Dr. Meenakshi Jhala

New guidance on the use of AI scribing tools has now been released. - gives a good overview of the tech - outlines benefits but there’s no link to evidence base of these benefits!! - no acknowledgment of the potential downsides & mitigation - there’s a lot of expectation on staff but surely some of this could (should!) be embedded as part of the tech https://lnkd.in/eY77pYK7

134

26 Comments

Explainable AI for Prediction of Patient Deterioration at University College London Hospitals NHS Foundation Trust, led by Julia Ive and Professor Hani Marcus. Nice to see CogStack AI being used to support developing explainable predictive AI based on verifiable clinical knowledge. Out now in Nature Portfolio Digital Medicine: https://lnkd.in/etciJzZp

111

6 Comments

British Medical Association: Principles for #AI Artificial Intelligence (AI) and its application in #healthcare AI in healthcare includes a range of applications aimed at enhancing efficiency, diagnosis, and treatment. Though AI lacks a single definition, it broadly refers to technologies simulating human intelligence to perform complex tasks, in the context of healthcare this includes: healthcare administration, clinical decision-making, improving diagnostics, personalised treatment, providing digital therapies, population health data analysis, and biomedical research. The BMA advocates, as set out in the principles at the end of this paper, for AI in healthcare to prioritise safety, efficacy, ethics, and equity. Each AI implementation must be rigorously assessed in real-world settings and continuously monitored to ensure it improves care quality and job satisfaction without exacerbating inequalities. Strong governance and up-to-date regulation are essential to protect patient safety. Involving staff and patients in AI development and providing them with the choice to opt out or dispute AI decisions is crucial. Training for healthcare professionals and robust IT infrastructure are necessary for effective AI integration. Legal liability must be clear, ensuring developers are accountable and doctors can challenge AI decisions. This briefing sets out how AI is being used in healthcare delivery currently or in the near future and examines both the potential benefits and drawbacks with respect to patients, clinicians, and the efficiency of the overall healthcare system. We conclude that AI in and of itself is only a tool – if implemented appropriately and effectively, it can improve outcomes; if not, it can worsen them. Therefore, we also set out some principles the BMA believe must be adhered to for AI to safely, effectively, ethically, and equitably improve healthcare outcomes.

151

2 Comments

Finally got a moment to put down some reflections from the HIMSS #AI in #Healthcare Forum 2025 in NYC. In the first place, why attend a healthcare AI conference at all when there's so much content online and every other post is from a self-proclaimed AI expert? Because sometimes, you need to cut through the noise. You need to hear from people actually doing the work — building tools, failing fast, leading change inside hospitals, startups, and health systems. You need real talk, not hype. That’s exactly what this forum offered: genuine, candid conversations and a chance to see AI tools in action — what works, what doesn’t, and what still needs fixing. Here are some of my #reflections on achieving value at scale with AI in Healthcare: 1. You don’t have an AI #strategy if you don’t have a #data strategy. Strong data foundations are essential — before anything intelligent can happen. 2. Start with #WHY. Every AI initiative needs purpose, measurable outcomes, and an exit path if it doesn’t deliver. 3. #Trust is earned, not assumed. We can’t look AI in the eye — it must earn trust through reliability, #explainability, and workflow harmony. 4. #Startups: Build with someone who’s walked the floors. Deep workflow knowledge beats generic AI skills every time. 5. #Governance first. AI second. Unlike other tech, early adopters of AI can run ahead of governance. That’s why our governance models must evolve fast — to match new behaviors and risks. No scaling without ethics, model drift monitoring, and transparent decision logic. 6. Avoid long-term #vendor lock-in. Unlike legacy tech, AI contracts should be short, flexible, and tied to utilization — not a 5-year leap of faith. 7. #Ambient is here. But #UX still breaks it. #Rage clicks are real. If a tool frustrates users, even good AI won’t survive adoption. 8. #Tech stack is nothing without a #human stack. From Minnesota’s playbook: Tech = data, models. Human = skills, governance, trust. Build both if you want adoption. 9. Choose use cases wisely. Start with #high-impact, #low-risk tasks (like ambient scribing or scheduling), then move toward high-risk, high-impact domains like clinical decision support. 10. #ROI isn’t just numbers. Burnout relief, cognitive ease, and clinician retention are just as valuable as hours saved. AI should give time back, not take more away. 11. The road ahead? Think ambient assistants, personalized nudges, passive monitoring — and tools that improve the #patient #experience too: managing unstructured data, PROMs, and supporting real-time feedback. These are just a few takeaways — there was so much wisdom from the trenches, shared by those leading the charge with curiosity, caution, and clarity. Kudos to the incredible speakers, vendors, and the HIMSS team for curating such an immersive, grounded, and forward-looking experience !

50

3 Comments

1 in 5 UK doctors use unregistered AI tools. I explore why in my latest article for Healthcare Today. The demand for AI in healthcare is real. Clinicians need: ⤷ AI assistants to cut through admin overload ⤷ Predictive models that reduce human error ⤷ Smart automation that streamlines processes Yet these tools are held back by: ⤷ Unfriendly infrastructure ⤷ Unclear regulation policies ⤷ Unsupportive local procurement rules If the UK is serious about modernising its healthcare system, it must treat AI as a lifeline, not a luxury. ➡️ Read the full piece in Healthcare Today: https://lnkd.in/dmpYfndD ➡️ Head to our website to learn more about Rhazes: https://www.rhazes.ai/

27

4 Comments

When Validic launched in 2013, the biggest thing we heard from physicians was, “We don’t need more data.” What a difference a decade makes… Today, we don't provide "more data", we deliver key insights clinicians need from wearables and in-home health devices to make smarter and faster decisions, right in the EHR. And physicians love it. They are saying things like: “Without Validic, we would have never been able to figure out this patient.” (Direct quote!) Connected health data from wearables and in-home health devices has advanced from 𝐥𝐢𝐟𝐞𝐬𝐭𝐲𝐥𝐞 𝐜𝐮𝐫𝐢𝐨𝐬𝐢𝐭𝐲 𝐭𝐨 𝐜𝐥𝐢𝐧𝐢𝐜𝐚𝐥 𝐧𝐞𝐜𝐞𝐬𝐬𝐢𝐭𝐲. Everyday connected health data gives clinicians greater insights, helps patients better self-manage conditions, and enables true team-based healthcare. Physicians no longer view connected health data as “one more thing to do.” Now the data is AI-enhanced and available in the EHR workflow. Everyday patient health insights are now critical to how they practice top of license and deliver the best care for their patients. This is the future of care — more data is creating smarter insights, which is leading to improved health outcomes so people can live their best lives.

80

4 Comments

🧠✨ A Breakthrough Moment: AI and the Future of Mental Health Care A new #randomizedcontrolledtrial, funded by Dartmouth College, tested an AI chatbot named TheraBot for adults experiencing clinically significant symptoms of depression, anxiety, and eating disorders. The results? https://lnkd.in/evtYDNYa 📉 Significant symptom reduction across all groups 📱 Over 6 hours of average use 🤝 Therapeutic alliance ratings comparable to human therapists This is the first RCT to demonstrate that a fully generative AI therapy chatbot can effectively treat clinical-level mental health symptoms. This is not just impressive—it’s transformational. 🧬 In Smarter Healthcare with AI, I highlight similar breakthroughs that are reshaping how we care for people’s minds and bodies. From adaptive AI companions supporting veterans with PTSD to scalable virtual therapists designed to meet the mental health crisis head-on, this field represents one of the most urgent and promising frontiers in modern medicine. Our mission at HumanCare Technologies, Inc. is clear: deliver solutions rooted in purpose, personalization, and partnership, using innovation to uplift the human experience. 💡 The promise of AI is not to replace human care but to expand access, deepen empathy, and bring healing to more people than ever before. If you’re as inspired by these results as I am, I invite you to join me in this work. Bruce Alan Kehr, M.D. | Heather Leigh Flannery | Sean Manion | Severence MacLaughlin, Ph.D | Qasim Butt, MD | Patrick Daugherty | Andrew D. Plummer, MD, MPH | Michelle Padgett | Caesar Junker MD, Col, USSOCOM (ret) | Peter Walker | Scott McKeithen | Tim Dasey | Robert Pearl, M.D. Together, we can build a smarter, more caring, and more accessible future for everyone, everywhere. #SmarterHealthcare #MentalHealth #AIinMedicine #HumanCare #DigitalTherapeutics #GenerativeAI #HealthcareInnovation #Therabot #ArtificialIntelligence #BehavioralHealth #HealthcareLeadership #AIForGood

38

4 Comments

How are large-scale AI models being used in healthcare? 𝐋𝐚𝐫𝐠𝐞 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐋𝐋𝐌𝐬) 𝐢𝐧 𝐌𝐞𝐝𝐢𝐜𝐢𝐧𝐞 LLMs such as MedPaLM process and analyse clinical text, including electronic health records (EHRs), medical literature, and patient histories. They assist in medical question answering, personalised treatment plan generation, and decision-making support for healthcare professionals. LLMs also contribute to medical education, particularly in preparing for exams like the United States Medical Licensing Examination (USMLE). 𝐕𝐢𝐬𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥𝐬 𝐟𝐨𝐫 𝐌𝐞𝐝𝐢𝐜𝐚𝐥 𝐈𝐦𝐚𝐠𝐢𝐧𝐠 AI-powered convolutional neural networks (CNNs) enhance diagnostic accuracy in skin cancer, lung cancer, and breast cancer detection. Vision-Language Models (VLMs) such as CLIP and SAM-based models improve anomaly detection in medical images, even in zero-shot settings. Diffusion models help generate realistic medical images and aid in pathological image synthesis. 𝟑𝐃 𝐋𝐚𝐫𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 𝐢𝐧 𝐌𝐞𝐝𝐢𝐜𝐢𝐧𝐞 These models use 3D convolutional neural networks (3D CNNs) to process volumetric medical data, such as CT and MRI scans. They enable tumor segmentation, virtual surgery simulations, and anatomical modeling, improving surgical planning and prosthetic design. 𝐌𝐮𝐥𝐭𝐢𝐦𝐨𝐝𝐚𝐥 𝐌𝐨𝐝𝐞𝐥𝐬 𝐟𝐨𝐫 𝐈𝐧𝐭𝐞𝐠𝐫𝐚𝐭𝐞𝐝 𝐃𝐢𝐚𝐠𝐧𝐨𝐬𝐢𝐬 These models integrate data from clinical text, imaging, and genomics to provide more accurate and holistic patient diagnoses. Used in early-stage cancer detection by combining CT images with clinical records. Support personalised treatment planning by analysing diverse data sources. 𝐋𝐚𝐫𝐠𝐞 𝐆𝐫𝐚𝐩𝐡 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐋𝐆𝐌𝐬) 𝐟𝐨𝐫 𝐁𝐢𝐨𝐦𝐞𝐝𝐢𝐜𝐚𝐥 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 Graph Neural Networks (GNNs) are used in brain network analysis, medical knowledge graphs, drug discovery, and protein interaction modeling. They improve disease prediction and biomarker identification by analysing gene-disease relationships. Applications like TxGNN and GraphGPSM support drug repurposing and protein structure prediction, aiding in new treatment discoveries. 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞𝐬 𝐚𝐧𝐝 𝐋𝐢𝐦𝐢𝐭𝐚𝐭𝐢𝐨𝐧𝐬 𝐂𝐨𝐦𝐩𝐮𝐭𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐝𝐞𝐦𝐚𝐧𝐝𝐬: Large models require significant processing power, limiting widespread adoption in clinical settings. 𝐈𝐧𝐭𝐞𝐫𝐩𝐫𝐞𝐭𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐢𝐬𝐬𝐮𝐞𝐬: AI models, especially black-box models, require better explainability to be trusted by healthcare professionals. 𝐃𝐚𝐭𝐚 𝐚𝐯𝐚𝐢𝐥𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐚𝐧𝐝 𝐛𝐢𝐚𝐬: AI models must be trained on diverse, high-quality datasets to ensure accurate and unbiased medical decision-making. 𝐄𝐭𝐡𝐢𝐜𝐚𝐥 𝐚𝐧𝐝 𝐫𝐞𝐠𝐮𝐥𝐚𝐭𝐨𝐫𝐲 𝐜𝐨𝐧𝐜𝐞𝐫𝐧𝐬: The integration of AI in healthcare demands strict validation, ethical guidelines, and regulatory compliance.

37

1 Comment