Jay Nanduri

It is applaudable for an Indian AI Healthcare company, DeepTek led by Dr Amit Kharat to receive WHO recommendation for its AI-based SaMD for #TB #screening. It helps reduce the cost of identifying TB cases by half, and strengthens healthcare economics, but also enhances quality of life by enabling timely triaging, even in the subclinical stages of the disease. DeepTek solution has regulatory clearance from both the US #FDA and EU #CE #MDR. In addition to identifying #tuberculosis, the #AI system can detect more than 21 #lung #abnormalities, such as nodules, masses, and pneumothorax. It can also interpret chest X-rays for children aged four and above, enabling its use across broader public health and routine screening contexts. #Healthcare #DigitalHealthcare #AI #India #indigenous Digitalhealthnews digital health news https://lnkd.in/gitpDQKU

8

Headline: A Regulatory Reset: Is this The End of "Feature Checking" in Health IT? The Assistant Secretary for Technology Policy (formerly ONC) just dropped the HTI-5 proposed rule, and it’s one of the most significant deregulatory shifts we’ve seen in years. By removing 34 certification criteria and updating 7 others, the government is signaling a major pivot: they are stepping away from regulating "how software looks" and doubling down on "how data moves." Here are the 4 biggest takeaways from HTI-5: 1) Slashing the Compliance Burden: ASTP estimates this rule will save developers an average of 4,000 hours each in the first year (NOTE: the 'per each' is per each org, not an individual developer). By eliminating "topped-out" requirements that are already standard in the market, they are freeing up developers to focus on innovation rather than federal paperwork. 2) Goodbye to "Safety-Enhanced Design" (SED): In a bold move, the rule proposes removing formal usability testing requirements (SED) and Accessibility-Centered Design. The philosophy? These are now industry norms or covered by other laws (like the ADA). Developers get more UI flexibility, but the pressure is now on vendors to maintain usability without a federal "stick." 3) The AI "Transparency" Rollback: The rule proposes removing the requirement for AI Model Cards. ASTP noted a lack of evidence that these "nutrition labels" for algorithms were actually improving clinical outcomes. This clears a major hurdle for AI-enabled interoperability and rapid tool deployment. 4) Data Standardization > Feature Regulation: While features are being cut, data standards are tightening. HTI-5 pushes the industry toward USCDI v4 and v5. The Goal: A standardized, inclusive baseline for patient demographics (SOGI, granular race/ethnicity, and Tribal Affiliation). The Catch: While it provides the "ingredients" for better patient matching, it still stops short of mandating a universal matching algorithm—leaving that "recipe" to the vendors. The Bottom Line: We are moving from an era of "Meaningful Use" feature-checking to an era of "API-First" efficiency. The government is handing the keys of UI/UX and AI transparency back to the market, while keeping a firm grip on the FHIR pipes. For decades, I've regarded certification as a waste of time because, generally, real use cases are not tested. In fact, many developers create feature flags for certification testing, and many of those features are not used in real-world examples. So this stance by ASTC is commendable. #HealthIT #Interoperability #HTI5 #EHR #DigitalHealth #ASTP #ONC #HealthTech Source: Article by Andrea Fox of Healthcare IT News:https://lnkd.in/gHGVUQHQ)

6

🪜8 Step Guide to Building a Compliant Healthcare AI Agent - Build smart🙌. Deploy safe🦺. Get cleared✅. 1️⃣ 𝗗𝗲𝗳𝗶𝗻𝗲 𝗜𝗻𝘁𝗲𝗻𝗱𝗲𝗱 𝗨𝘀𝗲 & 𝗖𝗵𝗼𝗼𝘀𝗲 𝘁𝗵𝗲 𝗥𝗶𝗴𝗵𝘁 𝗠𝗼𝗱𝗲𝗹 Before code, write the guardrails: ↳ Indications, contraindications, clinical setting ↳ Fit for deployment? (text, image, multimodal, SLM vs LLM, Edge vs cloud, HIPAA/GDPR ready?) 📌 Tip: Don’t pick a model before you define what it’s solving in a regulated workflow. 2️⃣ 𝗕𝘂𝗶𝗹𝗱 𝗥𝗶𝘀𝗸-𝗔𝘄𝗮𝗿𝗲 𝗖𝗼𝗻𝘁𝗿𝗼𝗹 𝗟𝗼𝗴𝗶𝗰 ↳ Agent logic = risk logic:bReAct or Plan-then-Execute? ↳ Keep a human-in-the-loop for key actions. ↳ Data Minimization and Audit Logging. 📌 Tip: Patient safety > automation speed. Design for oversight, not just output. 3️⃣ 𝗘𝘀𝘁𝗮𝗯𝗹𝗶𝘀𝗵 𝗖𝗼𝗿𝗲 𝗜𝗻𝘀𝘁𝗿𝘂𝗰𝘁𝗶𝗼𝗻𝘀 & 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗥𝘂𝗹𝗲𝘀 Instructions = medical SOPs + AI behavior ↳ Follows published guidelines (RSNA, ESGE, WHO, etc.) ↳ Defines escalation style and interaction loop 📌 Tip: You can’t validate a black box. Make rules explicit. 4️⃣ 𝗜𝗺𝗽𝗹𝗲𝗺𝗲𝗻𝘁 𝗮 𝗣𝗿𝗶𝘃𝗮𝗰𝘆-𝗣𝗿𝗲𝘀𝗲𝗿𝘃𝗶𝗻𝗴 𝗠𝗲𝗺𝗼𝗿𝘆 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝘆 ↳ Memory ≠ surveillance. ↳ Use only what’s needed: Short-term: Sliding window and Long-term: Summarized, consented, secured ↳ Differential privacy where possible ↳ Every access = auditable 📌 Tip: Comply with GDPR & HIPAA by default. Don’t collect what you won’t defend in court. 5️⃣ 𝗘𝗾𝘂𝗶𝗽 𝘁𝗵𝗲 𝗔𝗴𝗲𝗻𝘁 𝘄𝗶𝘁𝗵 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗲𝗱 𝗧𝗼𝗼𝗹𝘀 & 𝗛𝗲𝗮𝗹𝘁𝗵 𝗜𝗧 ↳ Only use validated tools (EHR, PACS, SaMDs) ↳ Match tool purpose to clinical use 📌 Example: An agent uses an AI polyp classifier—only if lesion size is within validated age range. ⸻ 6️⃣ 𝗖𝗹𝗮𝗿𝗶𝗳𝘆 𝗥𝗼𝗹𝗲 & 𝗞𝗲𝘆 𝗧𝗮𝘀𝗸𝘀 (𝗪𝗵𝗮𝘁 𝗶𝘁 𝗱𝗼𝗲𝘀, 𝗮𝗻𝗱 𝗱𝗼𝗲𝘀𝗻’𝘁) ↳ Scope = safety: Who is the agent helping? What tasks are guideline-backed? Where is bias and risk highest? What shouldn’t it do? 📌 Tip: Agents don’t replace clinicians—they augment them. Document limits clearly. 7️⃣ 𝗠𝗮𝗸𝗲 𝗢𝘂𝘁𝗽𝘂𝘁𝘀 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗮𝗯𝗹𝗲 𝗮𝗻𝗱 𝗘𝘀𝗰𝗮𝗹𝗮𝘁𝗶𝗼𝗻-𝗥𝗲𝗮𝗱𝘆 ↳ Explain the reasoning + evidence ↳ Flag uncertainty + escalate with context 📌 Tip: Build trust by letting users understand—not guess—why the agent escalated a task. 8️⃣ 𝗗𝗲𝗽𝗹𝗼𝘆 𝗠𝗼𝗱𝘂𝗹𝗮𝗿𝗹𝘆 & 𝗥𝗶𝘀𝗸-𝗙𝗶𝗿𝘀𝘁 ↳ Validate via human-in-the-loop trials ↳ Benchmark via RCTs / multi-reader multi-case studies ↳ One agent = one function and tool ↳ Shadow test → progressive autonomy 📌 Tip: Deployment ≠ flip-the-switch. Use safe rollout stages: shadow ⏭️→ co-pilot🤝 → assistive autonomy💪🏽. Build it right. Regulate it well. Deliver it safely. Detailed LinkedIn Article following soon! Happy (healthcare-grade) agenting! #agenticAI #aiforall #aiforhealthcare #safetyfirst #riskmanagement #aiforgood

12

Is your AI solution just a feature—or a real product? In this powerful clip, Amol Nirgudkar, CEO of Patient Prism , warns healthcare leaders about the rapid commoditization of single-feature tech tools like call summarization or voice AI. Amol explains why true innovation in healthcare AI must go beyond isolated solutions. A must-watch podcast episode for anyone building or investing in healthcare tech. Episode drops tomorrow! #Healthcare #AI #PatientPrism #HealthcareInnovation #PatientExperience #HealthcarePodcast #RoyBejarano #JasonSchifman #SCALECommunity #SCALEHealthcare

8

Introducing Banterop: Testbed for Language-First Interoperability :-) We've historically advanced healthcare interoperability by defining structured data models and workflow-anchored API transactions. This approach struggles with the long tail of diverse healthcare interactions (and if we're being honest: also struggles with the center of the distribution). What if, instead of pre-defining every detail, we could enable systems to negotiate requirements and share knowledge to get jobs done in real time? This is the core question behind "Language-First Interoperability" (LFI). At the upcoming HL7 Connectathon in Pittsburgh, Mark Kramer and I are excited to bring this concept into the HL7 community for early-stage, hands-on testing. This track has an exploratory spirit, reminiscent of the early days of FHIR. We're not bringing a finished solution; we're bringing a hypothesis and a set of open questions that we hope to explore with the community. What are the common failure modes of agent-to-agent conversations? How do we ensure reliability? What are the right patterns for grounding agentic interactions in high-quality data and user preferences? We expect challenges and failures, and that's precisely what will make the track a success -- because we will learn from them. To facilitate exploration, I've developed Banterop: an open reference platform and testbed. For background, see this intro video: https://lnkd.in/gfHVBsx4? The platform's scenarios focus on common, information-asymmetric interactions where one agent has deep knowledge of a patient's clinical history and the other holds the rules, policies, or eligibility criteria for a specific process. Think of a clinician's agent interacting with an agent for a payer, a clinical trial, or a public health registry. The goal is to lower the barrier to entry, allowing participants to bring their own technology (whether it's a client or a server, using Agent-to-Agent or Model Context Protocols) and immediately begin testing against a simulated counterpart to see how they negotiate a path to a decision. This is an experiment in building more adaptable and intelligent health IT infrastructure. If you're interested in the intersection of LLMs, interoperability, and real-world healthcare processes, come and roll up your sleeves! ## Learn more and get involved: Connectathon Track Details: https://lnkd.in/g-cnk4U9 Intro to Banterop Platform: https://lnkd.in/giGFgEgJ

161

36 Comments

12 days of the CMS ACCESS Model Continues! ACCESS runs on FHIR R4.0.1. The ACCESS Model mandates four APIs for every participating organization: Eligibility API — Verify Medicare coverage and model eligibility before enrollment Alignment API — Enroll beneficiaries to your organization with documented consent BCDA API — Pull Medicare claims data (Parts A, B, D) for care coordination Reporting API — Submit baseline and quarterly outcome measures to CMS The Reporting API is where I think most organizations will face technical gaps. CMS requires LOINC-coded FHIR Observations with source-verifiable timestamps. Blood pressure readings need to transmit directly from validated devices, NOT manual entry. PROMs need structured submission as coded resources. I think this model is a serious boost for platforms that have optimized for compliant patient report outcome measure collection through multimodal capture. This self-reporting stack for patients is how you get paid in this model, and I think a lot of organizations have figured this out manually depending on their clunky system of record to capture this information. Payment reconciliation runs on these FHIR submissions. If your quarterly BP data doesn't arrive as properly structured Observations, the payment adjustment calculation fails. I get the sense many remote monitoring platforms built proprietary dashboards, not standardized FHIR APIs. Most outcome data lives in systems that can't auto-populate FHIR resources. BCDA integration requires OAuth2 workflows and bulk FHIR processing many teams haven't implemented. 📹 90-second technical overview below I hope you enjoy the Techy Surgeon explainer below. I have a lot of fun making these 🤣 . I have more animated short explainers like this for almost all of the articles in this series. Article Link in the comments. I also have links to the available FHIR APIs from CMS in the article (some are pending). I have a technical integration details HTML that we'll also be sending out tomorrow. Special thanks to Hadi Javeed, my CTO and Co-Founder at RevelAi Health for breaking this down. If you're evaluating ACCESS participation, I'm hoping this series and the resources we've put together will save you significant time. Please reach out if you have questions or have other collaborative resources you think we should add on TechyPolicy.com and the Techy Surgeon SubStack!

46

11 Comments

Aditya Narayan and I are proud to share a Health Affairs Scholar piece on CMMI's new ACCESS payment model. ACCESS will pay for care delivered by AI (with or without a clinician) for Medicare patients. This is the first time ever Medicare will pay for care without a person in the loop. It is for chronic-disease management: recurring, condition-specific payments tied to clinical improvement, rather than billing for discrete encounters or tightly defined remote-monitoring activities. The promise is straightforward, more reimbursement for technology-enabled longitudinal care and more choice and competition for patient care. It could lead to lots of startup and innovation. It could also lead to lots of hard to cancel subscriptions and FWA. There are still many unanswered questions about payment levels, how quality is measured, how risk is adjusted, and how these patients’ facing technology enabled services will coordinate care with the traditional delivery systems. https://lnkd.in/grDfXDH5

148

23 Comments

In the latest episode of FYI – For Your Innovation, Rohit Agarwal sits down with Uma Ratnam Krishnan to discuss how we’re building a culture where bold ideas, empathy, and collaboration drive real impact. From empowering teams to take thoughtful risks to partnering with startups through platforms like CNBC TV18 HealthX Elevate, we’re shaping the future of healthcare. Watch the episode to see how innovation is being prioritized across Optum Global. http://spr.ly/6041AiCDc #ForYourInnovation #OptumGlobal #HealthcareInnovation

15

India just moved from "AI Vision" to "Sovereign Execution." 🇮🇳 With the formal launch of IAIRO in New Delhi and Minister @Ashwini Vaishnaw’s announcement of India AI Mission 2.0, the mandate is clear: We aren't just building chatbots; we are building a national "ISRO for AI." We are moving from "Renting" general-purpose models to "Owning" Sovereign Intellectual Property. As an Architect, what excites me most isn't just the 38,000 GPUs or the $70B investment. It’s the focus on Neurosymbolic frameworks and Compact Domain-Specific Models. In 2026, the "Sovereign Architect" knows that "Bigger" isn't always "Better." The real enterprise value lies in: Domain-Specific Sovereignty: Models built for Indian Healthcare, Finance, and Public Infra that run on cost-effective, local infrastructure. Architectural Integrity: Integrating logic-based AI with neural networks to ensure mission-critical reliability (The Neurosymbolic Shift). The Agentic Mesh: Moving beyond point solutions to build an autonomous enterprise fabric that actually takes action on data. Leaders like @Aditya Agarwal have long championed that "Your CTO Should Actually Be Technical." Never has this been truer. To lead in this Mission 2.0 era, you can't just manage a vendor; you have to understand the First Principles of the stack you are building. Are we ready to stop being "AI Consumers" and start being "AI Architects"? The next $70 Billion depends on that answer. #IndiaAI #IAIRO #SovereignAI #EnterpriseArchitecture #DigitalIndia #TechLeadership #SouthParkCommons

1

1 Comment

Big Step, Daniel! SAMA’s 2025 momentum is clear: Version 1.0 of the use cases, risk evaluation framework, and implementation guide is now adopted. Participation includes 170+ participants across 35 states. Next up is a model state Medicaid AI policy aligned to NIST and built for real program operations. If your team wants a walkthrough or to learn more about SAMA, message me or Daniel and we'll set it up. #Medicaid #AI #ResponsibleAI #NIST #SAMA #AIinHealthcare

11

1 Comment

Raheel Retiwalla, Chief Strategy Officer, Productive Edge, provides the best explanation and examples of how agentic AI can be used to power healthcare workflows! 🤖🦾🏥 #AgenticAI #AIinHealthcare #AutonomousAI #HealthTech #FutureOfHealthcare #AIforGood #HealthcareInnovation #ResponsibleAI #SmartHealthcare

5

Both Epic and Oracle Health recently announced incorporation of AI into current and future EHRs. 𝘊𝘰𝘰𝘭 𝘚𝘵𝘶𝘧𝘧 𝘕𝘰𝘸: 𝘈𝘐 𝘊𝘩𝘢𝘳𝘵𝘪𝘯𝘨 (https://lnkd.in/g9dRe68x), a video by Bradley Fox, MD, Director of Clinical Informatics, announced a number of current and planned future AI enhancements to Epic's widely used hospital EHR that he says will "simplify tasks across the health system from the back office to the point of care". I found these interesting: Clinical summaries include care that took place elsewhere. Patient notes will "translate medical jargon into approachable language". Future patient facing capabilities will explain test results, explain billing balances, answer questions and help make appointments. It extracts findings and recommends follow-ups from radiology reports. He says the ambient dictation "is smart enough to skip details that aren't relevant". In the future, it will "cue up orders and diagnoses based on the conversations". It helps nurses "with their flowsheet documentation" and "end of shift notes". Finally, for billing, it suggests codes and can draft appeal letters when claims are denied. Oracle Health just introduced its new Health EHR 𝘧𝘰𝘳 𝘱𝘩𝘺𝘴𝘪𝘤𝘪𝘢𝘯𝘴 in which, it says, "AI is seamlessly embedded into every layer —from data and services to the user interface—to create a scalable, integrated solution for personalized, data-driven care." I found these interesting features reading the Oracle product brief (https://lnkd.in/gAPTunZx) which is clear that it includes future features. It is integrated with health information exchanges, including Oracle Health Seamless Exchange, Oracle Health Data Intelligence, and TEFCA Qualified Health Information Networks (QHIN) "to retrieve and consolidate longitudinal patient records". The list of current QHINs is at https://lnkd.in/g85B4CBb. It includes Epic Nexus and Oracle Health has been "accepted into the project planning and testing phase". Physicians can interact with the EHR's embedded Clinical AI Agent 𝘣𝘺 𝘷𝘰𝘪𝘤𝘦 "across multiple functions, including chart review, documentation, and task automation". From the day's schedule it "provides a 'fish-eye view' AI-generated summary of each patient's visit purpose and relevant background, with highlighted sources". I've written about Retrieval Augmented Generation or RAG which Oracle apparently plans to use (although the brochure is aimed at physicians and doesn't say this) "for the clinician to be able to use voice commands to pull near real time, patient-specific insights from a multi-source catalogue, seamlessly integrating clinical intelligence into the workflow." Sources will include drug databases, clinical guidelines, pharmacy records, and immunization histories. I've long advocated offering physicians a timeline view of their charts, and Oracle says its new EHR will provide one.

34

2 Comments

🚀 Have you seen what ARiSE is working on? Best in class benchmarking for healthcare AI: impartial, independent and patient oriented within actual clinical scenarios. This is the work of numerous experts, including Ethan Goh, MD, Jonathan H. Chen & Adam Rodman, and it has never been more important. Check out my full article covering this for Forbes 👇

14

1 Comment

Google CEO Sundar Pichai takes action to crack down on the misuse of #AI As someone whose professional life centers on navigating the intersection of healthcare, AI, and commercialization — this conversation highlighted the need for strategic thinking and implementation when using #AI. We are at a turning point where #AI innovations bring immense promise — but also real responsibility. As Sundar underscores, it’s not just about building transformative technology — it’s about ensuring misuse doesn’t outpace intent. In healthcare, the stakes are even higher: patient safety, privacy, ethical use of data and algorithms — these aren’t optional enhancements; they’re foundational. It’s up to executives, policymakers, and industry leaders alike to embed guardrails now — so innovation doesn’t compromise trust or safety. If we succeed, the next decade could deliver breakthroughs in diagnostics, personalized medicine, and global access. When Sundar Pichai talks about cracking down on AI risks, the implications for our industry are enormous. Healthcare is now the sector most vulnerable to misinformation, model hallucinations, and misapplied algorithms — because the stakes are literally life and death. Here’s the uncomfortable reality we don’t say often enough: Most healthcare organizations are adopting AI faster than they are preparing for the consequences. Everyone wants AI for speed, scale, and efficiency… …but few have real guardrails, governance, or clinical validation in place. In our world, misuse isn’t just political or social — it can mean: -Wrong treatment pathways -Misread diagnostic patterns -Fabricated clinical insights -False patient communications -Compromised medical decision-support -Growing distrust across the entire patient journey Healthcare leaders must stop treating AI risk as an IT issue. It’s an operational, clinical, and ethical imperative. We need stronger validation protocols, tighter governance, and far more transparency before these tools touch patients, providers, or payers at scale. If we get this right, AI will transform diagnostics, care coordination, revenue cycle, adherence, and global access. If we get it wrong, the fallout won’t be theoretical — it will be measurable in outcomes. Are we moving fast enough to protect the system we’re trying to modernize? Fox News MediaFox News TVFox Business NetworkShannon Bream Molly Line #AliRaja Lila Sciences Andrew Beam Massachusetts General Hospital Brigham and Women's HospitalGoogle DeepMindAnthropicChat GPT.ChatGPTAdvaMedScott Whitaker Biotechnology Innovation OrganizationJohn F. CrowleyJyoti Gupta, PhDGE HealthCarePhilips https://lnkd.in/esDkdi2N

4

Listening to Vinod Khosla talk about AI in mental health, and he nailed something that's been our vision: Heard Nithin Kamath and Vinod Khosla on a podcast - People by WTF | Episode 12. They speak about mental health and AI at 55:00 The scarcity problem is real. India doesn't have enough therapists. People are embarrassed to seek help. Access is the bottleneck. AI can solve that at scale. But here's where I think the conversation is incomplete. Vinod talks about AI diagnosing conditions (93% accuracy on talk-based mental health). That's real. The classification part? AI can do that better than most human therapist. It doesn't stop there, diagnosis isn't recovery. That's where it gets interesting. And that's where we're building differently. Here's what we're seeing: A patient gets diagnosed. Let's say depression. Now what? They need a journey. A structured path forward. Because recovery isn't a moment—it's a process. A therapist would say: "Here's what we're working on together. Here's what you do between sessions. Here's what I'm monitoring." But between sessions? That patient is alone. Seven days. With triggers. With doubt. With old patterns. That's where recovery breaks. This is what we're building at Cadabams. Not just diagnosis. A complete recovery journey. First: Baseline assessment (AI does this accurately). Then: Personalized recovery journey (research-backed, structured). Then: Daily micro-tasks, reflections, tracking (the in-between support). Then: Agent guidance every single day (continuous reinforcement, adaptation). Then: Doctor gets complete visibility (can adjust, monitor, plan next session). It's not AI replacing the therapist. It's AI making the therapist infinitely better. The vision Vinod's talking about free, accurate, accessible mental health at scale? We're building that foundation. Today: We're bridging the gap between episodic therapy and continuous care. Tomorrow: AI as complete recovery partner, with doctors guiding at the right moments. Eventually: What Vinod said. Quality mental health accessible to everyone in India. At scale. Affordable. The path isn't "AI takes over." It's "AI handles what it's good at (consistency, presence, data), clinicians handle what they're good at (relationship, judgment, complexity), and together they change outcomes." That's not the dystopian AI story. That's the real accessibility story. https://lnkd.in/e_CGBqZD Sandesh Cadabam Neha Cadabam

9

A must read article on AI in healthcare from the Indian context, penned by my colleague Dr Arun How AI is a tool and how to master training it to be a servant and not the master!

17

The Evolution of MS-DRG Complexity Factors: Examining CC/MCC Trends MS-DRG, which plays a central role in inpatient reimbursement, gets updated annually by CMS. As I download and analyze all the versions from the last 15 years at mimilabs, I've discovered some interesting patterns. One fascinating aspect of DRG is the CC and MCC (Major) designations - these characterize whether the patient had other comorbidities on top of the base DRG, often resulting in higher reimbursement rates. The chart below shows the number of unique ICD10 diagnosis codes assigned to CC (not MCC) per version. Interestingly, the universe of CC has been steadily expanding over the years. I wonder if this continuous expansion of CC designations has unintended consequences for inpatient reimbursements. Are hospitals seeing appropriate payment increases reflective of true clinical complexity, or could this trend potentially impact the healthcare system in unexpected ways? What are your thoughts on how these designations should be determined to ensure fair reimbursement that aligns with actual resource utilization?

38

3 Comments