Healthcare AI Guy Weekly Newsletter | 9/5

Google expanding access to healthcare AI, ChatGPT accuracy at 72% in clinical case studies, Anthem expands AI-enabled plans, and more

Good morning everybody — hope you all had a great Labor Day weekend!

In this week’s edition, we cover the following developments:

  • Google expanding access to healthcare AI

  • ChatGPT’s accuracy at 72% in textbook-drawn clinical case studies

  • Anthem to launch AI-enabled virtual-first plans in more states

  • Link-worthy content, funding updates & new tools/partnerships

Our Picks

Highlights if you’ve only got 2 minutes…

1/

Google expanding access to healthcare AI

Google is expanding access to Med-PaLM 2, the company’s healthcare-specific LLM, to additional organizations to continue refining the product. Previously, access to the Med-PaLM 2 model was restricted to customers like HCA Healthcare, pharmaceutical giant Bayer, EHR company Meditech and digital health startups Infinitus Systems and Huma.

In early 2023, HCA launched a pilot program to test Google’s AI tech and now is considering further expansion to more hospitals later this year. The companies “will continue to improve the product, but physicians have already said they like the new tools.”

The real idea here is to keep experimenting with niche LLM models that are trained on narrower sets of data. Without tailoring models towards specific use cases, such as healthcare, you run the risk of just having a Swiss army knife, which is not the best knife, and it's also not the best screwdriver. And it's also not the best toothpick, etc. The goal is to run these models through rigorous quality evaluation and version control, which you would expect from any standard product. (link)

2/

ChatGPT’s accuracy in clinical decision-making at 72% in textbook-drawn case studies

It’s clear and with good reason that AI tech remains controversial in some medical communities. As we know, bad inputs = bad outputs, and clinical decisions require a lot of nuance, which many of these generalist models naturally can’t do well yet.

Physicians debate what is an acceptable success rate for AI-supported diagnosis and whether AI’s reliability under controlled research conditions will hold up in the real world. Mass General Brigham researchers tested ChatGPT’s performance on textbook-drawn case studies and found the AI bot achieved 72% accuracy in overall clinical decision-making, ranging from identifying possible diagnoses to making final care decisions.

What we need to find out now, is what the accuracy rate of the average physician is. However, not sure if I’m personally comfortable with 72% yet either way. That’s why I really think going after the administrative bloat with the efficiency AI brings is the first frontier. Clinical support still needs some work. (link)(tweet)

3/

Anthem to launch AI-enabled virtual-first plans in more states

Anthem Blue Cross and Blue Shield recently launched a new virtual-first plan that uses AI to streamline health services for members. Now, Anthem is expanding this plan to additional states, including Connecticut and Virginia, starting at the beginning of 2024. The plans first became available to large group self-insured employers in California, Missouri and New York as well as large group self-funded and fully insured employers in Georgia starting in July.

Members can access these plans through Sydney Health, a digital member engagement platform that includes access to benefits, tools, resources and provider care. Sydney Health also offers an AI-driven symptom checker that intuitively uses the information members provide to narrow down millions of medical data points and assess specific symptoms before seeing a doctor. (link)

Miscellaneous 🔍

News, podcasts, blogs, tweets, resources, etc…

  • Babylon Health, failed telehealth startup, sells assets to eMed (link)

  • Great Twitter/X thread on Babylon Health collapse (link)

  • Healthcare AI sees major investment from a16z & other VC firms (link w/out paywall)

  • You can now get a PhD at Harvard in AI in Medicine (link)

  • AI may influence whether you can get pain medication (link)

  • OpenAI on pace to generate more than $1B in revenue (link)

  • Doordash launches AI-powered voice ordering (link)

  • Tiny island nation Anguilla profiting 10% of GDP from .ai frenzy (link)

  • AI to represent 11% of healthcare budgets in 2024 (link)

  • OpenAI launched ChatGPT enterprise: more security, unlimited GPT-4 (link)

  • Behind the AI boom, an army of overseas workers in ‘digital sweatshops’ (link)

  • Could AI be spark that ends opioid epidemic? (link)

  • Students develop AI-powered therapist (link)

Deal Desk 💸 

Spotlight on latest capital raises, M&A and investments…

📈 EvoluntionaryScale: an AI biotech startup led by ex-meta researchers raised $40M in funding. Lux Capital was the sole investor. (link)

📈 Intuition Robotics: a provider of eldercare robots based in Israel raised $25M in funding. Woven Captial led the round and was joined by OurCrowd, Toyota Ventures, Western Technology Investment, and others. (link)

📈 QuantHealth: an AI-powered clinical trial startup based in Israel raised $15M in Series A Funding. Bertelsmann Investment and Pitango HealthTech co-led the Series A funding. (link)

Tool Box 🧰

Latest on business, consumer, and clinical healthcare AI tools and partnerships…

🔧 Google Cloud + Ginkgo Bioworks: Google Cloud partnered with Ginkgo Bioworks, an American biotech company, to build a generative AI platform for biology and biosecurity. (link)

🔧 MIT: MIT researchers have developed an AI technique called smoothing that simplifies contact-rich manipulation planning for robots, enabling them to manipulate objects with their whole bodies, potentially leading to the use of smaller, more mobile robots in factories and exploration missions. (link)

AI Images of the Week 📸

Funny memes and pics from around the web…

Checks out

US healthcare doing something right? Or just correlated to smoking?

See you next week 👋

That’s it for this week friends! Back to reading — I’ll see you next week.

Stay classy,

— Healthcare AI Guy (aka @HealthcareAIGuy)

PS. I write this newsletter for you. So if you have any suggestions or questions, feel free to reply to this email and let me know