Jan Beger’s Post

Success of Clinical Decision Support requires TRUST not only from Physician-CDS, but also from Patient-Physician. I am not reading much into the results given the artificial nature of the study, but it lays important foundations for the academic study of patient trust related to CDS. #CDS #futureofhealth #digitalhealth

The veterinary industry is behind when it comes to studying the effects of AI use by vets, on clients' perceptions. Luckily our industry isn't navigating these waters alone. Human healthcare is generating a lot of research whose conclusions can provisionally be applied across species boundaries. Here is one interesting paper on the topic of client/patient trust, thanks to Jan Beger . Worthwhile read!

AI in Medical Decision-Making: What Could Go Wrong? By Paul Pender, MD AI can suggest to physicians what clinical choices to make to achieve best outcomes, but such outcomes may not be realized, for several reasons.[1]  A Yale researcher reviewed the results from a study of blood clots in children, whereby one pediatric group had AI-generated recommendations for treatment while the other group did not.  There was no statistically significant difference between the outcomes of the two groups.  The study was conducted out of concern that children in the hospital risked blood clots from venous catheters and other sources.  Whether the risk of developing blood clots outweighed the risk of developing complications from anti-coagulation was a clinical decision to be made by the attending physicians. However, not all AI-directives for treatment were employed.  The author argues that even when faced with reliable data derived from many sources, doctors may not trust that data for making decisions. He refers to this phenomenon as the “Cassandra problem,” taking his cue from mythology.  Cassandra was given the gift of seeing the future, but no one would believe her pronouncements. “This is the gap between accurate prediction and the ability to change outcomes for patients. A prediction is useless if it is wrong, for sure. But it's also useless if you don't tell anyone about it. It's useless if you tell someone but they can't do anything about it. And it's useless if they could do something about it but choose not to.” As we learn more about the pros and cons of AI (including what are termed “hallucinations” derived from faulty assumptions) we need to continue to update our experience with AI as a tool for medical decision making.  Understanding what algorithms are at play may shed more light on why one clinical path is recommended while another is not.  Getting physicians to trust the output from questions directed to any AI-platform is a work in progress.  [1]https://lnkd.in/eq4vv48i

Why would hospitals not use life-saving medical research that’s already been discovered? Nishith Khandwala faced this shocking reality in 2020 when his father had a heart attack. Despite developing an AI algorithm at Stanford for detecting arterial blockages, he found it wasn't in use - not even at Stanford University itself. This raised a critical question: Why aren't groundbreaking medical algorithms like these in widespread use, even in their own institutions? Bunkerhill Health is Nishith's answer to this glaring gap in innovation and everyday medical practice. They operate on a two-part system: first, they create a consortium of top academic medical centers, fostering collaboration and validation of AI algorithms. Second, they handle the commercial aspects, managing FDA clearance and distributing these validated algorithms to hospitals. Key Highlights: - Bunkerhill Health ensures comprehensive algorithm improvement and validation by connecting AI researchers with a network of hospitals. - Post-FDA clearance, Bunkerhill Health commercializes these algorithms, sharing revenue with both the researchers and contributing hospitals. Their algorithms are already in use by top health systems including Cleveland Clinic and Mayo Clinic. - Bunkerhill Health has established a consortium with top research institutions like Stanford University, Johns Hopkins Hospital, and Harvard University, streamlining AI algorithm development and validation in healthcare. - Bunkerhill Health has raised $30.4M, with key investors including Sequoia Capital, Felicis, Y Combinator, Pacific 8 Ventures, and SciFi VC Follow Founders You Should Know to learn more about interesting companies we’re featuring. Apply for our upcoming showcase on Feb 7: foundersysk.com

"A growing number of AI tools are being used to predict everything from sepsis to strokes, with the hope of accelerating the delivery of life-saving care. But over time, new research suggests, these predictive models can become a victim of their own success — sending their performance into a nosedive and generating inaccurate, potentially harmful results... when patients start faring better, the models became far less accurate at predicting the likelihood of kidney failure and mortality. And retraining the models and other methods to stop the decay didn’t help. The latest research findings serve as a cautionary note at a time when few health systems are tracking the performance of AI models over time, and raise questions about what potential performance degradation means for patient outcomes, especially in settings that have deployed multiple AI systems that could be affecting each others’ performance over time." "the longer health systems use predictive models without accounting for this feedback loop of degraded performance, the less reliable they’ll become. “It’s like a ticking time bomb.” “We’ve long recognized that successful implementations affecting patient outcomes and downstream feedback within EHR data will require new approaches to model updating,” Sharon Davis, a professor of biomedical informatics at Vanderbilt University Medical Center, wrote in an email to STAT. “The interactive effects of the sequential and simultaneous release of AI-based tools are another layer of complexity for model managers that will need innovative solutions.” #bioethics #healthcare #hospitals #machinelearning #medicine #clinicalcare https://lnkd.in/ePycjWxB

a critical question is to have the application explain why it is making this recommendation and how it arrived at this decision

The proliferation of investor interest in AI and LLM is spawning marketing noise that clinical end-users must sieve through efficiently. Erkin Otles provides a template for doing this below.

Where does AI fit in medicine? A fantastic, thought provoking episode from the esteemed Eric Topol, MD and Adam Rodman Exceptional content, certainly worth the listen. Some key takeaways: 1. Like AI today, the stethoscope faced resistance before becoming essential to clinical practice. 2. LLMs could have a key role in bedside medicine helping to condense and understand large volumes of information. 3. Some evidence suggests that general-purpose LLMs can perform clinical reasoning and diagnoses on par with or better than humans in certain experimental scenarios. 4. LLMs could empower patients by helping them understand their medical records and participate in their care more actively. 5. Medical education must adapt to incorporate AI and LLMs into the curriculum, with particular focus on working effectively with AI rather than mastering its inner workings. 6. AI and LLMs have the potential to reduce burnout among healthcare professionals by automating tasks like documentation and allowing more time for patient care. https://lnkd.in/eqTx5VTn

🚨 new blog post on the 3 questions every doctor should feel comfortable when confronted with a new AI system! 🩺💻 tl;dr, you should ask: * how was it made? * is it any good? * how is it being used? # why do docs have to ask these questions? 👩🏾⚕️🤦🏾♀️ there's been crazy growth in the number of medical AI systems. many of them have been validated, but its unlikely that they've all been evaluated for all the settings they are being used in. thus, the onus of policing these systems comes down to clinical users. this isn't ideal, but it is the hand we've been dealt # how was it made? 🛠️ creation of AI models is called development. our goal is to create a model that can estimate unknown outcomes using known information. this process is super dependent on data. thus, a clinical perspective is critical to make sure the right data are included (and excluded) # is it any good? 📊 validation is the technical term for assessing the performance of a model. there's a lot of overlap between AI validation, evidence based medicine, and biostatistics. validation population and results should be easily intelligible to docs. additionally, there should be some form of continuous assessment # how is it being used? 👨🏾💻 in my opinion, the use of AI models in clinical practice (aka implementation) is the trickiest part. how do you integrate model into clinical workflows in a beneficial way? i'd argue that docs are the experts in the workflows and should feel comfortable asking for changes to summarize when confronted with a medical or healthcare AI/ML system ask: * how was it made? * is it any good? * how is it being used? everyone from medical students to attending physicians should feel comfortable asking these three questions. and our engineering friends should be able to answer them easily read the full blog post here: https://lnkd.in/g-x7M4BY #artificialIntelligence #machineLearning #healthcare #medicine #hit

A study published in The Lancet Digital Health evaluates the effectiveness of Foresight, a generative transformer in temporal modelling of patient data, integrating both free text and structured formats, to predict a diverse array of future medical outcomes, such as disorders, substances, procedures, and findings. One of the study's contributing authors is Anthony Shek, PhD, who participated in a panel on artificial intelligence (#AI) at the #SNOMED CT 2023 Expo and who is giving the opening plenary at our April 2024 Business Meetings. The study found that Foresight is a general-purpose model for biomedical concept modelling that can be used for real-world risk forecasting, virtual trials, and clinical research to study the progression of disorders, to simulate interventions and counterfactuals, and for educational purposes. Learn more: https://lnkd.in/gueG8-G9