Vahid Kowsari

Healthcare tends to lag behind in tech adoption, but that might be its biggest strength as AI transforms our industry. Julie Yoo, General Partner at a16z, recently made a compelling point: Unlike other industries, healthcare isn’t weighed down by billions of dollars of investment in legacy software systems. This lack of "sunk-cost bias" means we have a unique opportunity to leapfrog directly into the benefits of AI without the burden of ripping out decade-old tools. Our industry is slow to adapt not because of a lack of awareness, but because the mindset is deeply rooted in established ways of working. The key to advancement lies in shifting this mindset. Luckily, this opportunity couldn’t come at a better time. Healthcare is facing a critical workforce shortage, with hundreds of thousands of doctors and nurses short of what’s needed to meet growing demand. AI can help fill this gap by scaling clinical expertise, making certain that the best practices are shared widely and that every patient receives the same high standard of care, regardless of location. However, none of this matters if we don’t build trust in the AI tools being developed. Healthcare already has strong regulatory frameworks, with the FDA approving hundreds of AI-driven clinical products. But trust is built slowly, and the human element must remain central to any transformation. At PROXDIS, we’re focused on how AI can drive positive change, without sacrificing the relationships and trust that are the core of patient care. Technology has incredible potential for improving care, but it can never replace the relationships and trust that are at the heart of patient care. As Julie pointed out, healthcare can use this opportunity to jump ahead—but only if we guarantee that this innovation is always centered around the real needs of providers and patients alike. Are we ready to adapt to AI at a pace that allows for meaningful change while keeping the human element front and center? Let me know your thoughts.

14

Let's talk AI with payers and providers—some interesting numbers to consider from the latest research (Sept 2024) (KLAS Research & Bain) Here are four eye-opening takeaways: 1. AI Strategy Surge: • 15% of providers now have an AI strategy, tripling from 5% last year • 25% of payers are leading the pack with established AI strategies • Both sides are bullish on generative AI's potential 2. Tech Spending Spree: • 75% of surveyed orgs boosted IT investments • Top priorities: cybersecurity, clinical workflow optimization, and data platforms • 70% of orgs doubling down on security 3. Payer Priorities: • Focus on care coordination, utilization management, and claims processing • 65% grappling with legacy tech challenges • Trend towards streamlining tech stacks and favoring existing vendors 4. AI Adoption Hurdles: • Regulatory and legal concerns top the list • Cost and accuracy issues (like AI hallucinations) are giving pause • Growing need for robust governance and ethical frameworks The report here: https://lnkd.in/gxjKPUaW Is your organization leading the charge or playing catch-up? #HealthcareAI #TechInnovation #FutureOfHealthcare

66

1 Comment

In our last blog, we explored the potential impact of a Trump 2.0 administration on healthcare and its implications for HealthScoreAI™. Today, we expand on that analysis, dividing this discussion into three parts: (1) recent developments in the Trump 2.0 administration, (2) the current state of healthcare, and (3) the role of AI in this evolving landscape.

1

As the discourse shifts from models to compound AI systems / agents, we need better AI benchmarks to evaluate multi-modal and multi-step task performance, especially in healthcare and life sciences. When we wrote the first paper demonstrating ChatGPT's performance on the USMLE, we chose the US Medical Licensing Exam as a benchmark for accessibility, speed, and ease. This benchmark was never intended to represent AI model performance on real-world clinical tasks. Today, I still see so many research teams and startups using benchmarks (like the USMLE) that are ill-fitted for assessing the true clinical or scientific performance and utility of the models they are developing for real-world contexts. Benchmark development may be seen as a "less sexy" area of research, but it is of paramount importance. Years after the rise of the transformer, we still lack adequate benchmarks for so many single-step tasks in biomedicine. With compound AI systems (i.e., architectures that integrate multiple AI models to perform complex tasks) emerging, we need new benchmarks for agentic behaviors. I'd argue that developing an agent with novel capabilities without at least proposing a companion benchmark (if an industry standard does not yet exist) may hinder the adoption of said agent, especially for high-stakes workflows. Designing more benchmarks that capture/simulate real-world clinical and scientific workflows will help us mitigate the major discrepancies observed between in silico and in vivo performance and better support safe + effective deployment of AI in biomedicine. There are already brilliant people focused here, and we need more. DMs are open if you're researching or working in this area of multi-step/multi-modal benchmarking in healthcare and life sciences! #healthcare #ai #artificialintelligence #generativeai

100

18 Comments

Here’s your recap of last week’s health IT news 🗞️ 👇 🤖 AI Clinician Assistant • Suki is partnering with Zoom to generate clinical notes as well as other integrated AI features.    • Net Health announced it would integrate AI ambient scribe Tali AI. • Aidoc is partnering with NVIDIA to create a guideline to accelerate AI adoption, the Blueprint for Resilient Integration and Deployment of Guided Excellence (BRIDGE). 📞 AI Contact Center • Authenticx released a new AI assistant that helps users uncover business insights from their data. • Infinitus Systems, Inc. closed a $51.5M Series C with participation from Memorial Hermann Health System and others.   🏥 AI Facility Management • Cognosos, Inc. acquired Cox Prosight. 🌡️ AI Remote Patient Monitoring and Triage • Biofourmis has merged with CopilotIQ. 📈 Data & Analytics • Avandra Imaging launched out of stealth; it plans to sell to biopharma, medical research, and AI companies. • Gradient Health acquired DataAppraisal. • HealthEx launched and announced $14M in Seed and Series A funding. 📋 Patient Administration • Notable released Flow Builder, which allows users to develop their own AI agents for operational workflows within EHRs and other systems. • MIla Health Inc released its flagship AI solution of the same name, and announced implementations at provider organizations including Aayu Clinic, MIMIT Health, and Best Practices Inpatient Care. 🔁 RCM • Oracle’s patient accounting solution has been integrated with FinThrive, which enables improved functionality, performance, and a unified login experience. • Blue Shield of California and Salesforce are partnering on a solution that promises to give patients prior auth responses within the course of their appointment. ⚙️ Clinical Operations • Dyania Health closed a $10M Series A, including participation from Cleveland Clinic Ventures. 📱 Patient Support & Communications • Psych Hub released a new solution designed to support patients seeking mental health support, directing them to both non-clinical and clinical resources, thereby assisting behavioral health providers to reach new patients. • Mayo Clinic has partnered with Lin Health to support patients after treatment for fibromyalgia.   🍎 VBC • WellSky acquired Bonafide Health, allowing it to provide a more integrated platform for providers to run their businesses. 🧠 Food For Thought • 𝗦𝘁𝗿𝘂𝗴𝗴𝗹𝗶𝗻𝗴 𝘁𝗼 𝗸𝗲𝗲𝗽 𝘂𝗽 𝘄𝗶𝘁𝗵 𝗮𝗹𝗹 𝘁𝗵𝗲 𝗔𝗜 𝗴𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸𝘀? Andrew Hines, CTO at Canvas Medical, shared a breakdown (plus some hot takes) on all of the many entities currently vying for market share in the AI governance space https://lnkd.in/eMNpXw4t --- Want the latest healthcare tech news delivered to your inbox each week? Sign up here 📬 👉 https://lnkd.in/e2UJ3ckP

65

4 Comments

At Iodine Software we have been harnessing the power of AI for a long time to reduce the administrative toil and waste in our healthcare system. Take a few minutes to tune in and listen to this podcast from William Chan, our Founder and CEO to learn how AI can be leveraged responsibly to address key pain points in the mid revenue cycle #HealthcareAI #healthcareinnovation

26

3 Comments

I’ve spoken with 50+ CMIOs / CIOs over the past year on how they’ve piloted new Health Tech for clinicians (e.g. AI scribes, generative AI drafts for patient portal messages, etc.). There’s no one right answer, but there are 5 main approaches I’ve heard for choosing clinicians to pilot with: 𝟭. 𝗠𝗼𝘀𝘁 𝘁𝗲𝗰𝗵 𝘀𝗮𝘃𝘃𝘆 𝗰𝗹𝗶𝗻𝗶𝗰𝗶𝗮𝗻𝘀 You want to move fast so you look for clinicians who raise their hand and are eager to try the solution. The risk? Tech savvy clinicians may be biased in favour of new tech. So they may not be highly representative of the population. 𝟮. 𝗟𝗲𝗮𝘀𝘁 𝘁𝗲𝗰𝗵 𝘀𝗮𝘃𝘃𝘆 𝗰𝗹𝗶𝗻𝗶𝗰𝗶𝗮𝗻𝘀 Or you do the opposite. If the tech is so intuitive, so easy to use that even the least tech savvy folks can and will use it… then you can be confident it can scale enterprise-wide. The risk? Maybe you don’t actually care if the least tech savvy folks can use it, because even 50% adoption is good enough to have a meaningful ROI. E.g. data I’ve seen shows clinician adoption of AI scribes can range 40% to 80% at health systems. But if you piloted with the least likely users (who don’t end up using it)… you would call the pilot a failure and miss out on the up to 80% who would’ve used it. 𝟯. 𝗖𝗹𝗶𝗻𝗶𝗰𝗶𝗮𝗻𝘀 𝘄𝗶𝘁𝗵 𝘁𝗵𝗲 𝗯𝗶𝗴𝗴𝗲𝘀𝘁 𝗽𝗮𝗶𝗻 𝗽𝗼𝗶𝗻𝘁𝘀 If you’re looking to demonstrate the biggest ROI, you could pilot with the clinicians who would benefit the most. For Generative AI drafts for patient portal messages, this is often primary care physicians as they spend more time in their inbox than other specialties. For AI Scribes you could check EHR data to see who is spending the most time on clinical documentation, who is doing the most pajama time in the EHR, etc. The risk? You might end up investing in a solution that works for one clinical group but not for others. 𝟰. 𝗟𝗮𝗿𝗴𝗲𝘀𝘁 𝗴𝗿𝗼𝘂𝗽 𝗼𝗳 𝗰𝗹𝗶𝗻𝗶𝗰𝗶𝗮𝗻𝘀 More data = shorter time to determining value. So it would make sense to pilot with largest specialty with the most clinicians. The risk? Again you would be assuming a solution or approach that works in one specialty is scalable to all… it may not be. 𝟱. 𝗦𝗼𝗺𝗲 𝘂𝗻𝗶𝗾𝘂𝗲 𝗶𝗻𝘀𝗶𝗴𝗵𝘁 𝗮𝗯𝗼𝘂𝘁 𝗽𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹 𝘂𝘀𝗲𝗿𝘀 For example some health systems have chosen AI Scribe pilot groups by asking medical office assistants to identify physicians who care the least about how a note is written and thus are more willing to delegate to an AI scribe. The risk? It’s super custom to the situation and it may be hard to replicate this approach with various new tech. 𝙒𝙝𝙖𝙩 𝙖𝙥𝙥𝙧𝙤𝙖𝙘𝙝 𝙬𝙤𝙪𝙡𝙙 𝙮𝙤𝙪𝙧 𝙝𝙚𝙖𝙡𝙩𝙝 𝙨𝙮𝙨𝙩𝙚𝙢 𝙩𝙖𝙠𝙚? #digitalhealth #healthcareinnovation

53

21 Comments

"AI has got to go!" "Trust Nurses Not AI" from Nurses protest AI at Kaiser Permanente in Becker's Healthcare shows how much we have lost the plot. And its *all* the fault of the vendors in this Industry. IBM started with the mythical presentation of Watson for medicine. Then you have vendor after vendor making claims that studies do not show; videos of capabilities their products do not have; and muddling the water of AI is and can do so much that hospital technologists will say "thats not real AI" b/c your reality does not match their fantasy. We desperately need a Peterson Center on Healthcare for AI, but it would have to be 1000X the size to deal with the amount of BS in the market. In the mean time, what can help. 1. To Hospitals/Health Systems - Force vendors into trials and real world evaluation. Allow them to shape the environment to enable speed for your benefit and theirs - but make them show it work with real patients. No more 137 point RFP evaluations before you evaluate one solution. This is best practice b/c each AI requirement is worth 100 other ones (if it works) and you need to cut beyond the marketing. 2. To the Industry/Influencers - We need a SOC2 / HITRUST for AI. It does not have to validate models work, just that its real data, real training, force transparency, etc. This will not solve the problem but will eliminiate a significant portion. This is far better than government stepping in. 3. To Vendors - every video should be of your real product working (for computer vision AI), every demo should include an anonymized real data set (for predictive models) and every generative AI company should have a transparent set of test cases that can show the capabilities and boundaries of the tech. Also, would love to learn of educators who are helping teach healthcare professionals about AI evaluation in practical ways. As to the topic, our Director of Nursing at LookDeep Health (Tiffany Wyatt, RN BSN ) and many others have made the point that done well, AI can enable vs abstract the relationship between patient and nurse. One of the most valuable and scarcest resource in the hospital is clinical attention of nurses and doctor - AI can be the ultimate sorting hat that helps that expertise always be in the right place. Patients, nurses and even the financial drivers of the hospital can be aligned because of AI - but we need to show nursing and providers more generally thats what we are actually aiming for. #healthcare #AI #hype https://lnkd.in/ge8mW6WT

26

11 Comments

Lots of dollars flowing in to these sorts of solutions. As someone who managed a large call center for a primary care group, the success of these technological solutions will come down to two things: A) Can these solutions clearly reduce labor costs across a large physician group by streamlining low-risk patient interactions? B) Can it deliver a 'delightful' experience for patients? The physicians I've worked with have hated two things - the feeling that they're 'wasting money' on labor costs, and patient complaints about bad experiences on the phone.

8

🌟 A Better Future: Greater Trust of Generative AI in Healthcare 🌟 Just released: My latest Forbes blog, where I share thoughts on some surprising statistics that were revealed in a recent survey from Deloitte. For all the upside that generative AI will offer to patients, consumers, and organizational stakeholders, we must respect the crucial cornerstone of trust. As AI and generative AI are pouring into healthcare biggest challenges and needs, how it lands matters. We need to respect that patients are at the center of care and related transactions. Setting the right foundation is extremely important and wholly necessary. And in my latest blog, I share 6 strategies to help healthcare organizations who will be utilizing generative AI. #Trust #GenerativeAI #ForbesBlog #FutureOfHealthcare

35

2 Comments

What are important benchmarks you use with LLMs with #AIinhealthcare? Can #LLM excel in medical encoding? Yes! But what is your measurement of success? That you found a billable code or is it the RIGHT billable code? Does it help you ALSO find the RIGHT non-clinical code because of Regulatory pressures like #USCORE and #USCDI? Or for Research, Adverse Events, Observational outcomes, etc...? Questions you should be asking yourself in measurements of success: - Did you benchmark manual review vs LLM review? Was the encoded encounter the same code that left the organization on the bill? Was there a human in the loop to confirm its accuracy? - Was the encoded encounter changed from point of care to #CDI? Did CDI recommend a change after LLM? - Is the LLM recommending other code sets for the encounter because encoding is only a portion of the clinician's need? Are they the right codes or purely a crosswalk? - Did the LLM provide you an international version of your encoding requirement or did they give you a US version (AKA training Hallucination)? Here at IMO Health, we leverage terminology services as an expert system to couple IMO Clinical AI to train LLMs. This greatly improves the accuracy and support of medical encoding and supporting the overall F-Score of what truly occurred during the encounter. If you’re curious about how technology can positively impact medical coding accuracy, this read is for you. https://bit.ly/3VBZFqn  #DataQuality #NLP #ClinicalAI #ClinicalTerminology

10

2 Comments

Watch this.

29

3 Comments

I'm seeing a lot of "hot takes" about the The New York Times article* discussing autonomous AI being better than hybrid (human+AI) in clinical settings, where the latter has been dogma for a long time (based on a neat paper published in JAMA**. There have been a few other similar studies published recently. The Google AMIE*** paper is my personal favourite!). *https://lnkd.in/e5pGzfyY **https://lnkd.in/ed5Je3p5 ***https://lnkd.in/eVrs3p7G All I'm going to say is... experimental paradigms are important. And if that's too subtle, then please refer to the image below (which I first saw in a phenomenal MS/Neurodegen lab I worked in during medical school... if you ever want to chat oxidative stress and eIF2α, shoot me a message). Okay, back to work. P.S. I'm not saying autonomous isn't > hybrid, but rather the studies to date don't *prove* anything. No need for the existential panic, just yet. We've still got the hard part left, i.e. actually generating some real world evidence based on an outcome that human beings care about...

95

12 Comments

🔄 The Case for an Enterprise GenAI Platform: Promoting Reusability to Drive Innovation 🚀   Building on my recent post about the necessity of an Enterprise GenAI Platform (https://lnkd.in/eeQBtV9p), let's delve deeper into its key benefits, starting with reusability. In my view, reusability is a crucial factor that can dramatically accelerate innovation within your organization. Why? Just consider the resource and budget fragmentation occurring in your company as each LoB (line of business) attempts to create their own GenAI POC. The majority of these use cases employ a RAG (retrieval augmented generation) design pattern, essentially applying NLP (natural language processing) to their corpus of unstructured data. Although we’re using the same fundamental architecture, there are numerous distinct codebases now being developed & maintained by different teams, leading to significant redundancy in efforts in large enterprises!   Reusability isn't just a technology component—it's a catalyst for efficiency in the GenAI ecosystem and a game-changer for innovation. By harnessing pre-built design patterns and common components, organizations can: 1/ Accelerate Development: Drastically reduce time-to-market for new GenAI solutions, 2/ Ensure Consistency: Maintain uniform quality and standards across projects, and 3/ Optimize Resources & budget Dilution: Focus on novel challenges instead of reinventing the wheel.   A well-architected Enterprise GenAI Platform should encompass these pivotal features (though this isn't an exhaustive list) to amplify reusability: 1/ Modular Architecture: Enables easy integration and reuse of components across various applications. 2/ Component Registry: A centralized repository for discovering and utilizing reusable assets like pre-trained models, APIs, data pipelines, and even GenAI applications. 3/ Standardized APIs: Facilitates seamless connections between GenAI models and internal systems. 4/ Version Control: Ensures consistency and allows teams to track changes in reusable assets. 5/ Pre-built Templates: Accelerates development by providing templates and best practices for common use cases (think document summarization, formatting documents, summarizing transcripts to generate minutes of the meeting, etc.) By focusing on reusability, organizations can significantly accelerate the development of GenAI use cases. While the exact impact would vary by org size and complexity, reusability through an Enterprise GenAI Platform should lead to substantial efficiency gains.   💬 How are you currently managing resource allocation & fragmentation across different lines of business when developing GenAI solutions?   📢 Subscribe to my newsletter for more insights: https://lnkd.in/g3bdneR7   #aws #ai #EnterpriseAI #GenAI #Innovation #AIReusability #MLOps #AIPlatform #DigitalTransformation #AIStrategy #TechLeadership #FutureOfAI #GenerativeAI #cio #ciso #cto #cdo

11

To follow up on Dede Ainbinder recent post: If we use the current EHR data to build a AI clinical decision support tool it is unlikely to capture our clinical decision making process. EHR's are inherently designed to maximize billing and minimize risk of litigation. A clinical decision support tool based on this data will likely do the same. There is a large sunken cost fallacy here where we continue to hang on to garbage data. Just because the data gathering and storage process was capital and time intensive doesnt mean the data is valuable. To truly capture the clinical decision process we need to capture data which sits outside billing and litigation. Its hard to imagine a future where this data isnt used for billing and cannot be called upon during litigation. #EHR #healthdata #clinicaldecision

17

8 Comments

Gen AI is transforming healthcare for the better. Honored to be featured in Forbes by Lisa Chau, showcasing how Abstractive Health is making advancements in AI automated clinical summaries. “The goal of [GenAI] is to remove the administrative burdens currently placed on doctors so they can focus again on what brought them to medicine in the first place: caring for patients.” #genAI #healthtech #clinicalsummary

110

11 Comments

👉🏼 ChatGPT and retinal disease: a cross-sectional study on AI comprehension of clinical guidelines 🤓 Michael Balas 👇🏻 https://lnkd.in/exN8AyjC 🔍 Focus on data insights: - 📊 ChatGPT achieved an impressive average score of 4.9/5.0 in alignment with AAO PPP guidelines, indicating strong performance in understanding clinical protocols. - 🏥 The evaluation covered 130 questions across 12 domains, showcasing the breadth of topics addressed in retinal disease management. - 📉 Variability in scores was noted, particularly lower performance in surgical management, suggesting areas for improvement in AI training. 💡 Main outcomes and implications: - 🤖 The study highlights the potential of AI, specifically ChatGPT, to assist in generating guideline-concordant responses for common medical retinal diseases. - 🔍 Identified errors in diagnostic criteria and treatment options emphasize the need for continuous refinement of AI models to enhance accuracy. - 📚 The findings suggest that while AI can support clinical decision-making, clinician oversight remains essential, especially in complex surgical scenarios. 📚 Field significance: - 🌐 This research contributes to the growing body of evidence supporting the integration of AI in healthcare, particularly in ophthalmology. - 🧠 It underscores the importance of interdisciplinary collaboration between AI developers and medical professionals to ensure effective application in clinical settings. - 📈 The results may influence future AI training methodologies, focusing on improving comprehensibility and accuracy in specialized medical fields. 🗄️: [#AI #retinaldisease #clinicalguidelines #ChatGPT #ophthalmology #medicalAI #surgicalmanagement #healthcareinnovation]

2