Douglas Barnaby, MD, MSc, MBA

5 insights on leading the quantum shift in healthcare from Duke Health CEO Craig T. Albanese, MD, MBA: 1. Innovation starts with purpose 2. From problem to progress: easing documentation burden 3. Innovation from the inside out 4. Smart partnerships, shared values 5. Innovation that serves a mission "Start with empathy. Lead with insight. Measure success not in headlines, but in impact." https://lnkd.in/gqZnubjD

45

4 Comments

🔥 𝘖𝘶𝘳 𝘪𝘤𝘦𝘣𝘦𝘳𝘨 𝘪𝘴 𝘮𝘦𝘭𝘵𝘪𝘯𝘨, 𝘺𝘦𝘵 𝘸𝘦’𝘳𝘦 𝘣𝘶𝘴𝘺 𝘣𝘶𝘪𝘭𝘥𝘪𝘯𝘨 𝘯𝘦𝘸 𝘰𝘯𝘦𝘴 𝘰𝘯 𝘵𝘰𝘱 𝘰𝘧 𝘪𝘵 In healthcare 🩺 , we keep paying twice 𝐨𝐫 𝐦𝐨𝐫𝐞 for the same “innovation.” And “𝘱𝘢𝘺𝘪𝘯𝘨” isn’t just money: • 💸 duplicate contracts and overlapping tech investments • ⏰ leadership and clinician time wasted on repetitive pilots • 👨🏻‍💼 unnecessary hires or low-ROI teams • 🔄 duplicative committees and fragmented oversight • 🪫 innovation fatigue on the front lines • 🚫 lost trust when patients and staff see a system that feels uncoordinated But let’s be honest ⚠️ this isn’t just duplication. It’s patching 𝐢𝐧𝐡𝐞𝐫𝐞𝐧𝐭𝐥𝐲 𝐟𝐚𝐮𝐥𝐭𝐲 𝐩𝐫𝐨𝐜𝐞𝐬𝐬𝐞𝐬. Yes, Epic’s new AI features look impressive. But 𝘩𝘰𝘸 𝘮𝘢𝘯𝘺 𝘩𝘦𝘢𝘭𝘵𝘩 𝘴𝘺𝘴𝘵𝘦𝘮𝘴 𝘤𝘢𝘯 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 𝘪𝘮𝘱𝘭𝘦𝘮𝘦𝘯𝘵 𝘵𝘩𝘦𝘮 𝘦𝘧𝘧𝘦𝘤𝘵𝘪𝘷𝘦𝘭𝘺? It takes ✅ governance ✅ integration ✅ change management Otherwise, even the best tools just add another layer of digital noise. If the foundation is shaky, no amount of “AI” will fix it. 💡 The real opportunity is to stop patching and start 𝐫𝐞𝐝𝐞𝐬𝐢𝐠𝐧𝐢𝐧𝐠: ☑️ Align innovation with real clinical pains and workflows ☑️ Build governance that prevents duplication ☑️ Invest once in ways that compound toward outcomes Otherwise, 📛 we’ll keep digitizing the inefficiencies that are 𝐚𝐥𝐫𝐞𝐚𝐝𝐲 𝐛𝐮𝐫𝐧𝐢𝐧𝐠 𝐮𝐬 𝐨𝐮𝐭. Thanos Karras Ami Bhatt, MD Sanjay Gandhi Dr. Julia Dmitrieva Steve Metz Stacey E. Rosen, MD, FAHA Roxana Mehran Christopher Kramer David Wiener American College of Cardiology American Heart Association American Society of Echocardiography #HealthcareInnovation #LearningHealthSystems #ResponsibleAI #HealthTech

8

2 Comments

? "real hospital discharge notes with one inserted false recommendation" from MIMIC. Nothing about the notes generated in Epic, or any other EMR that copied it's effectiveness now haunting 352 million Americans charts is real. Not the patient, whose existence is a statistical ghost assembled from billing codes clicked to maximize revenue, not chronicle suffering. Not the physician, whose clinical judgment has been replaced by template logic designed by people who think documentation means compliance masquerading as care. Not the narrative, which was never about what happened to a human being but about what could be billed to an insurance company. The MIMIC database, like every electronic health record database, is a monument to Frankfurt's bullshit (link here: https://lnkd.in/dgY-k5rA unconcerned with truth, designed only to achieve effects, in this case the effect of extracting payment from Medicare while appearing to document medicine. The entire substrate is rotten, billing artifacts dressed as clinical documentation, the womb that birthed these algorithms diseased from conception, and now we're running elegant studies quantifying how often large language models accept misinformation when the information they were trained on was never information at all but billing optimization dressed in SOAP note formatting, click-generated context meant for cost savings and corporate-prized efficiency, and we're surprised, genuinely surprised, that nearly one in three medical AI responses accepts fabricated advice when fabrication was the generative principle of the training data? The study shows clinical documentation had the highest acceptance rate at 46.1%, nearly double the social media baseline, and the authors note that "authoritative tone increases risk" as if this were a bug rather than the feature Epic Systems sold to every hospital CFO in America, the ability to make billing codes sound like medicine, to make compliance sound like care, to make extraction sound like documentation, and physicians learned this language because the system punished them for speaking any other way, and now the AI has learned it too because that's what learning means when your teacher is a liar. They're abstractions several levels removed from the reality they claim to represent, mappings that fail in ways that make the failures invisible because we're no longer looking at patients but at representations of representations of billing behaviors that once upon a time might have had some relationship to actual clinical encounters. And we're training AI on this. And acting surprised when it hallucinates. Because hallucination is what you get when you train pattern recognition systems on patterns that were never real. The misinformation isn't being inserted into the notes. The misinformation IS the notes.

3

4 Comments

Health systems are feeling the financial pressure to move from experimenting to scaling gen AI and agent-based tools across the stack of operational and clinical workflows. What’s different today is that health systems want to move fast. Moving at speed however, requires discipline and measurement to prevent a disaster. Effective AI governance is critical to ensuring patient safety, regulatory compliance, and sustained clinical trust. This isn’t like traditional health IT oversight; it’s closer to the kind of governance used in safety-critical industries like autonomous vehicles or aviation, where real-time monitoring and system-level safeguards are non-negotiable. I worked with Dr. Kedar Mate Mate on a piece for Chief Healthcare Executive that lays out a framework for what health systems need to govern GenAI responsibly and at scale. Here are the six foundational activities: 1️⃣ Continuous real-time monitoring of LLM outputs to detect hallucinations, drift, and clinical inaccuracies as they occur. 2️⃣ Risk-aligned oversight tailored to use case severity, with human review for high-risk outputs and automation where appropriate. 3️⃣ Detection and governance of shadow AI use to mitigate untracked tools operating outside formal oversight. 4️⃣ Full auditability and traceability of outputs, including model version, prompt, and timestamp for compliance and QA. 5️⃣ Transparency and disclosure from vendors on model updates, fine-tuning activity, and safety testing protocols. 6️⃣ Integrated escalation pathways that route uncertain or high-risk outputs to human review within existing workflows. Where is your health system on the AI adoption curve? What systems are in place for this to go smoothly?

135

13 Comments

We need an AI app for Emergency Medicine, but not the one people think Most people imagine an ED AI app as a diagnosis engine. That is not the highest leverage. The ED needs an AI layer that reduces friction: auto-documentation that is clinically safe and structured real-time triage support that flags deterioration early pathway prompts that reduce variation in high-risk cases flow prediction: staffing, bed pressure, waiting time risk discharge safety: follow-up prompts, red flags, patient instructions The best ED AI app is not flashy. It is invisible, integrated, and measured on outcomes. If it does not reduce cognitive load and improve safety, it is a distraction. #EmergencyMedicine #AI #DigitalHealth #Workflow #PatientFlow #HIMS #HealthInformatics #PatientSafety #HealthcareInnovation #GCCHealthcare

2

1 Comment

Should Health Technology Assessments Count More Than Health Outcomes? NICE - National Institute for Health and Care Excellence's latest commentary in Value in Health offers crucial insights for healthcare policymakers grappling with a fundamental question: When evaluating the cost-effectiveness of health interventions, should we count only health outcomes and healthcare costs, or include broader societal impacts like productivity, education, and criminal justice? **The Debate:** A "societal perspective" has intuitive appeal—why not value childhood treatments that improve educational outcomes or mental health interventions that reduce criminal justice costs? Yet NICE reaffirmed its "health sector perspective" in 2022, with flexibility for exceptional cases. **Why?** ✓ **Methodological challenges** — No robust evidence on opportunity costs of displaced services across sectors ✓ **Ethical concerns** — Counting productivity systematically favors working-age populations over retirees ✓ **Evidence gaps** — Societal effects often rely on expert testimony rather than quantitative data ✓ **Coordination complexity** — Requires harmonized decision-making across education, justice, and health sectors **Implications for GCC Policymakers:** → The "perspective" decision fundamentally shapes which interventions get recommended and funded → There's no global consensus—50%+ of HTA agencies accept societal perspectives, but with minimal guidance → Balance comprehensiveness with timeliness—broader assessments delay patient access → Consider sector coordination mechanisms before adopting societal perspectives James Koh #HealthPolicy #HTA #ValueBasedHealthcare #HealthEconomics #GCC #NICE

83