This transcript is auto generated
00;00;00;00 - 00;00;32;08
Bryan Nowicki
Hello and welcome to Hospice Insights: The Law and Beyond, where we connect you to what matters in the ever-changing world of hospice and palliative care. Getting Ahead of the Curve: How Hospice Providers Can Build Smart AI Practices, Part 2. Taylor, welcome back to continue our conversation about AI in the health care space and how it is being used.
00;00;32;11 - 00;00;38;13
Bryan Nowicki
This, you know, in the grand scheme of things, it's a relatively new development, but here it is and we're all dealing with it.
00;00;38;16 - 00;00;54;12
Taylor Crossley
Yep. That's right, that's right. You can't you can't turn your blind eye. This is an and it's exciting thing to be able to use. But we've got to make sure that we're putting the right guardrails in place. I'm excited to be here to dive deeper into what kind of guardrails hospices can be considering, as they're implementing these tools in their in their patient care.
00;00;54;12 - 00;01;30;28
Bryan Nowicki
So great, great. So in part one, we kind of laid the foundation and you provided some great guidance on you know what where is it being used. How do you develop a framework to make sure you're monitoring and managing AI to avoid risk or at least minimize risk? And here we are, part two, where it's a bit of a dive into some more specific areas where you can share some of your insights into what our clients, what other providers should out there should be thinking about when they're implementing or making use of AI?
00;01;30;28 - 00;01;32;21
Bryan Nowicki
So why don't we get started?
00;01;32;21 - 00;01;33;11
Taylor Crossley
Yeah.
00;01;33;13 - 00;01;49;24
Bryan Nowicki
Let's go with, you know, something you touched upon before was vendors, health care provider, they contract with a lot of vendors. So what what's your guidance regarding how providers should deal with AI in the vendor context?
00;01;49;24 - 00;02;18;17
Taylor Crossley
Yeah, absolutely. So vendor contracts, that's a big area to be considering. Hospice organizations know, you know, that they need a business associate agreement, a BAA with vendors who handle protected health information. However, many organizations, existing BAA's are not written with AI in mind. And oftentimes standard vendor contracts. You know, talking about the underlying services agreement here are also rare
00;02;18;17 - 00;02;40;24
Taylor Crossley
rarely AI, AI ready and that gap creates some real exposure. So there's four main areas that I recommend clients, pay attention to kind of risk areas to look out for whenever you are engaging with or considering engaging with a new vendor. Or maybe you're looking at their contract and getting geared up to start working with them.
00;02;40;24 - 00;03;04;12
Taylor Crossley
The first area to think about is vendors using hospice patient data to train their AI models. You want to make sure that, you know, if then the vendor is going to use patient data to train their AI. And if so, build appropriate guardrails around their ability to do that. The second is vendor switching the underlying AI model without telling the hospice.
00;03;04;12 - 00;03;29;24
Taylor Crossley
So typically these AI, you know, programs, softwares is the ability for AI to be built into these, vendor products, comes from a certain AI model. And if they change the model that they're using without telling the hospice, that can lead to additional considerations that maybe the hospice didn't consider when they were first adopting this vendor.
00;03;29;27 - 00;03;54;03
Taylor Crossley
So that's something to take into consideration as well. A third is the AI features, being added through online terms of service that no one reviewed when the original contract was signed. So this is something I see a lot is, technology providers, or vendors I guess often have a standard online terms of service. That the provider is entering into.
00;03;54;09 - 00;04;19;14
Taylor Crossley
And then they also enter a business associate agreement. But frequently providers don't actually go to those online terms of service and make sure they're reviewing those to see what kind of carve outs are included there, and what kind of protections the vendor is putting into place. So making sure that, you know, what I features are actually being used in the product is really important.
00;04;19;14 - 00;04;33;22
Taylor Crossley
And that's something you can learn through the terms of service. And you can also learn, you know, maybe some carve outs that they're trying to throw in there through those online terms of service. So that's an important thing for providers to be looking at as well.
00;04;33;22 - 00;04;58;09
Bryan Nowicki
I agree, and I've encountered issues with these online terms of service. So who the contract will say the contract that signed will say, you know, we're incorporating by reference all our online terms of service, maybe get a link. And everybody's fine when the contract is signed. But often those terms of service say we can change these terms of service whenever we want to, with or without notice.
00;04;58;12 - 00;05;18;01
Bryan Nowicki
And the terms of service that appear a year after you sign the contract may be different from the terms of service that existed when you signed the contract. So, it's a good idea to make sure you're looking at those terms of service, but preserve a copy of them, see if they're able to be changed with or without notice.
00;05;18;04 - 00;05;38;28
Bryan Nowicki
That's kind of ever since, terms of service have been on the internet, that's been a problem. And, the fact that it persists even with AI companies is not surprising at all. But, you know, just from our past experience, we know that you got to know what the terms of the agreement are, and that's important to document and preserve all of that.
00;05;38;28 - 00;06;10;16
Taylor Crossley
Yeah, absolutely, absolutely. And I think, too, this is another area where we talked last time about, you know, kind of inventorying your AI use and maybe doing that as a part of your in your security risk assessment. Maybe you go through and look at your current vendors, maybe you're not engaging in new vendor, but during that security risk assessment, you can look at your current vendors and kind of look back at those online terms of use and make sure you know if they have maybe added AI into their software.
00;06;10;18 - 00;06;32;14
Taylor Crossley
Since you actually signed the contract because of that ability to change them without giving notice to. So that's that's a really good thing to be thinking about, both prospectively and retrospectively. The fourth and final kind of specific risk area to look for when contracting is when you have an agreement that does not define AI clearly enough to understand how and where it's being used.
00;06;32;17 - 00;06;57;13
Taylor Crossley
So the definitions that are used both in the terms of service and then also in your business associate agreements, is a really important thing to be taking into consideration. Don't let your vendor contracts remain silent on AI. If a contract is silent on AI, the vendor can integrate AI into its products and services without disclosing that use to the hospice.
00;06;57;15 - 00;07;19;17
Taylor Crossley
And that can create some real compliance concerns. And to ensure the transparent use of AI, we recommend requiring in writing that any use of AI in connection with the hospices data is subject to the hospices prior written consent. You know, and if the vendor will not agree to that broadly, then you can push for it at minimum
00;07;19;17 - 00;07;27;14
Taylor Crossley
where AI touches PHI clinical records or billing data. So just trying to build in those protections into your contract.
00;07;27;14 - 00;07;42;28
Bryan Nowicki
So when, when a provider is, talking with, a vendor, what what kinds of questions should they be asking to make sure they're educating themselves about how AI may or may not be part of this arrangement?
00;07;42;28 - 00;08;06;25
Taylor Crossley
Yeah, there's lots of questions to be asking. One is, does this product use AI and if so, how? Again, that's part of your inventory in the use of AI. Another good question is will I ever have access to our patient data or protected health information? You know, ask yourself or maybe ask the vendor, is there a BAA in place with AI specific provisions?
00;08;06;25 - 00;08;35;00
Taylor Crossley
And does it prohibit using patient or clinical data to train AI models? Another one is does the vendor disclose its hosting provider and the underlying AI model it uses? Ask yourself about indemnification. So does the indemnification provision in the service agreement cover AI specific claims like erroneous outputs, privacy violations, or clinical bias? You know, are AI obligations carved out from the general liability cap?
00;08;35;02 - 00;08;57;23
Taylor Crossley
And then kind of on that consent question too, requiring in writing that any use of AI in connection with organizations data is subject to the organization's prior written consent, making sure that that's something that you're considering as well as, your own consent for the vendor to be using your, your, your data, in the use of AI.
00;08;57;24 - 00;09;19;06
Bryan Nowicki
All right. Let's pivot to another area. More specifically of, of AI use. In part one, you referred to ambient scribes and the use of AI to, improve documentation, maintain documentation integrity. And and so what should hospice organizations be considering in that area?
00;09;19;06 - 00;09;54;23
Taylor Crossley
Yeah. So one one thing to be thinking about, is ambient tools that are embedded in services that you're all already using, like Teams and Zoom. Those platforms are commonly used today. And that can create a unique compliance problem. If those tools are already embedded into those, into those services. So one thing to be thinking about, if you're using those ambient tools for clinical encounters, is that one clinical encounter using those services can generate multiple records.
00;09;54;23 - 00;10;23;26
Taylor Crossley
Right. So you've got an audio file from the encounter. You have a transcript, you've got a summary that the AI generated, you've got a draft note that the AI generated. And you've got a system log. So you're generating, a lot more records than you maybe previously were for one single patient encounter. And what you've got to be thinking about is the privacy, retention, discoverability and privilege questions that are attached to each of those records. Another
00;10;23;26 - 00;10;47;00
Bryan Nowicki
We, I'm sorry. Yeah. Tell we, I hear you know, us as lawyers. One way we encounter AI problems is you hear of attorneys submitting briefs to the court that cite to cases that don't exist or make or court cases. And the, quote does not exist. These, I think they're called hallucinations. Some judges have gotten in trouble for that, too.
00;10;47;06 - 00;11;00;09
Bryan Nowicki
Is that an issue? I imagine it's an issue in the health care space. It's not, hopefully not just, something we lawyers have to deal with, but, Well, what what are some of the health care specific issues relating to those kinds of hallucinations?
00;11;00;09 - 00;11;44;01
Taylor Crossley
Yeah, the risk of AI hallucination is a big one for providers to be thinking about. Maybe you've got an a, an ambient AI listening tool that generates clinical findings, or even diagnoses that were not actually ever documented or observed by the provider. That's a that's a big issue that providers should certainly be taking into account, not only if they're using, you know, AI, ambient tools, you know, to listen in on the conversation, but also maybe if they're using that, on the back end to help them generate a clinical note without the listening feature, you know, all of these AI tools can really generate this issue of AI hallucination.
00;11;44;03 - 00;12;08;26
Taylor Crossley
And so in the hospice context, a hallucinated note that fabricates clinical findings, that might support a terminal prognosis is not just a documentation error at that point. It's a potential False Claims Act submission. And depending on the facts, it could raise fraud concerns. So this AI hallucination, just like lawyers, you know, can get us get in trouble for making up stuff in their court briefings.
00;12;08;29 - 00;12;27;15
Taylor Crossley
Providers can get in a lot of trouble for making up things unintentionally, right through using AI and then submitting that data to, for billing. And then that can raise False Claims Act considerations, which is a, a big, a big compliance concern to be thinking about.
00;12;27;15 - 00;13;02;17
Bryan Nowicki
Well, absolutely. And when you kind of the two phrases you use the use you mentioned the False Claims Act and billing and you know, when, when we're addressing, potential compliance violations and doing an investigation, we're always considering this, the kind of an issue that could result in just a repayment, an honest mistake, or or could it rise to the level of a False Claims Act issue where the hospice, whether intentionally or through reckless disregard, I think is one of the phrases of the one of the standards in the False Claims Act.
00;13;02;17 - 00;13;29;25
Bryan Nowicki
Did their billing system or the way they submit bills by the aid of AI, jeopardize or create some exposure? Because when you talk about the False Claims Act, the penalties are significant triple damages. So whatever you were paid or you need to repay, repay times three fines, attorney's fees. So certainly something you want to avoid. But put that into the context of AI in the billing area.
00;13;30;00 - 00;13;36;08
Bryan Nowicki
How how do hospices what what should they be considering, regarding billing issues and the False Claims Act?
00;13;36;08 - 00;14;00;11
Taylor Crossley
Yeah. So AI is a great tool, right, in all of these different settings. But the real important thing, kind of the bedrock in all of these different topics that we're talking about, is responsibility on behalf of the provider, the provider themselves, the hospice themselves needs to be responsible for the AI output, whether that's in the clinical setting, or that's in the billing setting.
00;14;00;11 - 00;14;29;02
Taylor Crossley
And so, under the False Claims Act, the hospice would be liable if an AI tool contributed to an overbilling problem under the False Claims Act. You know, the organization submitted the claim and it's responsible for what's in the claim. AI involvement doesn't shift that responsibility. So like Bryan, you mentioned under the False Claims Act, knowing, reckless disregard or deliberate ignorance of the accuracy of a claim is enough for liability.
00;14;29;04 - 00;14;52;18
Taylor Crossley
So if an AI assisted claim is inaccurate, the hospice is not automatically liable under the False Claims Act. However, by submitting a material false claim that was generated or assisted by an AI tool, or relying on AI billing outputs without meaningful human review, that may be a form of reckless disregard that could trigger, False Claims Act liability.
00;14;52;18 - 00;14;56;03
Bryan Nowicki
So what should hospices do to try to mitigate that risk?
00;14;56;04 - 00;15;23;17
Taylor Crossley
Yeah. So first audit AI assisted billing outputs the same way you would and human coder. Spot check regularly and build that review into the workflow of submitting claims. No AI generated claims should go out the door without human review and sign off. And, you know, third, finally is to make sure the clinical documentation supports what's being billed and that might be a part of the human review and sign off.
00;15;23;17 - 00;15;43;26
Taylor Crossley
Right, making sure that your reviewer is looking at the clinical documentation, attach that claim to check that the output from the AI is accurate, and accurately reflects what your clinical documentation says. The documentation integrity issue, and the billing liability issue are really the same issue.
00;15;43;27 - 00;16;03;22
Bryan Nowicki
All right. Let's talk about, another topic you you referenced in part one. This is communications with patients and families and transparency around that. Do hospice patients have a right to know when AI is being used in some way and in the context of the care they're receiving.
00;16;03;22 - 00;16;27;27
Taylor Crossley
Yeah. See, this is an interesting question because, like we've been talking about, AI is so new. And the framework, the government, the, you know, the legal framework, the regulatory framework is still building. So there's no universal federal disclosure requirement right now. And the regulatory landscape is unsettled. But state law is moving in different directions.
00;16;27;27 - 00;16;56;28
Taylor Crossley
So that's something to really be taking into consideration. As a provider, is what states do you reside in? And is your state regulating the use of AI in patient care? For example, if you think about California, in California, if a clinic or health facility uses generative AI to draft a message to a patient, that message has to disclose the use of AI and give the patient a clear path to reach a human clinician.
00;16;57;01 - 00;17;26;17
Taylor Crossley
So laws like that in California, laws like those, and there's a new law in Colorado as well that requires transparency with patients. So there's all of these different state laws that are developing and they're very recent new laws. And so it's unclear how those laws will be enforced. But regardless, there are state laws that are starting to require providers to give patients notification of how AI is being used in their care.
00;17;26;19 - 00;17;50;14
Taylor Crossley
And hospice patients and families are in a uniquely vulnerable position. You know that that's something that you want to be thinking about whenever you are using AI. So some organizations might be considering whether to disclose, you know, AI use to their patients and families, and our recommended practice is to build an AI disclosure into patient set paperwork.
00;17;50;16 - 00;18;21;11
Taylor Crossley
You can cover the key issues in kind of a single additional document as a part of your admission paperwork that covers these key issues. Lets the patient know that there is AI assistance with clinical documentation. Let them know how, administrative interactions, where patients or families may be interfacing with AI. Let patients know how that works in your organization, and then also let the patient know how it's being used for clinical decision support.
00;18;21;13 - 00;18;53;21
Taylor Crossley
Just a simple, you know, one paragraph disclosure about how the hospice is using AI. And, you know, taking responsibility, including that the hospice knows that they're responsible for those clinical outputs if they're using it. And that setting provides the transparency to the patient and also builds trust with your, with your patients as well. So our recommendation is to just provide that disclosure, even if there is no, you know, universal federal or state standard that requires it right now.
00;18;53;23 - 00;18;56;15
Taylor Crossley
I think it's a good practice to go ahead and get in place.
00;18;56;15 - 00;19;14;03
Bryan Nowicki
All right. And so we're coming to the end of part two. And Taylor, this is really been great information laying the foundation and then getting into the specifics. So how would you sum it up. What do you want hospices to walk away thinking about regarding AI.
00;19;14;07 - 00;19;33;08
Taylor Crossley
Yeah okay. So number one, I hope that we didn't scare you is the first thing. You don't need to stop using these tools. AI is a great tool that you can use to help innovate, right. And that's why I think that it's a it's a great place for organizations to be able to provide better care offerings to their patients.
00;19;33;08 - 00;20;01;24
Taylor Crossley
So don't stop using these tools. Don't be scared. These tools are genuinely useful and can make your staff more effective. But that said, you need to know where AI is operating in your organization. Hospices need to govern and intentionally and make sure that a human is being accountable. Right. Accountability is really the core kind of takeaway, for the use of AI. Make sure that you're being accountable for every clinical and billing decision that AI touches.
00;20;01;26 - 00;20;15;05
Taylor Crossley
So overall, get your policy in place, review your vendor contracts, train your staff. And that's a great starting point. As you continue on your, you know, your avenue to using AI, to help provide better patient care.
00;20;15;05 - 00;20;42;24
Bryan Nowicki
So great, great. Taylor. And I'm I'm so glad you're on our hospice team. So, as, as, hospices and home health agencies approach us with questions about AI. We certainly have, the expertise, the experience, and the the, the ideas that are going to help them really engage with AI to their benefit, make sure they're minimizing the risks as much as, as is practical and and making good decisions.
00;20;42;24 - 00;20;47;16
Bryan Nowicki
So thanks again, Taylor. I really appreciate you, sharing all this information with us.
00;20;47;20 - 00;20;49;00
Taylor Crossley
Yeah. Thanks so much for having me.
00;20;49;05 - 00;20;53;28
Bryan Nowicki
You're welcome. Take care.
00;20;54;01 - 00;21;14;03
Bryan Nowicki
That's it for today's episode of Hospice Insights: The Law and Beyond. Thank you for joining the conversation. To subscribe to our podcast, visit our website at huschblackwell.com or sign up wherever you get your podcasts. Until next time, take care.