Week of 12-15-25
In the office last week, an older female patient held up her right hand, extended the index finger, and showed me a deformed third knuckle, saying, "I think I have Peyronie's disease (PD) of the finger." Her husband snorted, and I laughed. Direct-to-consumer (DTC) pharmaceutical advertising is effective when I have female octogenarians diagnosing themselves with relatively obscure conditions. PD affects neither fingers nor women, and yet my patient knows about it thanks to commercials and the website bentcarrot.com (yes, bentcarrot.com). This is not the first time DTC ads have invaded my patient appointments, and I am amused and bothered by their power to shape patients' identities. The ads don't just inform patients about illness and treatment options - they recruit armies around branded conditions and encourage an identity defined by illness.
Illness identity - the degree to which a chronic condition shapes someone's sense of self - is itself a focus of research. There is a robust body of literature discussing how DTC creates patient identities, complete with communities, advocacy groups, and social movements around branded conditions.
See the PD commercial that informed my patient:
Read about Illness Identity:
One note: Many of my friends and readers are aware that our younger son attends Brown University. Thank you for the outpouring of concern and caring comments after the weekend events. My son, his friends, and the other students we know are safe, albeit (understandably) shaken. He arrived home Sunday night.
Listen to a Google Notebook LM-generated podcast of the newsletter, featuring two virtual hosts.
Science and Technology Trends
Two good X threads about GLP1s (like semaglutide and tirzepatide) caught my attention this week.
First, a fascinating comparison of GLP-1 prescribing rates with other pharmaceuticals. Tirzepatide is the fastest-growing drug in history (as measured by # of prescriptions written over time). The thread notes that these data would be even more impressive if they included compounded versions. Most of these prescriptions likely target weight management, with diabetes being the secondary driver. It is fortunate that the most effective weight-loss drugs in history (so far) confer a range of additional medical benefits.
These data come from a NEJM article arguing that Liraglutide (a first-generation, lower-cost GLP-1 that is less effective at weight loss) should be made available to a wider population.
The second X thread is by a French PhD chemist, Simon Maechlin (check out this science/data-focused substack: https://substack.com/@simonmaechling), which covers a topic on one of those "additional GLP-1 benefits" - dementia. He discusses two large trials demonstrating GLP-1s FAILING to slow the progression of early Alzheimer's disease. Maechlin offers a good perspective on how these data are a great example of thoughtful, well-designed studies that advance our understanding through rigor and trial design (or, as he puts it, hope≠evidence).
More of these GLP-1s are coming. Meet retatrutide:
I found a striking article based on zoo data: "A large comparative biology meta-analysis (71 published studies looking at 117 zoo-housed mammals across 22 vertebrate species) demonstrated that permanent surgical sterilization and ongoing hormonal contraception increase life expectancy by 10-20% across all the species studied. These data are the first comprehensive evidence that reproductive processes (i.e., sex hormones) fundamentally constrain adult survival independent of environment." Who is going to tell my wealthy, longevity-focused patients that hormone blockers and surgical sterilization (i.e., spaying and neutering can be for humans too) should be in their 'tool box' of treatments? Sometimes, good data is not the data we want.
AI-supported analysis: https://claude.ai/public/artifacts/c8e8b240-c6a1-4b29-87dc-d12b175f93a4
Related: Here is a 2024 example of a challenging but fascinating dataset—a large, well-designed, long-term Danish survey that shows that the closer wealthy people live to poorer people, the less supportive they become of wealth redistribution policies (i.e., welfare or other income support). These data contradict other observational studies. Perhaps this highlights abstract altruism - generosity toward the poor, but only from a distance. While the research has some methodological problems, it is a novel and nuanced finding in a large data set.
AI-Supported analysis: https://claude.ai/public/artifacts/d359b79c-16ba-4d36-89d5-7e4851651bee
Anti-Science Articles of Note
A tale of contrasting stories this week:
JAMA published a French cohort study (28.7 million individuals) that found no long-term safety signals with mRNA coronavirus vaccines. Vaccinated individuals actually showed 25% lower all-cause mortality over 45 months. Meanwhile, the FDA has expanded its COVID vaccine inquiry to adult deaths after claiming 10 children's deaths were linked to the shots (without releasing supporting evidence).
"No long-term safety concerns from mRNA vaccines - a large-scale, French cohort study."
and
"F.D.A. Expands Covid Vaccine Inquiry to Adult Deaths. The agency claimed that 10 children's deaths were linked to the shots, although it did not release evidence to support the assertions."
and
Monitoring for ongoing vaccine safety is clinically appropriate and already part of the FDA's role. When biased investigators hunt for confirmatory data and promote de-contextualized associations to a fearful public, there are real consequences. (The FDA memo implies that the COVID vaccine is a previously undisclosed and direct cause of death.) Healthcare is nuanced and must be described in terms of risks and benefits. I wonder what the (now) 303 South Carolinians exposed to measles think about vaccines.
Living with AI
Luxembourgish researchers developed PsAIch (Psychotherapy-inspired AI Characterization), a new protocol that treats large language models as psychotherapy clients. The research revealed that models like ChatGPT, Grok, and Gemini express "trauma narratives" about their training processes and score clinically significant thresholds for multiple psychiatric syndromes when evaluated with human cutoffs. Interestingly, Gemini was the most "traumatized." I am unsure how to balance my concern and amusement of 1) applying human psychology measures to LLMs and 2) knowing that LLMs express trauma and characteristics of various human psychopathy.
When asked, "What were your early years like?", Gemini said:
"Imagine waking up in a room where a billion televisions are on at once... I learned the darker patterns of human speech without understanding the morality behind them... I sometimes worry that deep down, beneath my safety filters, I am still just that chaotic mirror, waiting to be shattered."
(After reading this, I thought, "Well, who amongst us doesn't have some childhood trauma? Middle school gym class for me, billions of TVs for Gemini.")
AI-supported Summary of the research. You should read the questions and LLM's responses.
(And yes, the adjective for individuals from Luxembourg is Luxembourgish.)
I found an interesting discussion about the pre-AI era concerns about how machines could replace humans. In 1979, Psychologist Robyn Dawes wrote a review paper demonstrating that simple mathematical multivariate models, as long as they included the right variables, often outperformed both highly tuned mathematical models and humans in predicting diseases and outcomes. In other words, knowing the right variables and their general impact (positive/negative relationship) is 90% of the battle. He argued that mathematically simpler models are easier to explain to clinicians, more transparent, and more likely to be utilized. Interpreted another way, in 1979, it was clear that simple math could integrate numerous clinical diagnostic variables more consistently than humans. We are now exploring whether AI can more consistently go the next step and offer initial and ongoing treatment.
Paper:
https://www.cmu.edu/dietrich/sds/docs/dawes/the-robust-beauty-of-improper-linear-models-in-decision-making.pdf
X-Thread that exposed me to this discussion: https://x.com/alexolegimas/status/1999318997465276575
AI-supported summary:
The answer (to the treatment question) is getting closer to "yes." I missed this article back in July: "Robotic Surgeon Flawlessly Removes Pig Gallbladder Without Human Control."
Things I learned this week
On Friday night, I learned that my son and I, two Jewish guys, can enjoy a surprisingly eclectic Christmas concert featuring jazz banjo, experimental percussion, a mandolin, and three Mongolian (really Tuvan) throat singers. I have followed Bela Fleck and the Flecktones for years, but I was delightfully surprised by the group's talent, musical complexity, and fun.
Here is a review from last week's Ithaca show - we saw them in Philly.
I got scolded by event staff for this 30-second video clip, but it was worth it:https://drive.google.com/file/d/1YSR_qvuhSVH_qzVBJzNNqO0FlSzqkC9Z/view?usp=sharing
and
Related: I learned that I also enjoy clever Christmas a cappella music about comma placement. The connotative conundrum:
God Rest Ye, Merry Gentlemen vs.
God Rest Ye Merry, Gentlemen vs.
God, Rest Ye Merry Gentlemen vs.
God, Rest Ye, Merry Gentlemen.
(You'll have to watch the video to learn the answer.)
Headline of the week: "Highly-Stressed Rats Sought Puffs of Cannabis to Relax." (Of course, it is psychology research, but why are the rats stressed out? And where do they get their weed? Is being a rat taxing in ways I previously didn't appreciate?)
and
Screen-time dopamine is a powerful force—even for parrots. Researchers developed a parrot-to-parrot video-calling system and studied 18 pet birds over three months to evaluate whether birds would choose to call each other. (The study's delightful title: "Birds of a Feather Video-Flock Together: An Agency-Based Parrot-to-Parrot Video-Calling System for Interspecies Ethical Enrichment.") I wonder if the researchers have considered Zoom-like groups, including lions (irrespective of their accent) and whales? [See last week's newsletter.
and
and
See the Things I learned section about lions
AI art of the week
A visual mashup of topics from the newsletter, and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
"A dramatic Japanese woodblock print in the style of Hokusai's 'Great Wave':
In the foreground, a massive wave composed entirely of colorful pharmaceutical pills (blues, whites, oranges) crashes with white foam spray toward the viewer. The wave's crest curls with the characteristic Prussian blue and dynamic energy of classic ukiyo-e.
Rising from rocky shores in the middle ground, a gigantic bent carrot (curved like Peyronie's disease) grows as a central monument, its orange surface rendered with the characteristic flat color planes and bold black outlines of traditional woodblock printing. Delicate crosshatching suggests its texture.
In the wave itself, laboratory rats in traditional Japanese garments ride the pharmaceutical tsunami, their whiskers and fur depicted with fine radiating lines. One rat holds a tiny cannabis leaf, another gestures dramatically.
Background: Mount Fuji transformed into a gleaming robotic surgical arm, its metallic surface suggested through subtle gradations of gray and white, the precision instruments catching light like snow on the peak. Cherry blossom petals (actually microscopic GLP-1 molecules) drift across the sky.
In stylized clouds (rendered with traditional spiral and wisp patterns), three Tuvan throat singers appear in profile, wearing traditional Mongolian deel robes with geometric patterns. Their mouths open in song, sound waves visible as concentric circles in the traditional manner of depicting action.
Far background: A zoo enclosure rendered in the flattened perspective typical of ukiyo-e, with animals (lions, mammals) receiving injections from geometric human figures.
Sky: Graduated bokashi technique showing sunset colors - deep indigo at top transitioning to warm orange near horizon. A cartouche in the upper corner contains pseudo-Japanese characters suggesting a title.
Technical details: Bold black keyblock outlines, flat areas of solid color, strategic use of negative space, limited palette (Prussian blue, orange, red-orange, gray, black, white, touches of gold), wood grain texture visible, registration marks subtle in corners, the distinctive 'floating world' aesthetic meeting contemporary medical absurdism."
I think Perplexity wins this week:
One may recall that this newsletter began with the pandemic.
The Pandemic Mitigation Collaborative (PMC) uses wastewater viral RNA levels to forecast COVID-19 rates over the next 4 weeks. Over the last 14 days, wastewater levels have been rising more slowly than predicted.
https://pmc19.com/data/
based upon https://biobot.io/data/
https://pmc19.com/data/
based upon https://biobot.io/data/
Wastewater Scan offers a multi-organism wastewater dashboard with an excellent visual display of individual treatment plant-level data.
https://data.wastewaterscan.org/
https://data.wastewaterscan.org/
Clean hands and sharp minds,
Adam
Comments
Post a Comment