I’m starting a (nerdy) blog!
About medicine, AI, and the intersection
I’m starting a Substack! (which is a fancy way to say blog/mailing list, similar to saying Kleenex vs. tissue paper or Google vs. search). After publicly critiquing an AI in medicine paper on Twitter, I realized I wanted a more long form format to more thoroughly discuss opinions, learnings, challenges, as well as highlight advances in the field of artificial intelligence and medicine. I have strong opinions (and hopefully some insights) after both developing and deploying AI models in cardiology, and I wanted to have a space to share those opinions as well as discuss recent events. There’s undoubtably a lot of opportunity and potential in AI+medicine, but there’s also a lot of hype in the space, and I hope to cut through unrealistic expectations that fuel the Gardner curve.
This Substack will focus on artificial intelligence and medicine - topics dear to my heart (pun intended) and the intersection of the two (like the business, regulation, and advancement of AI+medicine). As a cardiologist with a focus on cardiac imaging and AI researcher/statistician, I think there’s plenty for clinicians to learn from computer scientists as well as engineers to learn from physicians. I have some practical experience which requires foundational medical knowledge as well as understanding of the math and statistics of machine learning (you can never know too much math and statistics in this space). Questions I’m hoping to answer include (“What does an AUC of 0.93 really mean? Like beyond the rough impression that it’s good, is that clinically actionable or reasonable?” as well as “What were the actual pain points of doing an AI RCT?”)
Working at this intersection, I’ve titled this series as “Intentional Intelligence”, because it’s not about artificial intelligence or human intelligence, but the intentional union of the two that will ultimately move the needle in AI in medicine. Posting every Wednesday, I hope to be intentional about providing my opinions on recent developments in AI and medicine. (As well as sharing lots lessons on things not to learn! - lots of bad habits in both disciplines and friction points in getting to deployment.)
Motivation
There’s a few motivations to starting this blog, and I hope to be transparent about this. First, one of the superpowers I’m most impressed by is the ability to write well. I already do a decent amount of writing in my professional life, but I’m always looking for ways to practice and improve. I’m hoping to use this platform for intentional practice as well as crystalize ideas about topics I’m passionate about. Making this series once a week, I hope a recurring deadline will motivate me to finally put pen to paper about many of the ideas that come up at various phases of research projects and clinical situations.
Second, I’m a big fan of other industry specific mailing lists and podcasts. Whether about tech (with Stratechery and All-In Podcast), finance (with Money Stuff or Bits about Money), medicine (with This Week in Cardiology), or the creator economy (with Colin and Samir), there’s an incredible amount of content on the internet sharing technical knowledge and I love to read how professionals think about their field. I look to these series as ways to think about adjacent fields I’m interested in, and I’m hoping to provide that perspective to academic medicine and AI in healthcare. There’s lots of “quirks and features” in medical AI as well as being in academic medicine that I think is generally interesting and lends color to the field. I’m aiming to have a similarly general audience, relatively jargon-free approach and offer my 2 cents about both achievements as well as shenanigans in the space.
Finally, I’ve been invited to write a few editorials about AI+medicine across a few medical journals, but the more editorials I write, the more I’m convinced it’s a thankless task. There’s relatively limited reach, a misalignment of incentives to truly highlight significant flaws in a manuscript, and often editorials are written primarily by the most junior member of the byline. I’m not senior enough in my career to have an open invitation to write an editorial, and the process of soliciting an editorial idea is often long, subjective, and unsatisfying. I’m convinced that a more casual medium without a paywall (few people actually personally subscribe to NEJM, JAMA, and etc and rather rely on their workplace to purchase) can capture a broader audience. The internet, by nature of reducing the cost of distribution, is a better medium to have farther to discuss interdisciplinary work, controversial ideas, and react to current events. The idea of building my own audience (even if small) that is more focused and relevant seems much more enticing and interesting to me.
What I’m gonna talk about
In no particular order, I’m hoping to focus this Substack on:
Challenges and successes in implementing AI in healthcare. In running our RCT, we encountered a few friction points as well as champions that really improved our experience, and what that means for AI deployment.
Important concepts in AI that are missed or under-appreciated in medicine. By now, everyone’s heard of an AUC metric, but what does that mean for clinicians?
Really great and really bad papers of AI in medicine. In our lab meeting, we frequently highlight behind closed doors practices we really want to emulate as well as behaviors we really don’t want to copy in the most recent manuscripts. I’ll probably post a few curated opinions of these highlights and lowlights.
The career trajectory and training pathway to be at the intersection of AI+medicine. I’ll briefly share the skills/classes/trainings I’ve found most helpful in getting me to my current role.
The business and regulatory landscape of AI in healthcare. Why are there so few AI RCTs? How come there are so many AI products out there being sold by salesmen without strong clinical evidence?
I’ll probably end up writing some posts that might be a little too technical (probably both erring on either too medical or too computer science-y at different times), but ultimately hoping to make the field of AI+medicine approachable.
Tldr
AI+medicine is a fun and expanding field, there’s lots of quirks and features that I hope to highlight, and if interested, I’ll post every Wednesday.