Apart from looking like he has just been on the losing end of a fistfight, and having occasional bouts of nausea, Ben Sasse seems to be doing as well as someone recently diagnosed with metastatic pancreatic cancer possibly could. Both the nausea and his face peeling off are because of daraxonrasib, a new drug which targets KRAS G12 mutations which are common in many cancers but are found in most pancreatic ductal adenocarcinoma (PDAC). As a reminder, PDAC is the one that Steve Jobs did not have, the one that has the dubious distinction of being both the most common and the most lethal cancer of the pancreas.

Well, daraxonrasib seems to be doing its job and doing it well, based on a company press release. Remember, most press releases should not count as evidence for anything. This particular one, however, is worth reading because it is (1) for a randomized controlled trial with (2) a “hard” endpoint of overall survival ⊕ [Note: OK, putting my pedant hat on, the pre-specified co-primary endpoints are progression-free survival (PFS) and overall survival (OS) in the RAS G12-mutant population. What is reported in the press release is only OS in the “intent-to-treat” which is to say both G12-mutant and wild type populations, which was a secondary endpoint. A bullet point at the beginning says that all primary and key secondary endpoints were met, so why not report both? Probably because one looked better than the other, but would it not be a tad suspicious that a less targeted population did better than the more targeted one? But this is just speculation, let’s see review the actual data once they come out. ] which will (3) be presented at the ASCO annual meeting, I imagine as a plenary talk, in early June of this year. The thing to look for there will be informative censoring, in particular early censoring of frail participants — the ones more likely to die early of their disease — who were randomized to receive daraxonrasib but then withdrew due to the “manageable” toxicity of a melting face. The fact that there are no participant numbers reported at all in the release makes me suspicious, though information on the number of patients enrolled is readily available: 501. That’s a lot of patients!

The company is certainly feeling optimistic: they have already received a National Priority Voucher from the US FDA and will now submit a New Drug Application. Kudos and congrats for designing and testing a working drug without using AI, because to read both professional and lay media the past two years it is a miracle there were any drugs being discovered until Large Language Models came along.

Yes, I had to invoke AI, because it is becoming exceedingly common for people to give algorithms credit where it is not due. This is what Tyler Cowen wrote yesterday about pancreatic cancer research:

AI and the pancreatic vaccine. More testing is needed, but there is a reasonable chance that we have a good treatment for pancreatic cancer, and AI was instrumental in that. It is mRNA as well, so a double burn on the haters.

The link is to a post on X by one Rotimi Adeoye, a “contributing opinion writer @nytimes” (one guest essay as of today which is one more than I have so congratulations, I guess?) who in true X fashion superimposed a screenshot from an uncredited journal abstract over someone posting a link to an NBC news article about the updated results of a phase 1 trial of an mRNA vaccine for pancreatic cancer. ⊕ [Note: For those not keeping track, you are right now reading a blog post about a blog post about a retweet of a tweet about a news article based on a press release. You’re welcome. ] These were presented yesterday at the annual meeting of the American Association for Cancer Research but were hinted at in a press release (?) from Memorial Sloan Kettering, where the vaccine — generic name autogene cevumeran which rolls right off the tongue doesn’t it? — was being tested.

Remember how a few paragraphs above I had implied that you should ignore most press releases? Well, news on academic websites should rank even lower as no one there has to answer to the SEC. The primary study was great for what it was, a first-in-human trial with laboratory endpoints meant to test whether the participants’ immune system responded at all to the vaccine. And it seems that it did, as shown in not one but two papers in Nature published two years apart. The number of original participants, all of whom had early-stage, freshly resected and otherwise untreated PDAC upon enrollment, was 19. Three of these did not make it to the vaccine as they had progression, died, or had toxicity from adjuvant chemotherapy before being dosed. Chemotherapy? Yes, in addition to the vaccine everyone also received “adjuvant” (meaning: there to “clean up” any residual cancer after surgery) chemotherapy (FOLFIRINOX, not for the faint of heart) and immunotherapy (atezolizumab which is in comparison to the chemo a walk in the park but even that has its side effects). There was no control.

Of the 16 participants, 8 were “responders” to the vaccine as measured by some highly sophisticated laboratory tests — not that the patients would care what their blood work showed — and in 7 of those the cancer hasn’t come back for 3 years as noted in the follow-up Nature paper or for 4-6 years as noted in yesterday’s update. This compares to 2 of 8 who were “non-responders” to the vaccine.

If you don’t have your calculator handy let me do the math for you: 9 of 16 patients, or 56.25%, with newly resected PDAC who received chemotherapy, immunotherapy and the vaccine were still alive more than 3 years after treatment. You may not know this, and I didn’t until I looked it up just now as it has been a while since I have treated patients with newly diagnosed early-stage pancreatic cancer, but the median OS after (modified) FOLFIRINOX alone in a recent large, randomized Phase 3 trial was 53.5 months, with 43.2% of patients still alive 5 or more years. Did the addition of atezolizumab and the vaccine change anything? I can’t tell and neither can anyone else until there is a randomized controlled trial, which isn’t to cast shade on the investigators — kudos to them as well for a successful first-in-human study — but let’s curb our enthusiasm.

So we have some updated results from a tiny trial that didn’t really move the needle one way or another, and yet Cowen et al. feel the need to push AI into the narrative. To be clear, there is absolutely no mention of LLMs, machine learning, algorithms or artificial intelligence of any kind anywhere in the autogene cevumeran literature. Granted, it is a “personalized” vaccine, meaning that every potential participant had their tumor sequenced and up to 20 vaccine targets identified among the newly mutated proteins. I am sure there was a lot of computation involved. But not every sophisticated computer analysis is AI, let alone an LLM, so I truly don’t see how they could legitimately be brought into the conversation.

And in case you were wondering, no, the screenshotted abstract did not in fact back up Adeoye’s claim. Best as I can tell this was the paper in question, a speculative review article in an obscure journal written by a Shanghai-affiliated group of authors who had nothing to do with BioNTech whose purpose was to be a never-looked-at reference for a false claim, that “AI played a critical role in advancing the vaccine”. Anything for the clicks, am I right?

Adeoye’s behavior was regrettable but Cowen’s is detestable, especially when paired with his look-at-the-sheeple attitude towards humans. ⊕ [Note: The linked to article from Cowen is particularly wrongheaded if you realize who the Luddites really were and that the label should in fact be a positive one. ] Cory Doctorow had warned about AI companies over-promising their capabilities for a short-term gain. But they don’t really need to: there are plenty of useful fools willing to promise on their behalf, giving it credit even where there is none.

• Monopolized by David Dayen, which Cory Doctorow recommended in response to my account from last week of the medical billing/phone scam rabbit-duck. Doctorow wrote about the book in more detail back in 2021 and yes it is now on the pile as is everything else below.
• The Credibility Crisis in Science ⊕ [Note: ↬ Joel Hamill ] by Thomas Plümper and Eric Neumayer, and if the subtitle “Tweakers, Fraudsters, and the Manipulation of Empirical Results” whets your appetite there is an excerpt available in Nautilus.
• The Art of Manliness by Brett and Kate McKay, who have had a blog of the same name for more than a decade, so it is a true mystery why an article in The Dispatch about the “gentlemanosphere” ⊕ [Note: ↬ Reader John ] chose to highlight overtestosteroned almost-douchebags such as Scott Galloway as the anti-manosphere crusaders rather than McKay. Haha, I’m joking, of course it’s not a mystery: Galloway gets more clicks, taps, swipes or rather the preferred method of interaction is nowadays. He also has a new book out, to which I shall not link.
• The Oxford Book of English Verse edited by Christopher Ricks, which someone recommended on The Honest Broker Podcast, a show that combines some of my least favorite things about this decade: Substack and using the word “podcast” to describe a video of two men talking. Still, the transcript makes for good reading and the book seems something worth keeping at the bedside, at least aspirationally.

• Bayesian Workflow, a new book from Andrew Gelman et al, coming out in June. Here is Gelman’s announcement and here is the book’s web page.
• The Alba Madonna, ⊕ [Note: ᔥ Tyler Cowen ] a painting of Raphael’s at the National Gallery of Art, which you can examine in detail on their website.
• The Classical Station, ⊕ [Note: ᔥ William Parker ] a no-nonsense, no-tracking stream of human-selected music. Here is their daily playlist.
• just the fracking QR code, ⊕ [Note: ᔥ Jahir Fiquitiva on Mastodon ] a no-nonsense, no tracking website that will give you the QR image for any text, file or hex code you desire.
• Jellyfin, a home media server which resembles Plex before it was enshittified. Note that (1) I am yet to try it out and (2) the person whom I first saw mention it does not in fact endorse it.

Per Tanner Greer, China is all-in on leading the new industrial revolution, which will of course first be a revolution in science and technology. My only insight into the tech part is that Chinese companies now make the most innovative earphones and the most electric cars, and from reading Apple in China I have a vague sense that they were contract manufacturers in those fields first, then perfected the craft and outdid their former employers.

Most relevant to the world is what they [Note: But who are “they”? CCP leadership which sets the tone, startups and conglomerates which do the actual legwork, or the provincial bosses who act as intermediaries? I won’t even pretend to know. ] have done and will do with renewable energy, batteries, quantum computing and the like. The most relevant to me is what they are doing with biomedical research, particularly in cell & gene therapy. And as the field progresses from discovery to manufacturing — both research-grade and commercial — to clinical trials and actual practice so does the extent of their involvement.

On the discovery side, Chinese authors have already overtaken their American counterparts in the number of papers in top scientific journals. A crude metric for sure — and let’s not get into the whole peer review debate — but it is a measure of general lab activity. Determining whether China overtaking the US is a direct or indirect result of American sinophobia or evidence that the sinophobia was justified I will leave as an exercise for the reader.

In manufacturing, China is in the “let us do your work for you” stage. Not a week goes by that I don’t get an email from a China-based CDMO [Note: Contract Development and Manufacturing Organization, which just to make things confusing is often shortened even further to Contract Manufacturing Organization, or CMO, which means so many different things within the same field (Chief Medical/Manufacturing/Marketing Officer, anyone?) that it has become ridiculous. ] and WuXi Biologics — headquarters in Shanghai — is among the top 5 companies of that type in the world. These companies specialize in being taught how to do things and then doing them at scale, picking up a few things about process development and optimization on the way. Why the US would not want CDMOs from China was so obvious that the bill forbidding federally funded entities from contracting with them sped through Congress, though not after being revised to not explicitly mention WuXi by name.

“They” are not taking the same approach with clinical trials: a wise decision on their part, since the only thing they could have possibly learned from American trialists was inefficiency and bloat. No, the big push is for investigator-initiated trials which can speed up approvals — they only require local oversight, not a national FDA-equivalent to give their blessing — and get cell & gene treatments to patients within weeks, provided of course that they are manufactured in China.

An eternity ago I wrote about two different types of trials, ones meant to introduce new treatments and others meant to establish guidelines and prune less effective therapies. I argued that the regulatory burden for the two should be different, since the latter — Set 2 as I called them — are clearly less risky and carry less uncertainty. We have of course already experienced the world of unregulated trials of new therapies, and it was not pretty. With its IIT push, China is speeding them up further. The risks are clear, but who benefits?

Top of mind are the patients who will get effective treatments sooner. The invisible graveyard will shrink! But of course most treatments won’t work and most patients who take part in IITs and their family members will have spent much time, some even money, to get false hope. A much larger proportion of the beneficiaries will be companies that hit a false positive, and in single-center IITs of small sample size there will be many of those. They will of course be willfully ignorant of the fact that their high response rate in a 10-patient single-arm open-label trial was spurious, but will instead get just enough conviction to pass on the asset to the greater fool. And so will the roulette wheel keep spinning, fueled by financialization that somehow ends up being the root cause of most of the world’s current ills.

I fully expect the American biotech establishment to learn the wrong lesson from Chinese IITs and instead of feeling relief that someone else is taking on the foolish risk be pressured into copying them. Here is to the steadfast box-checkers and blank faces whom I trust to put on the brakes.

• Raghuveer Parthasarathy: Space mirrors, solar panels, fools, and their money.

A few months ago I noted that the one of the main reasons biotech was not like tech was its almost unlimited freedom do bullshit. Well, people are able to raise money by BS in other areas as well, as this article shows, but an order of magnitude less because most investors are able to do back of the envelope calculations.

• Bryan Vartabedian: The Measure of Everything.

A take on Goodhart’s Law as applied to medicine, this time through the lens of instrumentalisation. If any of these articles tickle you and you haven’t yet read Zen and the Art of Motorcyle Maintenance, please do so now. Ted Gioia recently wrote about the book and its legacy. As an occasional note-taker I am in the minority club of Lila fans as well, though both of Pirsig’s books are due for a re-read.

• James Olds: The Chronology Problem.

The point is in the subtitle: “how our bias towards recency in scientific discovery hurts our understanding”. It rings true, and even reminded me of the 26 years it took for CRISPR/Cas systems to travel the path from an oddity to a gene editing platform, until I realized that those 26 years were not spent idling as this review in Cell describes in detail. So, the (lack of) developments in theoretical biology would be a much better example.

• Phil Price: Ted Williams and Me.

Ted Williams was, apparently, a base-ball player about whom John Updike had this to say: “For me, Williams is the classic ballplayer of the game on a hot August weekday, before a small crowd, when the only thing at stake is the tissue-thin difference between a thing done well and a thing done ill.” This applies to any profession you can imagine, and indeed things outside of one’s professional life. People who have the inner drive to do things well even in the absence of stakes could unkindly be called “perfectionists”, but let’s remember that they are the ones who keep the wheels of civilization in motion, in opposition to the hordes of blankfaces, lazy asses and morons.

• Aaron Cheng: On the emotional weight of a life in medicine. Thoughts of a newly-mintend hematologist which act as a good counterbalance to the popular fictional accounts. Feel free to peruse the few pre-2017 posts from the Infinite Regress archive for my own thoughts from that period.
• Fred Vogelstein: Enveda’s drugs from plants will turn pharma upside down. This is what Ruxandra Teslo warned us about a few months back.
• Joseph Howlett for Quanta Magazine: The Man Who Stole Infinity. On Georg Cantor, plagiarist. It does cause one to become more cynical, as we keep discovering that high-profile scientists built their reputiations on the backs of people with more scrupules and less need for attention. Taleb’s minority rule applied to sociopaths.
• Adam Mastroianni: The one science reform we can all agree on, but we’re too cowardly to do. It is to abolish for-profit scientific publishers, and I am all for it. But, people being people, I do not have high hopes that this prisoners’ dilemma will have a positive outcome.

• Derek Lowe: The Best Ideas Are Not Always Enough.

Drugs which look great in those cellular machinery flow charts with boxes and arrows pointing every which way, and which may even cure a few genetically monstrous and wholly artificial lab mice, tend to flop where it matters. Lowe links to 11 such examples and writes in more detail about the twelfth.

• Adam Kroetsch: Why clinical trials are inefficient. And why it matters.

An overly long article ⊕ [Note: ᔥNintil ] with which I don’t completely agree ⊕ [Note: For example, Kroetsch describes the role of a site investigator as resembling “that of a glorified data entry clerk - the investigator’s primary responsibility is gathering the data that the drug company needs and sending it to them”. This is incorrect: site investigators usually have clinical research coordinators and data managers to do it for them. But this deserves a post of its own. ] but which nevertheless provides a good overview of the many things wrong with how clinical trials are being conducted in the US, the biggest one being that they are reinventing the wheel each and every time they are done. The “lean trial” proposal at the end matches my own thinking.

• Ruxandra Teslo: A response to Dario Amodei on AI & clinical trials.

Teslo picks up on the tech bro magical thinking streak in which things you don’t sufficiently understand seem eminantly solveable using the most recent technological developments. Five years ago it was electronic medical records and blockchain, now it’s clinical trials and AI. The article gives the many reasons why things are not that simple. Now, if we all agreed on the set of LLM prompts that would provide an unbiased protocol and informed consent form review thus eliminating as many people from the loop as possible, well, then we may be on to something.

• Alex Tabarrok: If you have the right to die, you should have the right to try!

If someone qualifies for euthanasia, should they also not be eligible for every expanded acces, compassionate use, right-to-try scheme imaginable? Obviously: yes. Maybe not so obviously: there is a branch of my subspecialty aptly named desperation oncology which in the vast majority of cases leads to false hope, financial ruin and, worst of all, time misspent in doctors’ offices and infusion clinics instead with your loved ones. As a doctor and a human being I am partial to life, so I see state-assisted dying programs like Canada’s MAID as monstrous, but “you’d rather be dead so here, take this drug” is only a half-step above that qualifier and leads to the bad reputation of experimental therapies.

• Rebecca Robbins and Gina Kolata for the NYT: Grail’s Cancer Detection Test Fails in Major Study. Some of my earliest tweets — now protected, so apologies for linking if you haven’t already followed me — were about Grail and why their project to screen healthy people for early signs of cancer would likely fail. Nine years and billions of dollars later…
• Peter Bonate: Drugs Explained. A layperson’s primer on drug development from a pharmacologist, available on GitHub.
• Colin Gorrie: How far back in time can you understand English?. One thousand years of the English language in one blog post.
• Joel Hawksley: How I built Timeframe, our family e-paper dashboard. It was quite the journey, interrupted by the Boulder County Marshall Fire which burned down more than a thousand buildings in late 2021. “On the night of December 31–January 1, heavy snowfall put an end to the fire.”
• ConchCat: I accidentally ate lunch with a cult. The cult in question is Twelve Tribes, accused of exploiting its followers as free labor, along with assorted abuses of women and children. But the sandwiches were good!

• Martha Lane Fox: The Price of initiative just collapsed. On the lag between invention and implementation, from printing press to AI.
• Scott Shambaugh: An AI Agent Published a Hit Piece on Me. A taste of what’s possible, with follow-up.
• Steve Blank: You Only Think They Work For You. About contracted service providers. The article is about public relations, but applies just as well to contract research organizations (CROs) and these complex dynamics are part of the reason clinical trials are so expensive.
• Andrew Gelman: The 80% power lie. A clear example of why frequentist statistics, so favored by conservative regulators, institutional review boards and scientific review committees, are more often than not based on false assumptions.
• Dr. Drang: Chinese New Year and Ramadan. The pseudonymous doctor often shifts between programming and scripting languages, from R through Python to Mathematica which coincidentally matches my own amateur programmer journey. For this post he used Emacs Lisp and I am very much interested in Emacs and Lisp right now due to recent experimentation with Linux. Saliency FTW.

I admire the fight for clinical trial abundance, I truly do, but the longer I think about it the more convinced I am that making nips and tucks to federal law and FDA guidance is the wrong way to go about it.

I have been involved in clinical trials for more than a decade and have seen it through the eyes of an investigator, sponsor and IRB member directly, and owing to a few friendships with ex-FDA employees, from the eyes of regulators as well. The common thread between all cases of delay and all frustrated attempts to speed things up were not laws and regulations but humans. The blankfaces and boxcheckers. Those IRB members who prefer the list of possible toxicities in a consent form in table form as opposed to a list, or a list as opposed to a table, and with percentages of expected occurrences instead of qualitative statements of possibility, but can you please explain what does percentages mean in qualitative terms, and is it “toxicity” or “adverse event”, and you should list only true toxicities not hypotheticals, except when we require you to add each potential adverse event — sorry, toxicity — whether it happened in this or other trials or not, and for each round of these changes there is another 3-week turn of the IRB roulette during which a different person may view the edits and give their own two cents — leave their own fingerprint.

And that is only the IRB! Their are personal imprints to be made at every level of clinical trial desing and implementation, between different departments of the Sponsor, “key opinion leaders" [Note: Or “KOLs”. The whole KOL ecosystem is worth writing about at length, just not today. ] sitting on advisory boards, administrators and worriers-in-chief at clinical trial sites, and even, rarely, clinical trial investigators themselves although in the US at least they are the ones least involved in clinical trials. More often than note, the way to leave your fingerprint is by citing a rule or a law or precedent from the FDA or the IRB or some other state-level regulatory agency whether or not said rule, law and/or precendent apply.

To be clear: most people in the clinical trial ecosystem are not like this. But much like airport noise complaints, in which one household accounted for almost 80% of all complaints, it only takes a handful of people to gum up the works and make it appar like the whole system has conspired against smooth trial execution.

So ho do we achieve abundance? Cutting down on the laws that govern trials is an option, and would certainly reduce the degrees of freedom at which blankfaces and boxcheckers operate. But America is a country of min-maxers and I can easily see things going sour with drive-by trials and phase 1 clinics being run out of a storage unit in South Florida. ⊕ [Note: Which operate even now, so just you wait for the walls to come down and the minmaxing trial beast to be unleashed. ] Chesterton’s fence and all that.

Could we not reduce the number of gummer-uppers in the chain instead? There are plenty of trials that are quick to start, quick to execute, and not even all that costly, largely due to people being reasonable. Can we inject more reasonable people into the ecosystem, if we can’t inject more reason into people who are already there?

This thought brought me to Delaware, the second smallest state of the USA that has 60% of Fortune 500 companies incorporated there owing to the convenience, flexibility and predictability of its corporate law system. To my layperson’s eyes it seems, from the example of Delaware, that its not the law itself that makes the difference but rather who implements it and how — which was on spectacular display earlier this week.

Convenience, flexibility, predictability.

Now clearly the Delaware of clinical trials can’t be Delaware — it is far too small and wouldn’t have enough trial candidates. But California, Texas, Florida and New York could all vie for the spot, each having a population similar to Australia, the phase 1/2 promised land. Texas was the first to make some waves, tiny and of an uncertain direction, but in the ballpark of what would be needed. The only changes that federal law needs are those that would allow the states to compete among themselves in being clinical trial champions.

If that last image made you think of China and its competing provinces, it is for a good reason: the Chinese have been eating everybody’s lunch for the past few years, in large part because of the vicious internal competition elaborated on elsewhere. There are other lessons there, for some other time.