Augmenting Value Analysis with Artificial Intelligence: Frameworks, Tools and Ethical Guardrails

"Good morning, everyone, and thank you for joining me today. My name is Kyle Atkins, and I’ll be talking with you about one of the most important crossroads we face in healthcare innovation—how we can apply AI responsibly through ethics and strategy."

"The title of this talk says it all: ethics and strategy. Because when it comes to AI, the technology itself isn’t the hard part—it’s how we, as leaders, ensure it’s implemented in a way that builds resilience and trust in our systems."

"Here’s the roadmap for our conversation today. We’ll explore four key areas that shape how AI can be both ethical and strategic in healthcare."

"First, we’ll look at how to develop an AI governance strategy—because policies and oversight are what keep AI aligned with our values."

"Second, we’ll examine methods to minimize bias and risk, since AI can unintentionally reinforce inequities if we’re not careful."

"Third, we’ll talk about balancing innovation and trust—because the best technology fails if the workforce doesn’t buy in."

"And finally, we’ll anchor all of this in real-world case studies, both successes and failures, so we can take away lessons that apply directly to our organizations."

"But before we dig into frameworks and strategies, I want to ground us in a recent study that shows just how powerful—and risky—AI can be when compared directly with human physicians."

As we move forward, let’s consider a profound statement: “AI may outshine us in empathy and efficiency, but without ethics, it cannot outlast us in trust.”

This sentence captures a crucial insight. While AI has shown remarkable capabilities, especially in empathy and efficiency—as we've seen from the study where chatbot responses were preferred for their quality and empathy—ethics form the cornerstone of sustainable trust.

Trust isn't just about delivering empathetic responses; it’s about ensuring that AI operates within an ethical framework that people can rely on. Without ethics, even the most efficient AI systems may fail to gain long-term trust from users.

"Researchers in Texas asked a fascinating question: if you compare responses from physicians with responses from an AI chatbot—who do patients think does better?"

"The study pulled real patient questions from a public forum and had physicians and an AI chatbot provide responses. Then, independent reviewers compared the two sets of answers."

"The findings were eye-opening: nearly 79% of the time, evaluators actually preferred the chatbot’s responses. They rated the AI answers as higher quality and more empathetic than those of human physicians."

"On the surface, this might sound like a win for AI—but it raises a serious ethical question. If patients perceive AI as more empathetic than their doctors, what happens when those AI responses are wrong, biased, or lacking clinical context? Who is accountable?"

"So what does this study actually tell us? It shows that people can quickly build trust in AI—even to the point of preferring its empathy and quality over physicians. But trust without accountability is dangerous."

"If AI can win hearts so quickly, then leaders in healthcare must be proactive in ensuring those tools are transparent, accountable, and equitable. Otherwise, misplaced trust could lead to harm."

"This is why ethics can’t be a box we check later. It has to be baked into the foundation of AI strategy. Let’s look at the key ethical principles we need to hold as non-negotiables in healthcare AI."

"The chatbot study shows us something powerful—patients are willing to trust AI even when it hasn’t proven itself in the same way human clinicians have. That means we have a responsibility as leaders. Every move with AI should start with ethics. Let’s walk through the four principles that must guide us."

"First, transparency. AI can’t be a black box in healthcare. If clinicians can’t see how a tool arrives at its recommendation, they’ll either distrust it or, even worse, trust it blindly without questioning. Neither is safe."

"Transparency means explainability—dashboards, confidence scores, or at least plain-language outputs that allow us to understand the ‘why’ behind the answer."

"It’s not just about building trust—it’s about protecting patient safety."

"Second, accountability. AI doesn’t replace human judgment. We still need clear oversight and responsibility for every decision."

"If a model misclassifies a patient or recommends the wrong treatment, who is accountable? The AI industry partner? The IT team? The physician?"

"Leadership must set that chain of responsibility from the start. AI doesn’t absolve us of accountability—it sharpens it."

"Third, bias mitigation. We have to assume bias is baked into every dataset. Bias comes from history, from who is included or excluded, and from how we label data."

"Left unchecked, AI can reinforce disparities in outcomes—whether that’s race, gender, socioeconomic status, or even geography."

"Bias mitigation means continuously testing for inequities, retraining models, and auditing results. It’s not a one-time step—it’s an ongoing responsibility."

"And finally, patient autonomy and trust. AI should never undermine the clinician–patient relationship. Patients deserve to know when AI is being used in their care."

"AI should enhance informed consent, not replace it. It should support shared decision-making, not dictate it."

"Trust grows when patients feel respected, and if AI is implemented in a way that feels hidden or imposed, that trust evaporates quickly."

"So these four principles—transparency, accountability, bias mitigation, and patient autonomy—are not optional extras. They’re the ethical foundation we must build before we talk strategy. Next, let’s see what happens when organizations try to implement AI without these guardrails in place."

When those guardrails aren’t in place, the results can be devastating—not because the technology fails, but because it succeeds in the wrong way. Let’s look at a case where an AI system worked exactly as designed… but because ethics were ignored, it produced inequitable and harmful outcomes

What happened:

"This was a population health risk algorithm used by hospitals across the U.S. Its purpose was to identify high-risk patients for care-management programs. On paper, it worked—it stratified patients and flagged those at higher risk. But the algorithm used future healthcare cost as a proxy for patient need."

The outcome:

"The problem was that cost doesn’t always equal illness. At the same risk score, Black patients were significantly sicker than White patients. Because of the cost proxy, the algorithm only flagged about 17.7% of high-need Black patients, when in reality nearly 46.5% should have been identified if the model was based on actual illness."

Ethical breakdown:

Transparency: "Clinicians and administrators weren’t told the true target variable—the algorithm was optimizing cost, not health. That lack of visibility obscured inequities."

Accountability: "No governance structure existed to catch the problem before it scaled across millions of patients."

Bias mitigation: "No subgroup audits were performed, so racial disparities were baked in from day one."

Trust: "Ultimately, this undermined equitable access to services and eroded trust, particularly for marginalized communities."

That algorithm did what it was built to do—but it was built on the wrong foundation. It reminds us that just because AI works doesn’t mean it works ethically. The good news is, when these principles are applied the right way, the outcomes can be substantial. Let’s look at a success story where AI strengthened both clinical performance and patient trust.

"Now let’s look at a success story—an example where AI was aligned with ethical principles and delivered measurable value."

What happened:

"Viz.ai developed a stroke triage platform that identifies large-vessel occlusions using imaging and then immediately alerts the stroke team through a secure mobile communication system."

The outcome:

"At one center, this tool cut treatment times dramatically: door-to-groin time dropped by 39% during off-hours and by 32% overall. That’s minutes saved when we know every minute equals brain tissue in stroke care. These improvements accelerated life-saving interventions and directly supported patient outcomes."

Ethical alignment:

Transparency: "The workflow was clear and measurable, with metrics like time-to-treatment tracked and reported."

Accountability: "Clinicians remained the final decision-makers; the AI functioned as a communication and coordination tool, not as a replacement."

Bias mitigation: "Process times were monitored across different pathways—direct-arrival patients and transfer patients—ensuring equity in workflow gains."

Trust: "Because the system improved outcomes and supported clinicians, it built trust among both patients and providers."

The contrast couldn’t be clearer: one system eroded equity and trust, the other enhanced both. That’s the difference ethics makes. Now that we’ve seen the stakes, let’s move into strategy—how we as leaders can embed these principles into every AI initiative we evaluate or implement.

"Governance is the bridge between ethics and outcomes. Ethics tells us what matters, outcomes are what we want to achieve, and governance is the structure that connects the two."

"In healthcare, governance means the system of policies, people, and processes that ensure AI is implemented responsibly."

"Without governance, AI can drift away from its original intent—like the population-health algorithm we saw earlier. With governance, we create guardrails that make sure every AI tool is aligned with our organization’s mission, values, and patients’ needs."

"Think of governance as a playbook: Who approves AI tools? How do we test them for bias? Who monitors outcomes over time? And how do we communicate openly with clinicians, staff, and patients?"

"That’s why I like the bridge metaphor you see here. On one side, we have ethics—the ‘why.’ On the other, outcomes—the value we want to deliver: resilience, efficiency, better care. But the only way to cross from one side to the other is through governance—the ‘how.’"

"For value analysis professionals, governance is essential. You are often at the intersection of clinical, financial, and operational decisions. AI governance ensures that the tools you evaluate don’t just look good on paper—they are safe, ethical, and sustainable in practice."

"So AI governance isn’t just a committee or a policy—it’s the connective tissue that makes ethical AI real. Next, let’s talk about what governance actually looks like in practice, starting with the structures that make it work."

"Once we understand what governance is, the next step is structure. Governance doesn’t happen by accident—it requires the right people at the table."

"First, an Oversight Committee. This cross-functional group should include clinical leaders, supply chain and value analysis professionals, IT, compliance, and risk management. AI touches every corner of the organization, so representation matters."

"Second, Defined Roles. Governance fails when nobody knows who owns what. Who approves new tools? Who monitors performance and bias? Who reports outcomes to leadership? These responsibilities must be explicit."

"Finally, Executive Sponsorship. Without visible leadership support, governance becomes a paper exercise. An executive sponsor ensures AI governance is taken seriously and has authority across the organization."

"With the right structure in place, governance can move from concept to action. Next, let’s look at what governance in action actually looks like—the policies and processes that keep AI aligned with our values over time."

"Now that we’ve identified the structure—who needs to be at the table—let’s talk about how governance actually plays out in practice. Governance is not just another committee. It has to show up in policies, partner expectations, and regular monitoring."

"The first element is Policies. These are your playbook. Just like you have policies for purchasing, credentialing, or clinical protocols, you need policies that define how AI is evaluated and adopted."

"Policies should set standards for ethics, patient impact, safety, and return on investment. Having these criteria up front ensures that every AI proposal is judged fairly, consistently, and transparently—whether it comes from a vendor or from internal innovation."

"Second, Industry Partner Assessments. In value analysis, we already evaluate vendors for cost, quality, and outcomes. AI raises the bar—we also need vendors to demonstrate transparency, bias testing, and explainability."

"Ask questions like: Can you explain how this model works in plain language? What data was it trained on? Have you audited it for bias? How do you plan to monitor performance once it’s live?"

"If a vendor can’t provide those answers, then their solution is not ready for healthcare."

"And third, Audits. AI isn’t set-it-and-forget-it. Even the best models drift over time—data changes, populations change, workflows change. Regular audits are the safety net."

"These audits should check both technical accuracy and fairness. It’s not enough for the model to be 95% accurate overall—if that accuracy drops significantly for certain patient populations, that’s a governance failure."

"By building auditing into the governance process—monthly, quarterly, or tied to key performance cycles—we keep AI accountable long after implementation."

"So governance in action looks like this: clear policies, strong vendor accountability, and ongoing audits. These are the mechanics that turn structure into impact. But to make governance effective, we also need simple tools—checklists, grids, and templates—that help leaders and teams apply these principles consistently. That’s what we’ll cover next."

"We’ve talked about governance structures and processes, but here’s where it gets real. To make governance practical, leaders need tools—something they can pick up, apply, and repeat. Let’s look at three key tools that help put governance into action."

"The first tool is an Implementation Checklist. This is essentially a roadmap for AI adoption: define objectives, build a cross-functional team, establish ethical guardrails, train staff, monitor performance, and scale responsibly."

"It prevents ad hoc adoption—the kind where someone brings in a new AI tool because it’s shiny or trendy. Instead, it ensures every AI initiative follows a consistent, structured path."

"The second tool is the Risk Assessment Matrix. This is a structured scoring system across domains like clinical relevance, evidence base, bias and fairness, privacy, cost/ROI, scalability, and post-implementation monitoring."

"Instead of asking, ‘Does this AI feel good?’ you can quantify readiness and risk. For example, if a tool scores below 60, it’s not ready for adoption. If it scores 80+, it’s a strong candidate. This makes governance measurable, not just theoretical."

"The third tool is alignment with position statements and toolkits, like those from AHVAP and other professional organizations. These provide a foundation for ethical, safe, and effective AI adoption across healthcare."

"By aligning your internal governance with accepted external frameworks, you ensure your practices are not only consistent, but also credible within the broader healthcare community."

"So, these three tools—the Implementation Checklist, the Risk Assessment Matrix, and the broader Position Statements and Toolkits—move us from principle to practice. They make governance actionable, repeatable, and credible. Now the question is: how do we balance the push for innovation with the need to build and maintain trust? That’s where we’ll head next."

"We’ve just covered governance tools—checklists, matrices, and frameworks—that keep AI grounded in ethics and structure. But governance alone isn’t enough. As leaders, we face a dual challenge: encouraging innovation while building and maintaining trust. That balance is what allows AI adoption to be both resilient and meaningful."

"Innovation doesn’t happen in a vacuum—it requires safe spaces for experimentation. That means pilot programs, sandbox testing, and encouraging curiosity among teams."

"As leaders, we should reward curiosity and create environments where staff feel empowered to test new AI tools without fear of failure. The key is to keep these pilots structured within the governance framework so they stay aligned with organizational values."

"Innovation without trust is useless. If staff don’t trust a tool, they won’t use it. If patients don’t trust it, they won’t accept it."

"Building trust requires transparency—being honest with clinical staff, administrators, and patients about what AI can and cannot do. It means showing not just the benefits, but also the limitations, and explaining how we monitor and safeguard outcomes."

"When people see that governance and transparency are baked into the process, they’re much more willing to embrace innovation."

"Accountability is the glue that holds innovation and trust together. Governance shouldn’t slow things down—it should guide them."

"We have to establish ongoing oversight, encourage participation at every level, and celebrate wins while also acknowledging shortfalls. That’s how we build credibility."

"Accountability also means having the courage to identify risks quickly and address them before they undermine the value of the innovation."

"So, balancing innovation and trust is about creating safe spaces to innovate, being radically transparent with staff and patients, and keeping governance active throughout the journey. If we can do that, we don’t just adopt AI—we adopt it in a way that builds resilience for the future. Next, let’s shift from balance to practical application—what this looks like when you return to your own organizations."

"So let’s take this from theory into practice. When you walk into a meeting with a vendor next week, you don’t need to know how to code or build algorithms. What you need are the right questions—the ones that reveal whether this tool is truly ready for healthcare."

Checklist Talking Points (5 Questions):

Can you explain how the AI works in plain language?

"If they can’t explain it without jargon, they don’t understand it themselves—or they don’t want you to. Either way, that’s a red flag."

What data was it trained on, and is it representative of my patient population?

"AI trained on the wrong dataset can make dangerous assumptions. A model trained on urban tertiary hospitals may not work in rural community care. This is about context and fairness."

How do you test for bias, and can you share the results?

"Don’t take their word for it—ask for evidence. If they’ve done subgroup analysis, they should be able to show you results. If they haven’t, it means bias could be hidden under the surface."

What happens if the AI fails or drifts—who is accountable?

"Every tool eventually drifts. The question is: who owns that responsibility? Is it the vendor, the hospital, or both? Clarity here prevents finger-pointing later."

How will you support ongoing monitoring and updates?

"AI isn’t one-and-done. Just like software needs updates, AI models need continuous retraining and evaluation. Make sure the vendor has a support plan, not just a sales pitch."

"By asking these five questions, you cut through the hype and force vendors to address transparency, accountability, and equity. If they can answer, great—you’ve found a partner. If they can’t, the safest answer is ‘not yet.’"

"So, what can you actually do Monday morning when you go back to your organization? Here are five practical steps you can start without a new budget, without waiting for a board meeting—just with the authority you already have as leaders."

Action Steps Talking Points:

Form an AI Governance Working Group

"Pull together a small team—VA, clinical, IT, compliance, supply chain. This doesn’t have to be 20 people. Even a core group can establish momentum."

Review Current AI Tools/Vendors

"Start by asking: what AI are we already using? Many organizations discover AI hiding in tools they bought years ago. Apply the checklist retroactively."

Introduce an AI Implementation Checklist

"Standardize the process. The same way you wouldn’t bring in a new device without evaluation, don’t bring in AI without a stepwise checklist."

Educate Your Teams

"Offer a 30-minute primer on AI ethics and governance. This isn’t about turning nurses or supply chain professionals into data scientists—it’s about giving them confidence to engage with AI."

Align with National Frameworks

"Use resources like the AHVAP AI Position Statement. When you align with broader frameworks, your governance has credibility both inside and outside your organization."

"If you do these five things Monday morning, you’ll move AI from buzzword to practical, ethical practice. And when governance, innovation, and trust come together, that’s when AI creates real value. Let’s wrap up by pulling it all back to one simple equation: ethics plus strategy equals resilience."

"Today, we’ve walked through the real story of AI in healthcare. We began with ethics—why it must come first. We looked at failures and successes that prove governance is not optional. We examined the structures, processes, and tools that make governance real. And we talked about balancing innovation and trust so that technology actually delivers value."

"At the end of the day, it all comes back to this simple equation: Ethics + Strategy = Resilience. Ethics gives us the foundation, strategy gives us the roadmap, and governance is the bridge that connects the two. Without that equation, AI becomes either dangerous or useless."

"So here’s my challenge to you: when you get back home, form your governance team, review your AI products. Don’t wait for the next wave of technology to force the conversation—lead it."

"Because here’s the truth: Artificial Intelligence will not save healthcare. Leaders will. The future of AI isn’t in the code—it’s in the courage of leaders like you to guide it responsibly."