Posts in: medicine

Tuesday links, on science, medicine, technology and a bit of something extra

Why no scientist should hang their hat on a single pet theory, with real-world examples. The same problem haunts the world of biotech even as its denizens claim their superiority at drug development.

About a recent Nature Medicine article which found that LLMs were no better than Google at helping patients diagnose and manage their self-reported maladies. The reasons are those that I suggested two and a half years ago — ChatGPT can give you the correct answer from a properly structured clinical vignette, but the art and science of medicine are transferring the reality in front of you — the patient’s haphazard story, their hodge poge of medical records, the subtle physical exam findings — into something salient. Not saying AI won’t get there at some point, but it clearly still needs work.

Rao has collected 101 (!?) of his best Twitter threads and a few hundred single tweets into a book. A note on the title page says:

This book is LLM-friendly. Point your LLM to venkateshrao.com/twitter-book if you want it to explore it. A full interactive archive, explorable via an AI oracle, is under development.

Living up to his call to be (slightly) monstrous.

Yes, it is a person I hate making a good point, which is that the brutalist architecture of L’Enfant Plaza is out of place so close to the National Mall and should be kept where it belongs. I even prefer the proposed neoclassicist style to what Trump’s ego would want, which I imagine to be a Dubai [Note: And even Dubai would be better than what’s in the President’s id. ] on the Potomac.

Hacked to Pieces? The Effects of Ransomware Attacks on Hospitals and Patients

From American Economic Journal: Economic Policy about the effects of ransomware on patients: [Note: ↬Tyler Cowen ]

Ransomware attacks decrease hospital volume by 17–24 percent during the initial attack week, with recovery occurring within 3 weeks. Among patients already admitted to the hospital when a ransomware attack begins, in-hospital mortality increases by 34–38 percent.

The implication is that the computer systems being down has a huge detrimental effect on patient outcomes. What the abstract doesn’t get into — and I don’t have access to the full article — is how they calculated the in-hospital mortality among the already admitted patients. Many of them will have been discharged early or transferred to other hospitals if stable enough, decreasing the denominator and overestimating mortality. At least I hope that’s the case!

"In China, A.I. Is Finding Deadly Tumors That Doctors Might Miss"

So says this NYT headline (gift link). In reality, and in the article itself:

The tool might also be more useful for trainee doctors than for experienced specialists, said Dr. Diane Simeone, a pancreatic surgeon at the University of California San Diego. Some of the tumors that the tool caught in the Nature Medicine study should have been “super obvious” to well-trained radiologists even without A.I., she said.

But she acknowledged that it could be a valuable backstop for hospitals where specialists are in short supply.

This is based on the data So yes, A.I. is finding deadly tumors that an overworked and/or undertrained doctor might miss. Which is valuable, but a different message altogether from the one that the headline was trying to convey.

Separately, is “in China” becoming the new “in mice”? [Note: The link is to a PLOS One blog from 2021. The most recent post there as of the time of my writing this is a scathing and rather unfair review of the science of Pluribus. I refrained from adding it to my feed reader. ] What assumptions do writers have, and what emotions do they raise in readers, when they report about things happening “in China”? Was it the same with the Soviet Union? Whenever someone fans the flames of mimetic rivalry, I grab my wallet.

📺 The Pitt (2025)

First things first: The Pitt (2025) is miles better than two other era-defining medical shows, ER and House MD. The conceit — one hour per episode, one shift per season — makes for a more realistic pace. The case selection is good, if on the extreme end of any possible presentation. The medical staff personality types are spot on, [Note: They are all good, but I would like to highlight the charge nurse and the neuro-atypical first-year resident as commonly encountered phenotypes that TV shows never seem to get right. ] if not quite representative of the variety of English accents one would hear during rounds. And the battle between administrators and clinicians hit all the right notes, even if having the hospital’s Chief Medical Officer hover over ER staff at all hours of the day would be considered atypical for the role.

Kudos are also due for the use of prosthetics, sometimes quite grizzly, with an abundance of open wounds and mangled extremities. With so much exposed tissue I wondered why no one was wearing a mask during procedures even while, in a mid-season episode, admonishing an anti-mask patient about their beliefs. But that is, of course, another conceit, otherwise we would never be able to tell who was saying what. A more believable move was to have one of the medical students [Note: More kudos for making the two students smart, competent and lovable all at once. ] present for most of the cases, requiring everyone to explain what they were doing at an 8th grade level (our own 8th grader who was watching with us also appreciated this). Granted, the historical reminiscences and calling out different healthcare-related statistics were much less plausible: they reminded me of the most self-important parts of Studio 60 on the Sunset Strip that its then-arch rival 30 Rock so successfully parodied.

Admittedly, it is an unusual hospital. More than 20 ORs and so much house staff with only one attending physician during a day shift sounds… implausible. It does make for great tension-building, and it was no wonder that Noah Wiley’s character — spoiler alert — by the end of the season gets burnt out to a crisp. Another oddity is how competent and unflappable all of the staff were during — another spoiler — a major traumatic event that no one wanted to experience but everyone was prepared for. Color me skeptical that operations would have been that smooth.

Still. As fanciful as they were, ER and the less-remembered Chicago Hope were, to me at least and I suspect to many others of similar age, [Note: The less I say about House MD the better. ] a large part of the draw of medicine. It is good to know that there is a half-decent show out there that may keep the flame going.

A rare day-job update: we have not one but two papers out in the journal Nature Medicine this morning. The first is clinical and the other biomarker data from the same randomized trial, both open access. The last big paper was more than two years ago, and the post-publication feeling hasn’t changed.

Friday links, science and biotech edition, with extended commentary

The case for faster bench-to-bedside-and-back type of research, with which I agree. It is remarkable, however, how each generation interested in biomedical research reinvents the wheel without checking prior art. I would also argue strongly that the (correct) thesis of the essay is not a refutation of the biotech-as-casino hypothesis but rather its confirmation, unless you enlarge “biotech” to include academia and government research but then what are we even doing. Investors have no patience for nuance and view clinical trials as dichotomous regardless of how companies try to present them, and interpreting translational research results requires even more patience and tolerance of ambiguity.

Ginexi has been a program at the NIH for more than two decades, so caveat lector, but many POs are indeed mini-Moseses in their scientific domains. On one hand they perform important and valuable work, on the other the importance of a single human being to the careers of investigators young and old tend to favor those with soft skills of communication more than those of scientific and intellectual rigor. No judgements on my end because I genuinely can’t tell if the alternative would be any better.

Some genuinely good advice on how to write grants in a way to increase the odds of them being funded, with emphasis on accepting the reviewers’ comments and suggestions and approaching the grant resubmission as one would an offer to revise and resubmit a scientific manuscript, with much thanking and back-bending. Do keep that in mind when you read the next item.

This is true for most, as there are far too many academic right now for all of them to have soul in the game. However, academia continues to ask for more than it gives back out of too many people, while at the same time putting a negative selection pressure against people who are stubborn, single-minded and thus predisposed to a soul-in-the-game phenotype (see above). The only reason why the system survives at all is that the churn has been too low to fully reveal the tension, but it continues to creep towards the breaking point providing yet another case study of things that happen gradually and then suddenly.

Thursday links, for the academics

Wednesday links, one screw-up after another

Professional societies need to step up their online game, and so should we

“The internet is dying on the outside but growing on the inside”, wrote Yancey Strickler last month in a follow-up to his 2019 essay The Dark Forest Theory of the Internet. To avoid misunderstanding, malicious interpretation, competitive intelligence gathering and cancelation, conversations have been moving from the public-facing “social” “media” to gated, invitation-only services (e.g., your favorite Substack author’s members-only discussion forum) and private group chats (e.g., the Let’s Bomb Yemen Signal texts).

But some parts of this Cozy Web are growing faster than others, and as if often the case doctors and scientists are ruled by inertia. Both groups have the perfect setup, in the form of professional societies, to carve off some gated space in which to have potentially controversial discussions without providing fodder to “the enemy”. [Note: In these kinds of metaphors I always reach out to Venkatesh Rao’s The Internet of Beefs, which explains quite well why the public Internet has turned into a dark forest in the first place. ] And yet even the most developed online community program I know of — American Society of Clinical Oncology’s myConnection — is a stuffy, [Note: ASCO boasts as having more than 50,000 members. The two largest “communities” on MyConnection, “New Member” and “Women in Oncology”, have more than 9,000 members each yet the last post on one was 9 days ago (with zero replies) and 7 days ago (two replies). All of November, the more active WiO group had 9 posts with median 1 reply (range 0–20). ] formal messaging board that can barely be considered active. Most of ASCO’s online activity is still on X, where the official account has almost 150,000 followers and the hashtag for its annual meeting is heavily promoted. Other large hematology/oncology societies like ASH (hematology) and AACR (general cancer research) don’t even have that. Their “online community” is a member directory and heavy promotion of in-person conferences, which I can only assume are the true money-makers.

So I have to wonder, do they still deserve to call themselves “societies”? It is, after all, 2025 and much of life has moved online. By not providing an avenue for true internal discussion and instead promoting public debate, are they hurting their members’ cause more than helping? [Note: Yes, it was fun to post out in public when there was a slight chance that your favorite celebrity — or the POTUS — would retweet your post, but we have since learned that this is a liability more than a benefit and there are more high-follower accounts on X now that I would rather avoid. ] I have argued recently that scientists may want to button up their conversations if they are to keep or regain trust. Should these societies not be providing the means to do so, and not only once per year in a stuffy conference room? ASCO’s MyConection is on the right track, but much too formal. Yes, give people the opportunity to create subgroups and even more private chats as you do now. But if you think debating on X with millions of spectators is healthy, why not give all 50,000-plus members a chance to interact by default, and do so in a format that is not an early 2000s web forum?

Concluding the most recent article, Yancey Strickler provided a toolbox for people to create their own communities which he called the Dark Forest OS, of DFOS. While laudable, this effort is to put it bluntly too artsy fartsy for me. Strickler comes from the world of “creators” whose sensibilities are much different from those of doctors and scientists. But then science and medicine already have much of DFOS in place, from a members list to paying dues. The only thing we need now is for the said societies to build their walled gardens — with an app included! — which they would promote instead of X at the annual meetings and other conferences.

Where a SciMeDFOS would come useful is at smaller scale, for collaborative groups and maybe even large individual labs, where members are known but there are no dues, funds, or IT workers ready to build a custom Twitter clone. If I were to make one now I would probably use Hometown, which is a fork of Mastodon that enables local-only posting, though it being a single person’s passion project makes me a bit reluctant. But then what else do we have, Discord, WhatsApp and Signal? Whatever Dave Winer comes up with in collaboration with Wordpress? Maybe Squarspace could make creating private Twitter clones be as easy as creating websites? I will be on the lookout.

A Saturday NYT gift link splurge

Enjoy!