a formal diagnosis. Dodd pointed out that the software is not predicting a specific health outcome — only presenting relevant information for a clinician’s judgement — AI which does require U.S. Food and Drug Administration approval as a medical device.
For clinicians and patients, clarity alone can be a huge benefit.
“When you’re dealing with these types of high-risk patients, there’s so much going on,” said Hughes, a nurse practitioner. “The family says this, the patient says this, the labs say this. It’s so much information that you can miss something.”
While Hughes said she appreciates the nudge, AI adoption must continue to be treated with caution by clinicians, particularly when robot advice becomes so routine that it could be accepted without question.
Knowing exactly where and how AI should fit into the multidisciplinary treatment teams that drive palliative care is not obvious, said Selen Bozkurt, an AI ethics expert at Emory University.
“There’s a decision-making workflow, and it’s not clear where AI should be,” Bozkurt told Palliative Care News.
A December 2025 study by Bozkurt and colleagues found mortality prediction in cancer — which can lead to an end-of-life referral — was the most common application of AI in palliative care. She highlighted another 2025 study, one of the few randomized controlled trials of AI use in palliative care, that found bot use led to earlier referrals.
Bozkurt said many questions remain about safely using AI in palliative care: how to judge the efficacy of proprietary systems, whether a given AI model will make accurate judgments about under-researched subpopulations, and how a model can weigh intangibles like local resource limitations and the patient’s own concept of quality of life.
