Responsible AI in Hospice: What Clinical Governance Should Look Like

Hospice Leadership
Written By Tonia Barrier

AI is already inside many hospice and home health organizations. It shows up in EMR documentation assistants, intake tools that summarize referral packets, scheduling software, and staff using general chatbots on their own. In a lot of agencies, nobody formally decided to let it in.

That is the governance gap I worry about. Hospice documentation carries unusual weight. The record has to support a terminal prognosis, describe decline in specific terms, and reflect the care a team actually gave. A tool that writes smooth, generic narrative can quietly undermine all three.

I do not think the answer is to avoid AI. The answer is to govern it the way we govern any other clinical practice, with clear ownership, defined uses, and real oversight.

The standards you can anchor to

Hospice leaders do not have to invent AI governance from scratch. Several existing standards already point the way.

Nursing accountability does not transfer to software. The American Nurses Association's position statement on AI in nursing practice holds that appropriate AI use should support the profession's core values. It also restates, from the Code of Ethics, that nurses remain accountable for their practice even when a system or technology fails (ANA, 2022). For hospice, that means the nurse who signs a visit note owns every word in it, whoever or whatever drafted it.

Bias in decision support tools is a known risk. Section 1557 of the Affordable Care Act prohibits discrimination based on race, color, national origin, sex, age, or disability in federally funded health programs. HHS's 2024 final rule applied that principle to patient care decision support tools, which it defined broadly, from AI models to simple flowcharts and calculators (45 CFR 92.210). HHS has since announced it will not enforce the rule's specific requirements to identify and mitigate those risks. Even so, the rule text and the underlying law remain, and enforcement priorities can change. A tool that scores hospice appropriateness or prioritizes visits deserves a bias review either way.

Risk management needs a structure. The NIST AI Risk Management Framework, released in 2023, organizes AI risk work into four functions: govern, map, measure, and manage (NIST, 2023). It is voluntary, but it gives a compliance team a shared vocabulary and a place to start.

Patient information still belongs under HIPAA. Any vendor whose AI tool receives protected health information on your behalf is a business associate and needs a business associate agreement (45 CFR 164.502(e)). Staff pasting patient details into a free consumer chatbot is a privacy problem, not a productivity win.

Five governance practices for hospice and home health

This is what I would put in place first. None of it requires a data science team.

1. Make an inventory

You cannot govern what you have not found. List every tool in use that generates, summarizes, scores, or recommends anything clinical. Include EMR features that were switched on during an upgrade and tools staff found on their own. In my experience, the second list is usually longer than leaders expect.

2. Name an owner and a committee

Every AI tool needs a clinical owner who answers for how it is used. Larger organizations can route new tools through an existing compliance or QAPI committee instead of building something new. What matters is that someone with clinical authority says yes or no before a tool touches patient care.

3. Define approved uses in writing

A written policy should say which tasks a tool may support, which it may not, and what review is required. For hospice, I would draw a firm line around anything that drafts eligibility narratives or predicts prognosis. Those tools may help organize information. The clinical judgment behind a certification has to stay with the physician and the team.

4. Audit the output, not just the tool

Add AI assisted documentation to your existing chart audits. Look for notes that read the same across patients, decline language that is not backed by measurements, and details that did not happen at the visit. Copied or generic narrative has always been an audit risk. AI can make it faster to produce.

5. Train staff on both the tool and the limits

Nurses and aides should know what the tool does well, where it tends to get things wrong, and that their signature means they verified the content. They should also know which tools are approved for patient information and which are never acceptable.

Governance is what makes AI usable

I see AI governance as part of the same work as eligibility governance in my ALIGN framework. Both come down to one question: can your organization show that a clinical decision was made by the right people, for the right reasons, and documented truthfully? Agencies that can answer yes will be able to adopt useful tools with confidence. Agencies that cannot will find that AI magnifies whatever weaknesses were already there.

I speak with hospice, home health, and post acute leadership audiences on responsible AI, clinical governance, and compliant growth. If you are planning a conference session, leadership retreat, or board education program on these topics, you can reach me through the contact page on this site.

About the author

Tonia Barrier, MSN, RN, is the Founder and Principal Consultant of PAC Leadership Solutions and a DNP Executive Leadership candidate at Walden University. She is the former CCO & Compliance Officer of a multistate hospice organization, a trained End of Life Doula, and an HPNA Tennessee State Ambassador. She developed the ALIGN framework to help hospice organizations grow without trading away clinical integrity.

References

Tonia Barrier
Previous

Building a Scalable Serious Illness Care Model Without Losing Quality or Cost Control
Next

How a Hospice Can Attract Appropriate Referrals Without Compromising Eligibility