Single-source review
·
Science communication / debunk
·
July 2026
Jess Steier's Unbiased Science post of 11 July 2026 dismantles the revived claim that COVID vaccines cause fast-moving "turbo cancer," working through three sources driving the wave: an oncologist's clinical impression, a case-report review, and a large South Korean cohort study that has since drawn a formal expression of concern from its own journal. It matters because the piece is a clean worked example of how weak causal claims get assembled from real data, and because the same Korean paper's dose-response analysis argues against the claim it is being used to support. The caveat is that this is advocacy science writing rather than primary research, and it is arguing a position the author held before she started.
The Fallacy Inventory
Eleven reasoning failures called out in the piece. Several are not classical fallacies but epidemiologic biases, which is the more useful category here.
1
Post hoc ergo propter hoc
Cancer appeared after vaccination, therefore vaccination caused it. Temporal sequence stands in for causation.
A single clinician's caseload has no control group. You can observe co-occurrence, never attribution.
3
Selection bias by clinical setting
An oncology practice concentrates recurrence, delayed diagnosis, and heavy surveillance. It is the worst possible vantage point for eyeballing a new pattern.
4
Anecdote elevated to evidence
A stack of case reports is still a stack of case reports. Volume does not convert observation into inference.
5
Reverse causation / latency mismatch
Solid tumors do not arise within twelve months. A cancer found in that window predates the exposure.
6
Surveillance and detection bias
Vaccinated people were more engaged with the health system. Look harder at a group and you find more of what was already there.
The flagged cancers are the screened ones. Thyroid is the standing example: incidence climbed for years while mortality stayed flat.
Test dozens of cancer types and several clear the threshold by chance. No correction, no biological thread linking the ones that did.
9
Cherry-picking against the source's own data
The dose-response test came back flat at HR 1.01. Advocates cite the paper while discarding the analysis that would have supported the mechanism.
10
Suppression as vindication
Criticism recast as censorship, a server outage recast as silencing. The argument becomes unfalsifiable by construction.
11
Confounding by the more plausible exposure
By late 2022 nearly everyone had been infected. The unvaccinated comparison group carries more spike and inflammation, not less.
At a Glance
- Author & outlet
- Jess Steier, DrPH, writing at Unbiased Science. Public health doctorate, science communication practice, not the investigator on any of the studies discussed.
- Source type
- Newsletter debunk. Secondary synthesis, not primary research.
- Funding & conflicts
- Subscriber-supported newsletter with a paywall. The commercial model rewards clarity and reader trust, and it also rewards taking a firm side. Steier's professional identity is built on debunking vaccine misinformation, which is a real prior, not a disqualifying one.
- Load-bearing citation
- Kim et al., Biomarker Research 2025;13:114 — 8,407,849 Korean National Health Insurance enrollees. Expression of concern posted 22 October 2025.
The Research She Is Reviewing
What the Korean study did. A retrospective cohort drawn from the national insurance database, 2021 to 2023, matching vaccinated against unvaccinated enrollees and counting cancer diagnoses over a one-year window. It ran as a Correspondence item rather than a full research article.
What it found. Elevated one-year hazard ratios across six cancers: prostate 1.687, lung 1.533, thyroid 1.351, gastric 1.335, colorectal 1.283, breast 1.197. The authors state in their own supplementary materials that the findings do not establish causal relationships.
What the advocates skip. The dose-response analysis, which is the version of the claim that could actually be tested. Overall cancer risk across dose counts came out at HR 1.01, confidence interval sitting on the null. Lung, colorectal, breast, and thyroid showed no increase with additional doses. Leukemia fell. And the most-dosed group was older and sicker by design, which should have manufactured a spurious gradient. It still came back flat.
The paper's own null result is the strongest evidence against the claim it is being used to prove.
The population-level check. US cancer registries show incidence dipping in 2020 as diagnoses were delayed, then resuming its prior trend. Mortality kept falling. Passive and active surveillance systems caught myocarditis and adenoviral thrombosis within months of rollout. A system that sensitive would not plausibly miss a wave of aggressive malignancy.
Strengths and Weaknesses
Strengths
- Engages the strongest version of the opposing claim rather than the easiest. The dose-response framing is the sophisticated argument, and she takes it head-on.
- Uses the cited study's own internal data to defeat it, which is more durable than attacking the authors or the journal.
- Names a real limit unprompted: a decades-latency carcinogen would not yet be visible, so the long game is genuinely unsettled.
- Treats the clinical anecdote respectfully. A sharp oncologist noticing a pattern is where signals start, and she says so before explaining why noticing is not answering.
- Distinguishes the plausible question (does infection wake dormant tumor cells) from the implausible one, and reports that the human follow-up did not replicate the mouse finding.
Where I Would Push
- The piece describes the Korean paper as a research letter. It is indexed as Correspondence, Kim et al. 2025;13:114. Small, but if you requote this, use the citation, not the descriptor.
- The hazard ratios are characterized as "bumps" without being reported. Prostate at 1.687 is not a small number. Publishing the figures alongside the biases is the stronger move.
- Several links are described rather than named. For a relay-fidelity audit, the reader needs to reach the underlying sources.
- An expression of concern is not a retraction. The piece leans on it fairly hard, and expressions of concern sometimes resolve with no action.
- The closing line — "turbo-cancel this claim" — slightly overshoots a body of evidence she has just carefully described as incomplete on long latency.
Relay Accounting
Four layers between the original data and a reader's decision about their next dose. Layer one: Korean insurance claims, which record billing events, not clinical truth, and cannot see prior infection. Layer two: the Correspondence item, which compresses that into six hazard ratios and buries the causal disclaimer in supplementary material where most readers never go. Layer three: advocacy citation, which strips the disclaimer, the dose-response null, and the expression of concern. Layer four: this newsletter, which restores the missing context but strips the effect sizes going the other direction.
Every layer removed something. The pattern worth noticing is that a burial in supplementary materials did more damage than any single bad actor downstream. The authors said the honest thing. They said it where nobody reads.
CMIO Lens
Reviewer bias disclosure & dialysis angle
Adam's priors run with this piece, and that is the risk worth naming. A nephrologist and CMIO who has spent years watching claims-derived signals get overread is primed to accept a debunk that says surveillance bias explains the finding — which is exactly the diagnosis-momentum failure Woolever describes, arriving at the right answer through insufficiently examined agreement.
The dialysis-specific angle is not in the article and is worth carrying separately. ESKD patients are among the most intensively surveilled populations in medicine, with thrice-weekly clinical contact and frequent imaging. Every detection-bias mechanism described here is amplified in that population. Any future analysis claiming a vaccine-cancer association in dialysis cohorts will be more susceptible to these artifacts, not less.
The procurement translation: if a clinical decision support tool or retrieval layer ingested the Korean paper before October 2025, it may carry the six hazard ratios without the expression of concern attached. That is a version-drift problem, not a content problem, and it is the kind of thing that should be asked about explicitly during vendor review.
So What
Take it seriously. The reasoning holds and the fallacy inventory is genuinely reusable, though read it as a well-argued brief from someone who came in with a side, and go pull the hazard ratios yourself before you quote it.
Source-type triage: science communication, so GRADE, NNT, RoB 2.0, and CONSORT were dropped as inapplicable. Substituted lenses: relay accounting, prior plausibility, comparator identification, and reviewer-bias disclosure. Two anchor claims verified independently: the Kim et al. hazard ratios and the 22 October 2025 expression of concern.
Sources
Primary source: Steier, J. "No, COVID vaccines still aren't giving people cancer." Unbiased Science, 11 July 2026. unbiasedscipod.com
Load-bearing citation: Kim et al. "Increased cancer risk after COVID-19 vaccination." Biomarker Research 2025;13:114. Expression of concern, 22 Oct 2025