Brian McCarthy

SQL on FHIR is going to be a big deal for anyone feeding FHIR data into analytics, research, and quality improvement.

130

5 Comments

Worth a read: Moving healthcare from siloed, batch data to real-time, multimodal, AI-ready architectures—so AI can reason and act, not just run point solutions. It nicely highlights how scaling AI depends on modernizing the data backbone: unified pipelines, strong governance, and constructs like knowledge graphs that make context truly usable. Energized to help accelerate this journey at Mayo Clinic. #HealthcareAI #DataArchitecture #DataGovernance https://lnkd.in/gfdmXtEK

48

1 Comment

This McKinsey article provides a comprehensive overview of how generative AI is transforming the healthcare landscape, from optimizing logistics to revolutionizing patient engagement. 85% of surveyed healthcare leaders say they’ve already adopted or are actively exploring gen AI. Many have moved beyond pilot phases and into real implementation, especially for administrative efficiency, clinical productivity, and infrastructure gaps. The potential is tremendous. But as always, it’s complicated. Yes, AI can cut costs and boost outcomes. But success hinges on more than the tech itself. It demands thoughtful partnerships, responsible governance, and a willingness to evolve how care is delivered. And while it’s tempting to believe AI can “fix” healthcare’s deepest inefficiencies, we must resist reductionist thinking. Just as medicine isn’t only about metrics, gen AI isn’t a magic wand. Used wisely, it’s a powerful tool. Used blindly, it risks widening gaps and embedding bias. The challenge ahead? Implement AI with humanity so the systems we build reflect not just innovation, but care. Follow me for foresight on AI, health, and the future of work. https://lnkd.in/d8ur5eEw

44

6 Comments

Anthropic released #Claude for #Healthcare today at #JPMHealthcare. Looks like a thoughtful set of agents with connected, commonly used data assets. Here's a summary from all four corners of the healthcare game board. **Providers** Claude offers HIPAA-ready infrastructure that enables healthcare organizations to deploy AI for workflows involving protected health information. - **Administrative efficiency**: Speeding up prior authorization requests and helping with patient care coordination to reduce pressures on clinicians’ time - **Database access**: Connections to ICD-10 codes for medical coding and billing accuracy, the National Provider Identifier Registry for credentialing and network management, and PubMed’s 35+ million biomedical articles for literature reviews - **Interoperability tools**: FHIR development agent skills to help connect healthcare systems faster with fewer errors - **Claims support**: Accelerating insurance appeals with structured reasoning and evidence synthesis **Patients** Claude can summarize patient-uploaded health records, explain test results in plain language, detect patterns across fitness and health metrics, and prepare questions for appointments . Additional features include: - Integration with Apple Health and Android Health Connect (rolling out in beta for Pro and Max subscribers) - Private by design with explicit opt-in required, and health data excluded from model training - Help navigating complex medical reports and understanding patterns over time **Pharma & Life Sciences** New capabilities for life sciences extend from preclinical R&D into clinical operations and regulatory affairs , including: - **Clinical trials**: Drafting clinical trial protocols and tracking trial indicators - **Regulatory submissions**: Helping with regulatory submissions so that more life saving drugs can come to market faster - **Scientific research**: Deep integrations with scientific databases and lab tools, with biosafety guardrails built in - Improved performance on spatial biology analysis and other life sciences benchmarks with the Claude Opus 4.5 model **Payers** Connectors to the Centers for Medicare & Medicaid Services Coverage Database enable Claude to verify locally-accurate coverage requirements and support prior authorization checks . Additional capabilities include: - Building stronger claims appeals - Support for revenue cycle and compliance teams - Prior authorization review agent skills that can be customized to organizational policies All standard #AI disclaimers apply. Always protect your #Data. https://lnkd.in/gZxQHd7h

12

A major challenge in today’s enterprise landscape is navigating the rapid, often disruptive, evolution of AI. This demands a strategic approach, moving beyond hype to understand the practical implications for leadership, operations, and talent. A recent Gartner report, “Over 100 Data, Analytics and AI Predictions Through 2031,” provides a critical lens on these shifts. It outlines the key trends that will define the next decade for data and analytics leaders. The report highlights several key areas of transformation: • The Strategic Imperative for the CDAO: The role of the Chief Data & Analytics Officer is at a critical inflection point. The report predicts that by 2027, 75% of CDAOs who have not established themselves as indispensable strategic collaborators in their organization’s AI success will lose their C-level position. This underscores the shift from a technical leader to a core business strategist. • The Evolution of Analytics and Business Intelligence: Traditional methods of data consumption are becoming obsolete. Gartner anticipates that by 2028, 60% of existing dashboards will be replaced by GenAI-powered narrative and visualization. Simultaneously, by 2027, 50% of business decisions will be augmented or automated by AI agents, fundamentally changing how we interact with and act on data. • A Paradigm Shift in Talent and Skills: The demand for specific AI expertise will reshape the talent market. The report projects that by 2027, there will be three times more open positions for AI engineers than for data scientists. This signals a move from custom model building to a focus on integrating, scaling, and optimizing pre-trained AI solutions. • New Frontiers in Security and Governance: The proliferation of AI introduces new risks. According to the predictions, 40% of AI-related data breaches will be caused by the improper use of GenAI across borders by 2027, and 25% of all enterprise breaches will be traced back to AI agent abuse by 2028. This necessitates a renewed focus on data governance and AI security. I believe that for leaders, internalizing these predictions is crucial. The path forward requires not just adopting new technologies, but fostering a culture of AI literacy, building adaptable strategies, and proactively addressing the associated risks. Source: Gartner, “Over 100 Data, Analytics and AI Predictions Through 2031,” July 2025. #AI #DataAnalytics #Gartner #DigitalTransformation #Leadership #BusinessStrategy #CDAO #FutureOfWork

16

2 Comments

Unleashing AI for Clinical Data: GPT-4o Powers 3,458 Patient Records into i2b2 Key Takeaways: • 3 disparate data models (three synthetic AnVIL datasets) automated into a unified pipeline [Ensures broad applicability across spreadsheet and bespoke formats] • 3,458 fake patient records ingested seamlessly into i2b2 via ChatGPT-4o–generated Python scripts [Delivers immediate cohort discoverability for researchers] • Two-stage transform: data dictionaries → intermediate common format → i2b2 ontology [Guarantees semantic consistency using NCBO’s standard terminologies] • Original clinical files converted into i2b2 facts with no manual coding tweaks [Minimizes human error and accelerates deployment] • Prompt reuse slashes incremental setup for each new dataset [Drives scalable ETL workflows and substantial time savings] #ClinicalData #ETL #LLM Link: https://lnkd.in/edEvSzid

1

Enabling AI and discipline do not always coexist. Many technology leaders often approach AI as experimentation rather than starting with solid use cases grounded in business strategy. This approach contributes to AI governance frequently being a "catch-up" activity. To enable AI effectively—driving value, enhancing business outcomes, and mitigating risk—a disciplined approach is essential from the beginning. Transformation and governance must work hand in hand to achieve success. https://lnkd.in/emGzGDEn

39

4 Comments

Most AI failures in healthcare don’t come from bad models. They come from bad operating assumptions. This executive Q&A from Sam Obeidat, MGM makes one thing unmistakably clear: AI transformation is not a technology problem. It’s a leadership, operating model, and governance problem. A few signals that stood out: AI use cases shouldn’t be copied but they should be discovered. High-performing hospitals don’t start with a wishlist. They start with leadership alignment on outcomes, map real workflows, identify decision bottlenecks, and then prioritize a small number of use cases for 90-day delivery. Anything else is guessing. Infrastructure follows use cases—not the other way around. The minimum data, integrations, and security posture depend entirely on what you’re building. Teams that “build the platform first” often stall before value appears. Speed comes from focus, not shortcuts. A serious pilot is a 90-day sprint with clear ownership, metrics, and human-in-the-loop design. “Low risk” means narrow and controlled, not casual. Adoption lives or dies in the workflow. If AI adds steps or context switching, clinicians will reject it. The only viable path is embedding AI where it removes cognitive load and eliminates low-value work—with clinicians involved as co-designers from day one. Accountability must be explicit. Every AI use case needs a named owner, defined decision boundaries, and documented escalation paths. If it can’t be governed, it shouldn’t be deployed. Privacy handled “later” is already a failure. Access control, audit logs, data minimization, encryption, and regulatory alignment must be designed in from the start. Retroactive compliance doesn’t scale. Success isn’t demos—it’s repeatability. In the first 3–6 months, success is proven by validated KPIs and the ability to safely repeat the use case across teams and sites. One-offs don’t count. The throughline across all answers is consistent: Hospitals that scale AI successfully treat it as an operating capability, not a project. Start with leadership, governance, and cadence—and the right use cases become obvious.

1