📷 Nature is obviously a fan of Escher.

“It was a small trial, just 18 rectal cancer patients, every one of whom took the same drug. But the results were astonishing. The cancer vanished in every single patient, undetectable by physical exam, endoscopy, PET scans or M.R.I. scans.”

You shouldn’t judge an article by its headline, so how about the first few paragraphs?

“Dr. Luis A. Diaz Jr. of Memorial Sloan Kettering Cancer Center, an author of a paper published Sunday in the New England Journal of Medicine describing the results, which were sponsored by the drug company GlaxoSmithKline, said he knew of no other study in which a treatment completely obliterated a cancer in every patient. “I believe this is the first time this has happened in the history of cancer,” Dr. Diaz said.”

So far we have “astonishing”, “vanished in every single patient”, “obliterated”, and the “first-time-in-history” gambit. Anything else?

“Dr. Alan P. Venook, a colorectal cancer specialist at the University of California, San Francisco, who was not involved with the study, said he also thought this was a first. A complete remission in every single patient is “unheard-of,” he said.”

“Unheard-of”! So be it.

Not until paragraph 18, long after it praised lack of toxicities in a 12-patient trial of a drug with a known side effect profile, and shortly after mentioning the C-word sandwiched between “remarkable” and “unprecedented” is it revealed that

“The inspiration for the rectal cancer study came from a clinical trial Dr. Diaz led in 2017 that Merck, the drugmaker, funded. It involved 86 people with metastatic cancer that originated in various parts of their bodies. But the cancers all shared a gene mutation that prevented cells from repairing damage to DNA. These mutations occur in 4 percent of all cancer patients.”

The “mutation” in questions is mismatch repair (MMR) deficiency is not actually a mutation, but that is this report’s least egregious journalistic error.

MMR deficient tumors respond well to immune checkpoint inhibitors. One of them, pembrolizumab, has broad FDA approval for all “advanced” (i.e. metastatic or unresectable) cancers with MMR deficiency. The goal of this most excellent study which absolutely should have been done is to see whether it can be used even earlier, to avoid possibly debilitating but potentially curative surgery.

And lo and behold, it can! Congratulations to the study team, immune checkpoint inhibitor manufacturers, and most of all to those people who are yet to develop an MMR-deficient tumor which is still resectable, but now maybe doesn’t need to be resected because it will melt away with immune checkpoint inhibition. NB: this is significantly less than the quoted 4%, which includes people whose cancers are advanced and who can already receive ICIs.

Note that I did not learn any of this from the new article, but rather from being a medical oncologist and reading the NEJM paper which is at least — let it not be said NYT does everything wrong — linked to early on.

How will others read it, one wonders?

“The Cotswolds are like Disneyland, they said. And I suppose that was meant to be pejorative. But as a kid I grew up going to Disney World, over and over, … and having the time of my life.”

I like Disney World, but it looks like I would really like the Cotswolds.

“We can’t keep making so many scientists tear out their hair over well-intentioned but near-infinite administrative requirements, as if that doesn’t distract from the actual science that they are supposed to be doing.”

Yes!

“Many breakthroughs in scientific progress have required massive funding and national coordination. This is not one of them. All that needs to be done is allow expert research scientists to do the hands-on work that they’ve been trained to do.”

Spot on.

“…the sheer amount of regulation is so voluminous that if I had to actually read the guidelines that they want us to know about, I would never again be able to submit a grant”.

In communist Russia, the grant writes you!

Silly me, it’s not Russia, it’s the US of A.

To get yourself in the right frame of mind before reading this book, try watching a few optical illusion videos. There is no reason to think our visual cortex is any dumber than the rest of the brain — in fact, quite the opposite. That our inference can be so easily fooled in a domain which is supposedly our strong suit is humbling.

Our statistical inference is even worse, so a short book or two on statistical numeracy should be in everyone’s library. Gerd Gigerenzer’s Calculated Risks can be that book for most people. The assumption, easily defensible, is that “most people” will get more use out of understanding frequentist rather than Bayesian probability. After all, most probabilities people are bombarded with — your chance of dying from breast cancer with and without screening, the chance of your neighbor being the killer given a positive DNA match (you know, the day-to-day stuff) — is frequentist.¹

The only reservation to wholeheartedly recommending Calculated Risks to everyone is that it falls into the category of “blog post books”, if you believe that most non-fiction books should, in fact, have been just blog posts. Or, since blogs are out of vogue, a 15-minute YouTube video may do. Or perhaps a single sentence: use natural instead of relative frequencies (e.g. 1 in 10.000 instead of 0.01%). Let your faulty cortex fill in the rest.

📚 Calculated Risks was funnier than you’d expect from a book about statistical (in)numeracy. It’s healthy to laugh in the face of our inadequacies.