Stephen Trask, MT

Heading home from J.P. Morgan Healthcare Conference and one sentiment kept surfacing: anxiety around large language models disrupting healthcare software and services. Shares of RCM vendors and other vertical software providers were off based on LLM disruption narratives. But I'm not buying it. Here's what I know: healthcare is hard. Big tech companies have tried to enter this space by brute force before—most fell flat. The ones that stuck around had to acquire their way into relevance, not innovate their way in. Will large LLMs drive efficiency? Absolutely. Will they improve outcomes? Yes. But they'll do it alongside or in partnership with purpose-built healthcare software and services, not by replacing them. Companies that integrate LLM capabilities thoughtfully into existing workflows will thrive. The market is repricing based on fear. 25 years of watching healthcare evolution tells me: specialized vendors aren't going anywhere. They're adapting. #ICRHealthcare #JPM2026 #DigitalHealth #HealthcareIT #AI ICR Healthcare

37

3 Comments

Starting over a decade ago, cloud and data-consumption-based pricing change the economics of technology. But we watched lots of IT leaders keep making decisions with an on-prem mindset. In many cases, leaders didn’t even absorb the fact that forecasting was necessary (let alone, how to do it well). Sure, if you asked them "do you understand that this will cost money?" they would say "yes." But they were making decisions that said otherwise. This created a financial mess in many enterprises. It’s how companies got thousands of dashboards refreshing every hour, 24/7. It's how they got monthly reports built on near-real-time architectures. It's how they ended up with contracts signed that paid no attention to the ongoing cost of data movement, compute, and storage (and sometimes costing more that the revenue a contract brought in!) This is why I became obsessed in recent years with FinOps. (Also, I’m weird like that.) And I wasn’t alone: I’ve seen more attention to cost with cloud and data platforms. Hurrah…. Bravo! But... uh oh…. … now we have AI. To paraphrase Mark Twain: history may not be repeating itself exactly, but it sure is rhyming. Most companies aren’t paying close to enough attention on modeling cost. Some aren’t even acknowledging that you need to do so. This is a great discussion posted by Connor Deeks and Codestrap. My favorite line, which serves as a major theme: “Variable costs are this new thing that tech people thought was never going to exist in our industry… we need to adjust to this new reality.”

3

1 Comment

Clinical governance for AI in Healthcare is ever more urgent as Chatgpt5 and others increasingly focus on consumer/patient facing use cases. Clinical leaders must step forward to ensure safety, rigor and insight are not compromised or lost . HealthcareAI #AIGovernance #PatientSafety #ClinicalDecisionMaking #GPT5 #DigitalHealth #HealthTech #MedicalAI #HealthcareLeadership #AIEthics #HealthSystemStrategy #ClinicalOversight

12

1 Comment

Data quality, vetting and privacy are long standing issues and a critical pain point in AI. Building systems that correct for these basic flaws requires meticulous attention…but it can be done.

5

Explainable AI needs clear data context. #MDM provides semantic models, speeding clinician trust by 58%. Make AI transparent. Ready for trusted decisions? #Profisee #ExplainableAI #MDM

1

In this article, Biorasi's Evgenii Cherepanov, PhD makes an essential point: Data maturity is a necessity for AI. "Organizations that are genuinely ready for automation share one simple property: their work can be described. Their processes have edges, their decisions have principles, and their data has shape. "You do not have to guess what a spreadsheet means. You do not rely on the unspoken assumptions of a single individual. You do not see the same process implemented differently in every project. "You do not treat data as something produced for reports, but as the raw material from which operations are built." Well said, Evgenii!

28

OpenAI’s reported findings that GPT-5.4 outperformed specialty physicians on difficult clinical reasoning benchmarks should not be viewed as just another AI milestone. It signals the beginning of a much deeper governance problem: What happens when the system is statistically superior, but institutionally non-accountable? In medicine, accuracy alone has never been enough. We also require: * traceability, * explainability, * reproducibility, * liability, * and the ability to challenge a decision before harm occurs. That is why this moment is not simply about whether AI can outperform clinicians on benchmark reasoning tasks. It is about what happens when a clinician disagrees with a system that is demonstrably more accurate than they are. How does accountability function then? Aviation confronted a version of this problem decades ago. Autopilot systems routinely outperform humans in specific operational tasks, yet the industry did not remove pilots. Instead, it redesigned accountability around human-machine collaboration, layered oversight, escalation protocols, and procedural control. Healthcare will likely move in the same direction. The real danger is not “AI replacing doctors.” It is clinicians gradually becoming passive approvers of systems they no longer meaningfully interrogate or fully understand. That is not augmented intelligence. That is outsourced judgment. #AI #HealthcareAI #MedicalDevices #ResponsibleAI #AIGovernance #ClinicalAI #HealthTech

1

Most AI companies brag about how much data they can process. We measure success differently. How quickly can a clinician get to the one insight that matters? Because processing 10,000 data points means nothing if the doctor still can't answer: "What should I do for this patient today?" Speed to clarity > volume of data. #HealthcareAI #PrecisionMedicine

11

Here’s an excellent use case for AI in healthcare: Patient-Reported Outcomes (PROs) have long been a valuable tool for understanding symptoms, function, and quality of life. But traditional methods (structured questionnaires and fixed scoring) often miss the complexity of how people actually experience health. Generative AI changes that. Interpreting patient narratives in real time allows us to capture subtle, multidimensional insights that rigid forms can’t, transforming lived experiences into actionable data. This evolution can help clinicians better personalize care, close gaps in understanding, and bring us closer to truly patient-centered healthcare.

5

Excited to announce Community Care Connection Inc. (C3) new partnership and integration with OneChart , a partnership that brings together two powerful solutions with one shared goal: giving clinicians their time back. OneChart is an AI-driven clinical scribe designed to lift the documentation burden from healthcare professionals. By transforming real-time conversations into accurate, structured clinical notes, it’s redefining how clinicians chart — faster, more consistently, and with far less cognitive load. Pairing this with C3’s workflow ecosystem creates a seamless experience where intake, documentation, communication, and follow-up live in one intelligent platform. Together, we’re helping clinics and healthcare organizations: ✅ Reduce administrative fatigue ✅ Improve documentation accuracy and compliance ✅ Enhance clinical workflows through automation ✅ Free clinicians to focus on what matters most — their patients Watch the demo: https://lnkd.in/eRDMgHsb Glad to be working with these partners that prioritize innovation, efficiency, and better care experiences for providers and patients alike. Shreyas Suri Paolo Aiello Excited for what’s ahead. 💡 #innovation #AIscribe #healthcare #mentalhealth #alliedprofessionals #AI #clinicalintelligence Sign up for a demo below!

13

If AI is in your workflow, oversight belongs there too. Because once AI is in your workflow, it starts impacting real outcomes: → client advice → patient care → legal decisions → financial actions AI GRC is not just technical. It is operational. It is how the work actually gets done. Who is accountable when the AI is wrong? Who reviews the output before it impacts someone? What happens when it fails? These questions do not show up in a demo. They show up when it matters. Protect the people you serve, including yourself.™ #iGovernAI™ 🤓❣️ #GovernBeforeYouAutomate™ #GovernBeforeYouInnovate™ #EEEAIGRC™ #AIGRC #AIGovernance

1

AI is quietly changing the economics of data. For years, more data = more value. In AI/ML, more exposure = more risk, cost, and complexity. That tension is becoming the bottleneck. It’s no longer about collecting more data. It’s about how little of it you actually need to use. Because the future of AI may not be built on access to data… but on minimizing it. More in this interview with Pulse 2.0: https://lnkd.in/gvvCtPS4 #AI #MachineLearning #DataSecurity #AISecurity #CIO #CISO

13

NBC News recently visited Mass General Brigham (MGB), where physicians using ambient documentation technologies reported less after-hours work, greater satisfaction, and a 21% reduction in burnout within 84 days.      The segment explores research from Emory Healthcare and MGB on tools like DAX Copilot, now part of Microsoft Dragon Copilot. There's no greater motivation for innovation than hearing from doctors like Rebecca M. that AI is helping restore joy in patient care: https://lnkd.in/d58h8_tp 

43

4 Comments

🔍 Why Trust Is the Most Important Metric in Clinical AI (And Why Accuracy Isn’t Enough) We’re hearing more than ever about AI diagnosing diseases, predicting patient outcomes, and even outperforming doctors in controlled studies. But here’s the uncomfortable truth: no matter how accurate an AI system is, if clinicians don’t trust it, it won’t be used. In our latest research published in Diagnostics (https://lnkd.in/e_uiCSKh), we dug into one of the biggest roadblocks in healthcare AI: trust — not just in the model, but in how it communicates its confidence, reasoning, and limits. 👉 Imagine an AI tells a doctor: “I think this is heart failure.” Now imagine it says: “Based on elevated BNP, pulmonary edema on imaging, and progressive dyspnea, I’m 94% confident this is heart failure. Here’s why.” Which one would you trust more? That’s the power of transparency, confidence calibration, and explainability — and our study shows they aren’t just nice-to-have features. They’re essential for real-world adoption. Using over 6,600 cardiovascular cases from the MIMIC-III database, we built a dynamic scoring framework that combines: 🔹 AI-reported confidence 🔹 Semantic similarity to real clinician diagnoses 🔹 Transparency of the AI’s reasoning 🔹 Output clarity (yes, even length matters!) The result? We reduced unnecessary clinician overrides from 87.6% down to just 33.3% — not by making the AI smarter, but by making it more trustworthy. 💡 Key insight: When AI expresses high confidence and provides clear, logical explanations, clinicians are far more likely to accept its suggestions — safely and efficiently. But when explanations are weak or confidence is low, the system raises the bar for acceptance, ensuring clinicians stay in control. This isn’t about replacing doctors — it’s about designing AI that supports them like a thoughtful colleague, not a black-box oracle. As we move toward AI-integrated care, this lesson is clear: ✅ Accuracy matters. ✅ Speed matters. ✅ But trust is non-negotiable. And trust isn’t earned through performance alone — it’s earned through clarity, consistency, and collaboration. ➡️ Read the full study: https://lnkd.in/e_uiCSKh #AIinHealthcare #ClinicalAI #ExplainableAI #DigitalHealth #MedicalAI #ArtificialIntelligence #HealthTech #MachineLearning #PatientSafety #FutureOfMedicine #DecisionSupport #TrustInAI — 💬 I’d love to hear your thoughts: How do you think we can build AI systems that clinicians will not only use — but truly trust? — Proud to have led this research with an incredible team. Special thanks to my co-authors for their dedication to responsible, human-centered AI in medicine.

10

2 Comments

AI adoption isn’t just a project. It is a strategic shift. In our latest Twinning Strategy episode, Jeff Sigel, Eric Sigel and I sat down with Stephen Cooper, an enterprise AI implementation consultant, to unpack what’s really happening inside organizations racing to adopt AI. Here’s the truth: most companies focus on productivity and efficiency and invest in tools. Only few invest in the strategy and change management needed to make the tools work. AI adoption is not a one-time upgrade. It will be an ongoing transformation that will require new skills, new mindsets, and continuous learning. When companies start treating AI as a new way of working - that’s when the real shift will happen. Watch the episode (link in the first comment) to hear the full discussion. #AIAdoption #EnterpriseAI #DigitalTransformation #FutureOfWork #ChangeManagement #LeadershipInAI #AIIntegration #AITransformation

30

14 Comments

For everyone following this conversation on Implementing AI in Healthcare, mark your calendars for October 8th. Digital Medicine Society (DiMe) is launching The Playbook: Implementing AI in Healthcare, a foundational guide for the field. Michael Gorton Shauna M. Overgaard, Ph.D. Ann Allen Sreekanth Chaguturu Michelle Longmire M.D. https://lnkd.in/gp8nYyi2

29

1 Comment