Mechanism #19: Taxonomic Inertia
In February 2026, GRAIL released the results of the NHS-Galleri trial — the largest randomized study of a multi-cancer early detection blood test ever conducted. 142,000 participants. Three years of screening in NHS England. The test found four times more cancers than standard screening.
It failed its primary endpoint. There was no significant reduction in late-stage cancers — the only thing that actually matters for survival.
GRAIL's stock crashed 51%. The company is extending follow-up by six months, hoping the trends reach statistical significance. Detailed results are expected at ASCO 2026.
But the failure was not a surprise to the people who study overdiagnosis. H. Gilbert Welch, the Dartmouth screening researcher, called the fourfold detection increase "worrisome." Former NCI director Barnett Kramer warned the test would "dip more and more deeply into the iceberg of disease" — finding lesions that look like cancer to a pathologist but may never behave like cancer in a patient. Richard Houlston, a cancer geneticist, was blunter: stage shift is an "unreliable surrogate for real patient benefit."
The Galleri trial is not a story about a test that didn't work. It's a story about a system that cannot distinguish between finding disease and helping patients — because the word it uses for both is the same.
The Iceberg
The reason more detection doesn't mean more lives saved is that most "cancer" found by modern screening will never kill anyone. The evidence for this is not speculative. It comes from autopsies.
Prevalence of occult papillary thyroid cancer in autopsies when tissue thinly sectioned. Up to 100 million Americans carry subclinical thyroid cancer.
Of Black men aged 70-79 have prostate cancer at autopsy. 36% of white men. Most never knew. Most would never have been harmed.
Of papillary thyroid cancers diagnosed in the US (1991-2019) met the definition of overdiagnosis. Three to four times higher in women than men.
These reservoirs are not a secret. Ahn et al. (NEJM, 2014) documented South Korea's thyroid screening epidemic: a 15-fold increase in diagnoses from 1993 to 2011, with zero change in mortality. When screening was reduced after public concern, incidence dropped. Deaths stayed the same. Chen & Haymart (Nature Reviews Endocrinology, 2025) confirmed: most of what screening finds was always there, doing nothing.
Japan ran the most clarifying experiment. From 1985 to 2003, they screened infants for neuroblastoma — a cancer that can spontaneously regress. Twenty years of data: zero mortality reduction, massive overdiagnosis. The screening preferentially found slow-growing tumors that would have disappeared. Japan stopped in 2004. Canada, the US, and Germany followed. After screening ended, incidence dropped. Mortality was unchanged.
The iceberg was always there. Technology just got good enough to see it. The problem is what happens next.
The Word
What happens next is the word "cancer." And the word is not a description. It is an instruction.
Two randomized experiments prove this. In Omer et al. (JAMA Internal Medicine, 2013), 394 women without breast cancer history were presented with an identical clinical scenario describing DCIS — ductal carcinoma in situ, a condition where abnormal cells are confined to the milk ducts. The only variable was what they called it.
| What they called it | Chose surgery | Chose nonsurgical |
|---|---|---|
| "Noninvasive breast cancer" | 47% | 53% |
| "Breast lesion" | 34% | 66% |
| "Abnormal cells" | 31% | 69% |
Same cells. Same prognosis. Same clinical scenario. The word alone shifted 16% of women from surveillance to surgery. p < .001.
Nickel et al. (JAMA Otolaryngology, 2018) replicated this with thyroid cancer: 550 Australian adults shown identical low-risk papillary thyroid scenarios. "Papillary thyroid cancer" produced the most anxiety and the highest preference for surgery. "Abnormal cells" produced the least. Same cells. Different word. Different life.
The word "cancer" is not neutral information. It is a trigger that activates a cascade: psychological terror, family alarm, treatment expectations, surgical referral, insurance coding, malpractice liability. Once the word is spoken, the cascade is nearly impossible to stop. The patient who hears "cancer" and chooses surveillance must resist not just their own fear but their family's fear, their employer's concern, their doctor's training, and a legal system that they believe will punish inaction.
The Phantom
That legal system is a phantom.
Chang et al. (Annals of Surgery, 2024) searched every malpractice database in the United States — Westlaw Edge and LexisNexis, all 50 states, from 1990 to 2022 — for malpractice cases involving active surveillance for cancer.
of thyroid cancer physicians cite malpractice concern as a barrier to recommending active surveillance
successful malpractice suits for choosing active surveillance. Across all cancer types. In 32 years of litigation records.
Of the five AS-related cases found (all prostate): two were prisoners arguing that AS violated the Eighth Amendment (courts ruled AS was appropriate). Three were patients suing for NOT being offered surveillance after suffering complications from aggressive treatment.
The malpractice fear that drives overtreatment is empirically unfounded. The actual legal risk runs the other direction — toward patients harmed by unnecessary surgery. But the fear persists because it doesn't need evidence. It needs only the word "cancer" and the question: what if you're the doctor who didn't act?
Fifty Thousand Women a Year
DCIS — ductal carcinoma in situ — is the clearest case. It accounts for about 30% of all breast cancer diagnoses in the US, roughly 50,000 women per year. Only 10-15% of untreated DCIS progresses to invasive cancer at 5.5 years. Yet 98% of women diagnosed with DCIS receive treatment. 28% undergo mastectomy. One-third of those also get the other breast removed prophylactically.
If only 20-30% of those treated derive benefit, that is roughly 40,000 women per year receiving surgery, radiation, or both for cells that would never have harmed them. They endure the pain, the scarring, the anxiety, the follow-up appointments, the reconstruction decisions — all activated by a word.
The LORD trial in the Netherlands offered women with low-risk DCIS an honest choice between active surveillance and conventional treatment. When given that choice — without the gravitational pull of the word "cancer" dominating the conversation — 76% chose surveillance.
The patients, when honestly informed, already know what the system cannot admit.
The Proof It Can Be Done
In 2016, the Endocrine Pathology Society removed the word "carcinoma" from one specific thyroid diagnosis. The condition formerly known as "noninvasive encapsulated follicular variant of papillary thyroid carcinoma" became NIFTP — "noninvasive follicular thyroid neoplasm with papillary-like nuclear features."
One word changed. Carcinoma became neoplasm.
~46,000 patients per year worldwide are affected by this reclassification (~10,000 in the US). In the original prospective study, 109 patients treated with lobectomy alone were followed for 10-26 years. Zero adverse oncologic events. Over 50% had previously been receiving unnecessary radioactive iodine ablation. The WHO adopted the reclassification in its 4th edition (2017) and 5th edition (2022).
NIFTP is proof of concept. Remove the word "cancer," and the treatment cascade weakens. Outcomes don't worsen. Unnecessary harm decreases.
A caveat matters here. An Ontario population-based study (Annals of Surgical Oncology, 2019) that retrospectively reclassified older thyroid cases found a 9.4% disease failure rate among NIFTP patients over 15.3 years — higher than zero. But this used looser criteria applied to cases that predated the 2016 refinement. The prospective study with strict criteria showed zero events. The boundary of the label matters: rename too broadly and you under-treat; rename precisely and you save thousands from harm. The lesson is not that relabeling is risk-free. It's that it can be done carefully, and when done carefully, it works.
Thirteen Years of Nothing
If NIFTP proves the concept, the IDLE proposal proves the resistance.
In 2012-2013, an NCI-convened working group led by Esserman, Thompson, and Reid published a proposal in JAMA: reserve the word "cancer" for lesions with a reasonable likelihood of lethal progression. Everything else — intrathyroidal papillary thyroid cancer under 1cm, low-grade DCIS, Gleason 6 prostate — should be reclassified as IDLE: indolent lesion of epithelial origin.
The proposal was clear, evidence-based, and backed by the National Cancer Institute. It called for an Institute of Medicine panel to oversee the reclassification.
Thirteen years later, the word "cancer" still applies to all of them. DCIS still carries the C-word in its name. Gleason 6 is still called cancer (though "grade group 1" terminology is slowly spreading). Only NIFTP — one narrow thyroid variant — has been successfully relabeled.
Why? Thompson himself acknowledged the challenge: reclassification would be "an expensive undertaking" requiring prospective research that "may not be completed for 20 years." But the deeper barriers are structural:
The Mechanism
Every previous entry in this taxonomy involved evidence that was wrong, hidden, manipulated, aggregated incorrectly, or unfalsifiably defended. This one is different. The cells are real. The histology is accurate. The pathologists are right. The evidence is correct.
The damage comes from the category.
"Cancer" was a useful word when it meant "cells that will kill you without treatment." That was true for centuries. Diagnostic technology then advanced to detect entities that meet the cellular criteria — abnormal growth, atypical morphology — but not the behavioral criteria. The cells look like cancer under a microscope. They do not act like cancer in a body.
The word doesn't know the difference. It activates the same institutional response regardless: surgery, radiation, chemotherapy, anxiety, follow-up, billing. The response was designed for the lethal version. It runs on the indolent version because no one changed the label.
Taxonomic inertia: a classification category created in an era when it reliably predicted lethal outcomes continues to activate institutional responses designed for lethal disease, after technology detects entities meeting the category's criteria but not its outcome profile. The label does the damage. The cells do not.
What's Coming
The Galleri trial is not an anomaly. It is the mechanism operating at industrial scale. GRAIL's blood test did exactly what it was designed to do — find more cancers. The problem is that "find more cancers" and "save more lives" are not the same thing, and the entire screening industry is built on the assumption that they are.
The MCED (multi-cancer early detection) market is projected to grow rapidly. Galleri is already commercially available in the US — sold directly to consumers for $949 — before any randomized trial has shown it reduces mortality. Ruth Etzioni, a biostatistician who studies screening, noted that "the extent to which the fate of these cancers is being changed by being diagnosed by the test is unclear."
NCI's Philip Castle was more concise: "I'm sorry that it didn't work."
But it may never "work" in the way the industry hopes — not because the technology fails, but because the word that the technology delivers carries a weight the biology doesn't justify. You can build a test that finds every cancer cell in the human body. If the system then treats every finding as a threat, the test will generate treatment without generating benefit. More detection, same mortality, more harm.
The fix is not better tests. It's better words.
NIFTP proves it can be done. The LORD trial proves patients want it. The malpractice record proves the legal risk is imaginary. The NCI proposed it thirteen years ago.
Fifty thousand women a year are still waiting.