Naqi Khan, MD

Oracle Health re-announced their new AI native EHR today. Although I want to write a bit more deeply about Oracle soon, it did spur some thoughts: 1. Timing and competitive positioning This is obviously a pre-UGM move to get headlines before Epic’s conference next week, but it’s not really competitive with Epic yet. It’s ambulatory-only, so for now, the real competition is athenahealth and eClinicalWorks. Yes, large health systems could have half new and half old, but they won’t want to operate in two parallel EHR worlds, given the extra complexity and support burden. Early wins here will hinge on landing pilots, turning them into credible ROI proof points, and using those to expand. 2. Agents vs. copilots confusion Oracle’s press release blurs terms: “Our agents act as smart assistants that can dynamically surface critical insights and queue suggested actions while enabling clinicians to remain in control.” That’s a copilot, not an agent. Agents imply a higher degree of autonomy and task completion; copilots assist and suggest while a human drives. The distinction matters for expectations and trust. It reads sloppy to me (or as a deliberate marketing ploy). 3. Voice as a primary modality Oracle is betting heavily on foreground (active) voice — commanding the system to navigate, retrieve, and act. I’m skeptical clinicians will embrace it as the main interaction mode. Ambient voice (passive listening) has seen rapid adoption recently, but active voice has a long history of failed attempts: Nuance Communications SpeechMagic, Microsoft Kinect in ORs, Amazon Alexa for clinics. The friction points are well known: - Latency: Is saying “Show me all unread and high-priority messages” actually faster than clicking a filter? - Repetition tolerance: Yes, many clicks suck, but you know what sucks more? Saying things multiple times. You need to one-shot everything or users revert to a more reliable modality - Noise & privacy: Active voice requires an open intent window in noisy, shared spaces, where overlapping speech can cause errors and louder commands risk PHI exposure. Ambient capture is limited to private exam rooms, focused on the clinician–patient exchange, with irrelevant speech filtered out. - Mental model mismatch: Clinical work often requires scanning multiple data points at once, which voice typically reduces to a slower, one-at-a-time process. Ambient works because it piggybacks on natural conversation, stays invisible until needed, and delivers output in familiar channels (structured notes, summaries). Active requires a deliberate mode switch and often interrupts the flow of thought, which is why, outside of niche cases, it hasn’t beaten keyboard/mouse/touch. Active voice has a role, but it seems secondary to the tried and true user interfaces. And if it is indeed secondary and not revolutionary, the main upgrades in this AI-native EHR seem limited over competition and jury still feels out on whether it warrants a switch.

610

81 Comments

I love getting these notes from our partners: “We are excited to continue to drive differentiated outcomes for medically underserved patients.” Healthcare is a complex, interconnected system. Solving its toughest challenges, like improving access, driving health equity, and transforming outcomes, requires multiple organizations. It's what makes healthcare solutions hard. You are REQUIRED to collaborate. My advice: Start & End with the shared mission. Everything else is details. #healthcare #technology

88

1 Comment

Oracle Health recently launched the Oracle AI Center of Excellence for Healthcare to help organizations take advantage of the rapid advances in AI. Learn how this can help solve the industry’s most pressing challenges. https://lnkd.in/gqDiahVu

94

2 Comments

Oracle Health's new AI-native EHR is now “available” to ambulatory providers, with their AI hospital EHR coming out in 2026. My 4 thoughts… First, the cliff notes on the announcement: → A push for “voice-first” where clinicians can retrieve patient data by just verbally asking: “What meds is this patient on?” → Smart AI assistants to surface patient summaries, clinical insights, evidence-based guidelines, clinical trial opportunities, etc. → Built on Oracle Cloud Infrastructure (OCI) → “Open” - meaning that health systems can use Oracle’s AI agents, build their own agents or integrate with third parties → Note: Certification is pending final regulatory approval - not yet in production Now onto my 4 thoughts (or speculations): 1️⃣ The No. 1 goal is to motivate health systems to move to Oracle Cloud Infrastructure I don’t know how much Oracle truly cares about the EHR, but they’d love everyone to be on OCI - regardless of what systems sit on top of it. The healthcare cloud market will grow to $200B in 10 years and Oracle wants a big chunk of that. 2️⃣ The No. 2 goal is to maximize customer retention - by being competitive on AI for the clinician UI/UX to minimize churn. Epic’s been testing 100+ AI use cases in the EHR with health systems over the past year, so just to stay in lock step, Oracle needed to release something beyond the Digital Assistant and add more clinical value. 3️⃣ Voice has so many exciting use cases, but it’s not going to be enough to leap frog the competition. I think many people actually underestimate how much faster they could operate if they did things more by voice and less by text. Many of us don’t do “voice messages” or use the mic to write texts… but as someone who has started to try it, it’s WAY faster. For Oracle to make voice a killer feature, they not only have to make the tool available, they have to truly evangelize its use. BUT… it’s going to be table stakes. They need something much bigger to win back customers, especially the big AMCs. 4️⃣ Despite the messaging around an “open” system, I suspect the push for Oracle’s own AI assistants / agents will be prioritized over making it third party friendly - BUT I think flipping this and demonstrating true openness is an advantage Oracle should grab. There’s a reason why third party AI scribes have flourished with Epic health systems, and I can’t name a single health system on Oracle that’s using one of those third parties (I’m sure there are exceptions but they are clearly rare) So I think health systems will want Oracle Health to prove the openness is truly there. And they SHOULD… because that’s a way to compete with Epic whose own customers often raise concerns isn’t as open as it should be. The problem, of course, is that being “open” and maintaining so many integrations with third parties is a LOT of work… fair. Providing their own MCP endpoint and making it easier for third-party Agents to connect makes this possible. What do you think?

62

21 Comments

LLMs are moving fast into healthcare but deployment isn’t simple. In our new piece in the Journal of Clinical Medicine (JCM) MDPI, we outline five challenges that stand between promise and practice: Workflow misalignment Bias and equity Regulatory uncertainty Technical vulnerabilities Trust and human connection Mitigation strategies are emerging, but safe adoption will require more than technical fixes. It will take governance, oversight, and leadership that keeps patients at the center. Read the full article: https://lnkd.in/euFFHvEn Co-authors: Yaara Artsi, Vera Sorin, MD, CIIP, Ben Glicksberg, Panagiota Korisianos CPC, Girish Nadkarni, Eyal Klang What do you see as the biggest barrier or opportunity in deploying LLMs responsibly in care?

154

9 Comments

11 tools shaping public sector innovation in 2025 Public leaders aren’t just debating tech—they’re deploying it.   If you're working in digital health, AI governance, or government reform...   These tools should already be in your stack: 1️⃣ WHO Digital Health Atlas  Global registry for health tech deployments. Essential for scale + sustainability. 2️⃣ OECD AI Policy Observatory  Compare national AI strategies, governance, and ethics frameworks. 3️⃣ UN Global Platform  Open sandbox for geospatial + statistical innovation. SDG monitoring made smarter. 4️⃣ OpenMRS  Open-source EMR used in 40+ countries. Fast deployment in LMICs. 5️⃣ AI for Health (WHO + ITU)  Framework to vet clinical AI tools for safety, equity, and scale. 6️⃣ World Bank GovTech Maturity Index  Ranks 198 countries’ digital government readiness. A guide for reform + funding. 7️⃣ OpenHIE  Interoperability engine for EMRs, labs, registries, and analytics. 8️⃣ Health Data Governance Principles  Ethics + privacy guidelines shaping national legislation worldwide. 9️⃣ GovAI Observatory (Oxford)  Tracks foundation model use in government—including risks + red flags. 🔟 ReHealth Africa (AUDA-NEPAD) Africa-led digital health infra rooted in ethics, equity, and sovereignty. 1️⃣1️⃣ Global DARE Index Benchmarks healthcare digital readiness across workforce, infra, and policy. 📌 Save this for your next innovation roadmap. Follow Rizwan Tufail for deeper insights on public sector AI and system change.

40

8 Comments

🧬 We talk about “health data” as if it’s one thing, but it’s really hundreds of incompatible languages trying (and failing) to talk to each other. Every layer speaks a different dialect: • EHRs: HL7 v2, CDA, FHIR • Claims: X12 837, UB-04, CMS-1500 • Labs: LOINC, SNOMED CT • Devices: DICOM, IEEE 11073 • Genomics: VCF, FASTQ, BAM Each was built for a single purpose, not interoperability. The result? 🚑 A patient’s data is scattered across 40+ systems, each with its own schema, timestamps, and access controls. But things are shifting. Newer models are moving beyond formats to: • Graph-based data structures • Semantic layers • Federated architectures These approaches preserve context, not just content, across systems. FHIR paved the road. But the next frontier is semantic interoperability. That’s not just data exchange; it’s data understanding. 🧠 The future of healthcare intelligence isn’t in collecting more data, it’s in connecting meaning. #HealthTech #DataInteroperability #FHIR #HealthcareAI #KnowledgeGraphs #SemanticWeb

147

15 Comments

Patient Matching with FHIR + LLMs for Clinical Trials Did you know that up to 32% of a clinical trial's cost can be spent just on patient matching? This staggering figure, highlighted in a Deloitte report, underscores a critical industry challenge driven by complex, unstructured protocol criteria. While Large Language Models (LLMs) are already showing immense promise in this area, the true breakthrough comes from combining their power over unstructured data with the robust, structured data from FHIR. The synergy of LLMs and FHIR holds incredible potential, especially given the vast amount of untapped insights in unstructured clinical notes. In our latest video, we share an approach we're actively developing around this. It's a solution we're exploring as part of an ongoing project. If you're also navigating the complexities of patient matching, or exploring the frontiers of AI in clinical research, we'd love to connect, compare notes, and discuss future collaborations. #FHIR #AIinHealthcare  

161

18 Comments

At Epic's annual Users Group Meeting, CEO Judy Faulkner unveiled an ambitious vision for integrating AI across the company's healthcare applications, with tailored tools like "Art" for clinicians, "Emmie" for patients, and "Penny" for administrators. While the polished demos showcased AI's potential to streamline healthcare, Fred Bazzoli at Health Data Management questions whether the industry is ready to adopt these innovations, especially for smaller providers and less tech-savvy users. Read more: https://lnkd.in/dQNtVxmG

76

10 Comments

Medicare Advantage Star Measures Are Raising the Bar I’ve been refreshing my analytics code ahead of the October MA Star data release. With help from OpenAI Codex, I completed some trending and cut-point visualizations I’ve had in mind for years. The results are striking: plan quality distributions keep shifting upward—yet achieving 4–5 Stars is tougher than ever. Consider the Statin Use in Persons with Diabetes (SUPD) measure for MA-PD plans: - SY2021 – an 85% score earned 4 Stars - SY2025 – that same 85% now earns 2 Stars My accompanying chart tells the story: higher overall industry performance, but a much steeper climb for top ratings. Let me know if there are any particular measures of interest. If you try this at home, keep in mind that CMS can and does change the Measure ID year to year so you have to build a mapping table to do trending.

102

8 Comments

The EHR inbox is a burnout furnace, and everyone knows it. Clinicians are spending more time than ever on messages, yet many patients feel more disconnected from their doctor than ever. If messages have surged, why does it feel like access is getting worse and not better? We have had AI-powered automatic response generation in the wild for almost a year. It helps, but it has not come near to solving the problem. Why not? Because this is a two-sided issue. Patients face a fragmented front door with portals, phone trees, and apps that rarely meet them where they are. Clinicians face a misaligned set of incentives and exploring productivity and responsiveness expectations that treat inbox care as invisible work. Systems think they’re being innovative but rarely have protocolized care pathways that extend beyond order sets into the population health based asynchronous digital infrastructure that we need. What would change if patients could choose the channel that actually works for them, and if inbox care had protected time, clear routing, and real credit on the clinician side? What happens to burnout and satisfaction when AI handles triage and drafts, and humans focus on judgment and connection? By the way, that feeling of access suffering is fact, not fiction. Recent data shows an increasing number of Americans reporting they can’t get timely access to care, and this will only get worse with impending cuts to Medicaid and narrowing payor networks. While expanding AI to automate more, how do we expand access without further burying our healthcare workers? I dig into the data and outline a practical playbook in my latest Techy Surgeon article. Read and subscribe at the link below.

28

2 Comments

A recent CHIME Foundation survey, sponsored by CliniComp, reveals strong support among #healthcare #CIOs for integrating AI to automate administrative tasks. Over 80% of health IT leaders prioritize AI automation, with 81% specifically targeting the reduction of administrative tasks to alleviate clinician burden. Furthermore, CIOs identify enhancing clinical decision support and improving revenue cycle management as other significant priorities for AI adoption, emphasizing the crucial role of AI in optimizing provider workflows and enhancing patient outcomes. The survey also indicates that nearly half of CIOs consider it extremely important to embed AI capabilities natively within their EHR systems, aiming to reduce reliance on third-party software. While many CIOs are already utilizing or evaluating AI tools, they continue to assess the most effective integration strategies. Overall, CIOs view AI as essential for enhancing data access, analysis, and driving innovation within the healthcare sector. What's Next? 👇 - The industry continues to move toward a consolidated EMR-centric model. The key question: Will this momentum persist, or will we see a return to best-of-breed systems focused on integration? - Embedding AI isn't a plug-and-play fix. Clinicians and staff must redesign their workflows to apply AI-driven insights and automation effectively. - Healthcare professionals must still think critically and validate AI recommendations before acting on them. - While we all recognize AI will transform healthcare operations, we must stay focused on critical thinking and reworking operational processes to make that transformation meaningful.

27

2 Comments