The Dose That Never Arrives

This is the one subject in the season that arrives unasked, forwarded by people who are frightened and reading carefully. So we do it properly, and we start with the refusal rather than ending on it. The laboratory work is real. A 2010 drug screen found ivermectin kills acute myeloid leukemia cells in preference to normal blood-forming cells. A 2022 affinity purification pulled a named human protein, TELO2, out of a cell extract on an ivermectin column. A 2024 experiment stopped a lung cancer line from ever becoming paclitaxel-resistant. Then we do the arithmetic slowly, with the conversion shown on air. The concentration that halved a mouse lymphoma line's viability in 24 hours was 10.55 micrograms per milliliter — 10,550 nanograms. The peak measured in the blood of twenty-five onchocerciasis patients given a standard oral dose was 52.0 nanograms per milliliter. That is 203-fold, and 93.2 percent of what circulates is stuck to plasma protein, leaving about four nanomolar free to enter a cell. Raising the dose is tolerable, roughly proportional, and still does not close it. Then the whole human record, named and classed: one flagged-retracted self-report cohort whose authors sold the capsules, a six-patient series allocated by electrical readings at acupuncture points, a melanoma that shrank and then grew, a woman intubated after seizures, and one registered trial nobody has read a result from. Beside it, mebendazole, which has been tested further and missed; fenbendazole, which has liver injuries and a retracted remission series; and atovaquone, the one antiparasitic that measurably changed a human tumor at its ordinary licensed dose — which is the control that tells you the difference here is not politics.

Show notes

Two numbers, in the same unit, in the first minute, so nothing I say afterward can be mistaken for permission.

The first comes from a paper published in April 2025. Shukla and colleagues treated Dalton's lymphoma cells — a mouse T-cell lymphoma line, meaning cancer cells kept alive and dividing in a dish — with ivermectin for twenty-four hours and calculated the half-maximal inhibitory concentration, the concentration at which half the cells stop being viable. The shorthand is IC50. For ivermectin it was 10.55 micrograms per milliliter; for cisplatin, the standard anticancer drug compared against it in the same assay, 8.32 (Shukla 2025). Now convert, because the argument hides in the unit change. A microgram is a thousand nanograms. So 10.55 is 10,550 nanograms per milliliter.

The second has been measured in people and nobody disputes it. Baraka and colleagues gave twenty-five patients infected with Onchocerca volvulus — the worm that causes river blindness — one oral dose of 150 micrograms per kilogram, the standard human dose. Maximum plasma concentration: 52.0 nanograms per milliliter at 5.2 hours, with fourteen healthy volunteers measured alongside behaving much the same (Baraka 1996). A second series, nine patients, peaked at 38.2 plus or minus 5.8 (Okonkwo 1993).

So say it out loud. Ten thousand five hundred and fifty against fifty-two is a factor of 203; against thirty-eight point two, 276.

The refusal goes at the front, where nobody can clip around it. This hour is not telling you to take ivermectin. Not as a hedge — as the conclusion of that arithmetic, and because of a 73-year-old woman with metastatic breast cancer who read about ivermectin online, took high doses of it, and was intubated after seizures (Saperstein 2026). You will meet her at the end of the hour. I name her now because she is why these numbers are not an academic exercise.

One qualification that cuts against me. A twenty-four-hour assay in which the standard chemotherapy drug also needs micrograms per milliliter is an insensitive test system, so 10.55 is not ivermectin's typical potency in culture. And the same paper reports a substantial reduction in tumor size in tumor-bearing mice — it has an animal arm, and that belongs on the record beside the figure I just read — but it gives no blood concentration for the dose that produced it, which is why the IC50 is the only number in it that can be set against a human peak.

So what is the drug. Not a laboratory invention: one bacterium, one sample of dirt. Crump and Ōmura's own account states that it originates solely from a single microorganism isolated at the Kitasato Institute in Tokyo, from Japanese soil, in the late 1970s — Streptomyces avermitilis (Crump & Ōmura 2011). Ivermectin is a dihydro derivative of what it makes, and belongs to the macrocyclic lactones: drugs built around one very large carbon ring closed by an ester bond. The bottle holds avermectin B1a and B1b, 80 to 20 (Löscher 2023). Animals first, as a veterinary product; the 2018 review that gathered this literature dates approval for human use to 1987 (Juarez 2018). A Nobel Prize followed in 2015 (Van Voorhis 2015); by 2025 more than 5.9 billion treatments had gone out (Nicolas 2026). It is licensed in people for two worm diseases: river blindness, and strongyloidiasis — threadworm, the one worm American practice routinely names (Sharmeen 2010). That is what it is approved for, before you hear what it is being bought for.

Now the conflict in the record. That account is a review, not a new measurement — PubMed types it as one — co-authored by one of the laureates, with the Kitasato Institute printed as its affiliation: the institute the paper credits as the drug's sole origin. It does mean "originating solely from" was written by the party with most to gain from the word solely.

The molecule itself — the sixteen-membered ring that gives the class its name (Juarez 2018), the fused hexahydrobenzofuran, the two oleandrose sugars, all from the compound record, PubChem CID 6321424 — I hand to the ivermectin page, with two corrections for it to carry: that unit is not the flat benzofuran the page calls it, and the hydroxyl on it was shown to be needed for full efficacy at the glycine receptor, by comparing selamectin with ivermectin, not for potency at chloride channels. Why is it considered safe? Two mechanisms. The first: the antiparasitic effect runs through the glutamate-gated chloride channel, a pore in invertebrate nerve and muscle that opens when glutamate binds it and floods the cell with chloride, shutting it down. Vertebrates do not have that channel (Löscher 2023).

The second: P-glycoprotein — one pump under three names, because the gene names mdr1 and ABCB1 and the abbreviation P-gp all mean this protein. At the brain's blood vessels it throws ivermectin back into the blood; inside a tumor cell, as you will hear, it throws chemotherapy back out. Schinkel and colleagues deleted its gene: mice homozygous for a disruption of mdr1a were viable, fertile and phenotypically normal — and a hundred times more sensitive to ivermectin (Schinkel 1994). Mice, and a complete deletion, not the partial variation people carry. The pump's own history goes to the new P-glycoprotein page.

Now the half the word safe conceals. In mammals ivermectin is not selective. Löscher lists what it activates in us: GABA-A and glycine receptors, the brain's two main inhibitory brakes, and the nicotinic acetylcholine receptor, an accelerator (Löscher 2023). Chen and Kubo add the P2X purinergic receptors, which open to ATP outside the cell — ATP being the molecule a cell runs on and spills when damaged — a bile-acid sensor, and a potassium channel (Chen & Kubo 2018). Both are reviews, not new measurements.

So safe is a condition, not a property, and Löscher prints the condition. Ivermectin penetrates the mammalian brain poorly and exerts no effect through those receptors, quoting him, "unless it is used at high, potentially toxic doses or the blood-brain barrier is functionally impaired." And in animal seizure models, he reports, the evidence is equivocal, and in at least some of them the dose that works in half the animals sits in the range of the dose that kills half of them (Löscher 2023).

The clinical picture of a barrier that does not work is a dog. Ivermectin-sensitive collies carry a four-base-pair deletion in mdr1 that cuts off manufacture of the pump; two copies give the sensitive phenotype, one or none does not (Mealey 2001). That is where the story is usually left. Bissonnette and colleagues sequenced the gene in 28 dogs of other breeds that developed neurotoxicity on daily macrocyclic lactone treatment: twenty-seven had no copy of it, one was a carrier, and the authors point instead to other drugs handled by the same pump, excessive dosing — their word, one explanation among several — and variants elsewhere (Bissonnette 2009). A case series with no comparison group, so no rate, but the genotype test does not catch every case.

Now the laboratory work, at its true strength. I will name the evidence class inside every sentence; that is where this argument is won and lost.

Sharmeen and colleagues screened licensed drugs for anything that killed leukemia cells; ivermectin came out of it. In glassware it killed acute myeloid leukemia cell lines and fresh patient samples at low micromolar concentrations, in preference to normal blood-forming cells, and the mechanism was worked out, not asserted: chloride flooding in, the plasma membrane hyperpolarizing — the inside going more negative than normal, which silences the cell — and, hold this, explicitly no change in mitochondrial membrane potential. Reactive oxygen species rose and were functionally required for the killing. And it synergized with cytarabine and daunorubicin, the two drugs acute myeloid leukemia is actually treated with — which is where the one promising idea in this hour comes from. In mice, growth was delayed in three models of leukemia, in the paper's own words for that mouse work, “at concentrations that appear pharmacologically achievable” — and its closing line is that ivermectin “could be rapidly advanced into clinical trial for leukemia” (Sharmeen 2010). Those two sentences are what the human measurements contradict.

The PAK1 claim needs its date fixed first: it is usually cited as 2012, which is when PubMed indexed it; the paper is from 2009. Hashimoto and colleagues report that in glassware ivermectin blocks PAK1, a kinase — an enzyme that switches other proteins on by adding a phosphate — in human ovarian cancer and schwannoma lines, at 5 to 20 micromolar (Hashimoto 2009). Four pages, a small journal, no DOI, and its mouse result relayed at second hand.

Melotti and colleagues ran a repositioning screen — testing drugs already licensed for something else against a new target — for blockers of the WNT-TCF pathway, a growth-signaling route switched on permanently in several cancers. In glassware ivermectin repressed that pathway's target genes, and its low-concentration effects were rescued by switching the pathway back on with an engineered activator — the control showing the drug worked through the pathway, not around it. In mice carrying transplanted human tumors it selectively inhibited the growth of grafts that depended on TCF and not of those that did not (Melotti 2014). Inhibited growth: slowing, not shrinking. And its closing move urges a relative, selamectin, into clinical testing instead.

Yin and colleagues were studying something else: DDX23, an RNA-unwinding enzyme glioma cells use to make more of a small regulatory RNA called miR-21. Ivermectin appears there as an off-the-shelf helicase inhibitor. In glassware it lowered miR-21 and blocked invasion; in mice carrying transplanted human tumors it decreased glioma growth by, in the authors' own word, "potentially" inhibiting that enzyme (Yin 2015).

Dominguez-Gomez and colleagues found ivermectin by computational resemblance to salinomycin and tested it on one breast cancer line in glassware. Growth was inhibited between 0.2 and 8 micromolar — 0.2 is the lowest figure among the studies in this hour — and the stem-like fraction was hit harder than the bulk, the opposite of paclitaxel's pattern in the same experiment (Dominguez-Gomez 2018). One cell line. And in endocrine-resistant breast cancer cells ivermectin suppressed estrogen-driven proliferation (Rujimongkon 2026) — an abstract with no concentration in it. This work is real and consistent, and anyone who tells you it is junk has not read it. Which is precisely why the arithmetic matters.

One more, the strongest mechanistic result in the file. Yonezawa and colleagues immobilized ivermectin B1a on a solid support and ran cell extract past it — affinity purification: whatever sticks becomes a candidate for the drug's target. What stuck was TELO2, a cofactor of the phosphatidylinositol 3-kinase-related kinases: ATR and ATM, which sense damaged DNA, DNA-PKcs, which repairs it, and mTOR, which governs cell growth. Ivermectin bound it through a helix at the protein's tail end, and mutations in that helix conferred resistance — the experiment that turns a binding partner into a target. Ivermectin itself reduced TELO2, those kinases, and the phosphorylation of AKT and S6 kinase, two relays in the growth signal beneath mTOR (Yonezawa 2022).

It is the molecular partner to the 2014 pathway screen, and it is also, exactly, biochemistry and cell culture: no animal work, no concentration in its abstract. And several of its authors are at the Ōmura Satoshi Memorial Institute at Kitasato University, named for the drug's co-discoverer. An interest, not misconduct, and it belongs beside the finding.

Two corrections. The previous edition built its whole mechanism on this protein, calling it TEL2, and cited a 2023 review of the complex it sits in, which says only that Tel2 “has recently been found to be a target of ivermectin” (Bhadra & Xu 2023) — one sentence in an abstract about something else, naming no primary paper. Citing the review instead is how a finding gets weaker and louder at once. Second, proportion: one affinity-purification result from 2022, in a dish, is one of six results here, not the mechanism of ivermectin in cancer.

The immunology, and the one result that does not ask the drug to kill anything

One more preclinical result belongs here, because it does not ask ivermectin to kill anything by itself. Draganov and colleagues (2021), in mouse breast tumor models, reported that ivermectin induced immunogenic cancer cell death — cell death that leaves wreckage the immune system reads as a threat — and drew T cells into the tumor, raising the ratio of effector T cells to the regulatory T cells that restrain them. They call it “an allosteric modulator of the ATP/P2X4/P2X7 axis”: it does not open those surface channels itself but changes how they answer the ATP a damaged cell spills. Then the sentence that matters most: “neither agent alone showed efficacy in vivo.”

Combined with an antibody against PD-1, the brake on T cells that checkpoint drugs release, the pair limited tumor growth and promoted complete responses, relapse fell when it was given before and after surgery, and modeling confirmed what the authors call bona fide synergy, p of 0.007 and below. These are mice.

The status of that paper has changed twice since 2021. In February 2026 the authors published an Author Correction: errors in Figure 4 panels D and F, the panels blocked out because “the original data are no longer available to replace the incorrect images,” and “the overall results and conclusions have not changed” (Draganov 2026). On 5 June 2026 the journal posted an Editorial Expression of Concern — a notice that flags a paper while leaving it in print, stating no finding of misconduct. It records concerns about multiple highly similar images between Figure 4d and 4f, says the publisher found similarities among three further image pairs, and advises readers “to interpret these data with caution” (Editorial Expression of Concern 2026). Neither alleges fraud. Those two notices, a retraction notice, and PubMed's “Retracted Publication” type are four different things, kept apart on the new post-publication-review page. The figures in the favorable animal paper most often sent to me can no longer be checked against anything.

Underneath the result is a second problem: the axis itself. Nörenberg and colleagues (2012) tested ivermectin's reputation as the tool that tells P2X4 responses from P2X7. It potently enhanced human P2X7 currents, and potentiated currents in human macrophages grown from a person's own blood. On rat or mouse P2X7 it was “only poorly effective.” So one of the two receptors the mouse tumor result invokes is one ivermectin barely touches in a mouse and acts on strongly in a person. That cuts both ways: it is not the same experiment in the two species.

A correction here too. The summary version of this literature says ivermectin kills cancer cells by collapsing the mitochondrial membrane potential, the voltage a mitochondrion holds across its inner membrane, which falls when a cell commits to dying. Two of the better papers disagree. Shukla and colleagues (2025) report exactly that fall in their 24-hour lymphoma experiment. Sharmeen and colleagues (2010) found hyperpolarization of the outer membrane explicitly “without changing mitochondrial membrane potential.” The error is in neither paper; it is in reciting a contested mechanism as established. Reactive oxygen species: supported. Mitochondria: contested, unresolved.

Now the one result I think is genuinely promising, and what it is not. Hayashi and colleagues (2024) took A549 cells, a human lung cancer line, and made them resistant to paclitaxel the way resistance actually develops — long exposure to gradually rising concentrations. The resistance arrived with up-regulation of ABCB1 messenger RNA, the working copy of that gene's instructions, and overproduction of the protein it encodes: that same pump, doing the same job from inside a tumor cell. Ivermectin given during that exposure “completely abolished P-gp expression,” raised the paclitaxel inside the cells and kept them sensitive, by a route through the epidermal growth factor receptor. The class of that evidence: one cell line, in glassware, no animal arm, no concentration in the abstract.

Two readings make it a finding rather than a slogan. Jiang and colleagues (2019) found the same route in three drug-resistant human lines — colorectal, breast and chronic myeloid leukemia — and in two mouse models carrying those resistant cells. Ivermectin, at what they call a very low dose causing no obvious toxicity alone, reversed the resistance in dishes and in mice. And they state the mechanism plainly: ivermectin reverses resistance “mainly by reducing the expression of P-glycoprotein (P-gp) via inhibiting the epidermal growth factor receptor (EGFR), not by directly inhibiting P-gp activity.”

That contradicts the paper that started the multidrug-resistance claim. Didier and Loor (1996) reported that ivermectin restored the retention of two P-glycoprotein probes in resistant leukemia cells; their abstract gives no concentrations. The comparator, SDZ PSC 833, is valspodar: the history of pump blockade in one molecule. Baer and colleagues (2002) tested it in a randomized phase 3 trial in untreated acute myeloid leukemia patients aged 60 and over, and closed the valspodar arm after 120 patients for excessive early mortality: death in 44 percent against 20, p = .008. Now the half that cuts the other way: disease-free survival did not differ, 7 months against 8, about a third of each arm alive at a year — and the valspodar arm was given lower doses of two chemotherapy drugs to allow for the interaction, part of why the early deaths happened. The early deaths are the finding; the trial is not evidence that the pump is worth leaving alone.

So, the honest summary. Preventing resistance alongside real chemotherapy is a better hypothesis than replacing it, and the only one here that does not need ivermectin to reach a micromolar concentration by itself: it asks the drug to turn a gene down, not to kill a cell. It rests on one cell line in glassware, one cells-and-mice paper, and no human test at all.

What the drug would have to reach

Which brings us to the arithmetic, done slowly, because it is the hour. Laboratory papers report micromolar, a count of molecules in a volume. Human blood levels come in nanograms per milliliter, a weight in a volume. To join them you need the weight of a mole. Ivermectin B1a is C48H74O14, about 875 grams per mole, from the compound record, PubChem CID 6321424 — not from any paper here, because no abstract in the file states it. So one micromolar is about 875 nanograms per milliliter. Check it without trusting me: Schmith and colleagues (2020) call one antiviral target 2 micromolar; Jermain and colleagues (2020) call the same target 1,750 nanograms per milliliter. Two times 875. The new drug-exposure page carries the rest of the vocabulary.

Now everything in one unit. The lymphoma figure, 10.55 micrograms per milliliter, is 10,550 nanograms per milliliter: about 12 micromolar. The PAK1 range, 5 to 20 micromolar, is 4,400 to 17,500: 80 to 340 times the 52.0 measured on a standard dose. And the lowest concentration among the studies cited here, the 0.2 micromolar at the bottom of Dominguez-Gomez's range in one breast cancer line, is about 175: more than three times that standard-dose peak. Now the thing that cuts against me, because raising the dose closes that gap at its lowest end and nowhere else. Six hundred micrograms per kilogram daily for three days, measured in a moment, produced a peak of 165.5 nanograms per milliliter — within six percent of that line's 175. On total drug, the smallest number in the glassware and the biggest measured in a person are the same number. What raising the dose does not close is the figure that governs whether a drug gets into a cell at all.

But the measured peak is not the number that matters. Klotz and colleagues (1990) measured ivermectin's plasma protein binding in five healthy people by equilibrium dialysis — letting the drug settle across a membrane proteins cannot cross, so the free drug can be counted: 93.2 plus or minus 4.4 percent bound. Drug stuck to a protein cannot enter a cell; only the free fraction can, about 7 percent. So at a 52 nanogram peak the free concentration is roughly 3.5 nanograms per milliliter, about 4 nanomolar — against 10,550, a factor of about three thousand. Run the same sum on the high dose, and this is the comparison that survives it: 7 percent of 165.5 is about 11.6 nanograms per milliliter, 13 nanomolar — still roughly fifteen times below that 0.2 micromolar, and three hundred times below the 2 micromolar range most of this work used. Those divisions are done on air from separate papers, not lifted out of either; the ivermectin page gives the free share as roughly a fourteenth, and here is that fraction with the sum finished.

The obvious answer is to raise the dose. Follow it, and watch the trap close. Guzzo and colleagues (2002), at the manufacturer, gave 68 healthy adults, randomized three to one against placebo — so about fifty-one on the drug — up to ten times the highest approved dose of 200 micrograms per kilogram. It was generally well tolerated with no indication of central-nervous-system toxicity, and pupil dilation, the main safety measure, looked like placebo. Peak concentration and total exposure were generally dose proportional.

Here is the correction this show owes. The earlier script said that study showed how little a large dose buys. It shows the opposite: dose proportional means ten times the dose buys about ten times the peak. No ceiling in the exposure curve is what makes the modeling decisive rather than reassuring — if exposure is proportional, you can calculate where a bigger dose lands, and it has been done twice. Schmith and colleagues (2020) built a population pharmacokinetic model — a prediction from measurements in many people — and report that simulated total and unbound plasma concentrations “do not reach the IC50, even for a dose level 10x higher than the approved dose,” with predicted lung concentrations of 0.820 micromolar at ten times the label dose against that 2 micromolar target. Jermain and colleagues (2020) independently put peak lung exposure at 772 nanograms per milliliter against the same 1,750. Both are simulations, and both borrowed their lung-to-blood ratio from cattle, the weakest joint in each — and still two independent models landing in the same place.

And now a measurement rather than a simulation. Lapphra and colleagues (2026) gave ivermectin to 24 children with acute uncomplicated dengue at 400 or 600 micrograms per kilogram once daily for three days — open-label, sequential dose escalation, and in the authors' words not designed or powered to evaluate antiviral efficacy. The maximum observed concentrations were 107.50 and 165.50 nanograms per milliliter. Convert the higher one: about 0.19 micromolar. Six of the 24 had grade 4 neutropenia, a severe fall in the white cells that fight bacteria, recorded as serious adverse events; all were clinically well and none was judged related to ivermectin. Four times the standard dose, three times the highest approved dose, three days running, under hospital supervision, and the total blood level is a fifth of one micromolar while the free level is thirteen nanomolar.

One last number, which governs how long any of this lasts in a body. The half-life — the time for the blood level to fall by half — is about 18 hours in Guzzo and colleagues (2002), 35.0 in Baraka and colleagues (1996), and 56.50 plus or minus 7.01 in Okonkwo and colleagues (1993), by two different assays. Three studies, three answers, unreconciled; the drug-exposure page sets them side by side. Anyone who quotes you one settled half-life for this drug is quoting a preference.

What has actually been measured in people

So raise the dose and give it to a lot of people under supervision. That has been done several times, at scale, for a different disease. The cleanest: Naggie and colleagues (2023) reported ACTIV-6 — Accelerating COVID-19 Therapeutic Interventions and Vaccines — a publicly funded platform trial testing several repurposed drugs against one shared placebo group. It randomized 1,206 participants at 93 United States sites: 602 to ivermectin at a maximum targeted 600 micrograms per kilogram daily for six days, three times the highest approved antiparasitic dose, and 604 to placebo. Median time to sustained recovery was 11 days in both arms, hazard ratio 1.02, credible interval 0.92 to 1.13, the Bayesian form of the range of values the data are compatible with. Their conclusion: “These findings do not support the use of ivermectin in patients with mild to moderate COVID-19.” Nor is it alone. An earlier ACTIV-6 arm gave 400 micrograms per kilogram to 1,591 outpatients: 12 days against 13 (Naggie 2022). COVID-OUT randomized 1,323 adults: odds ratio 1.05, confidence interval 0.76 to 1.45 (Bramante 2022). Three large placebo-controlled trials, the same nothing.

Why is a virus trial in an hour about cancer? Because this is the only setting where a high dose has been given at scale, under supervision, against a placebo. The high dose was tolerable, which is the half that gets quoted. And it did nothing — the only direct test of the inference the cancer claim rests on.

The show owes a correction here, the template for the rest. The meta-analysis of 23 randomized trials by Hill and colleagues (2021) circulated most widely, and the version people shared reported a significant reduction in deaths. Their own journal flagged it on 9 August 2021 (Expression of Concern 2021) — and that notice has since been withdrawn with the paper, so its indexed title now reads “RETRACTED: Expression of Concern”. The paper was retracted on 5 February 2022 (Retraction 2022). The abstract PubMed carries today is the replacement, which excludes high-risk-of-bias trials and finds no significant survival effect: risk ratio 0.90, interval 0.57 to 1.42. Then the same authors published the autopsy: “the significant effect of ivermectin on survival was dependent on largely poor-quality studies,” one potentially fraudulent trial carrying a risk ratio of 0.08 against 0.96 in the low-risk studies. And they name it: a preprint from Egypt by Elgazzar and colleagues, retracted from Research Square on 15 July 2021 over ethical concerns, with duplicated data for about 79 participants, deaths dated before the trial began, and plagiarism in the text (Hill 2022).

Retractions cut both ways here, and saying so is what makes the rest of this hour worth your time. PubMed now types as a Retracted Publication a randomized trial of one ivermectin dose in 100 asymptomatic people in Lebanon (Samaha 2021); a Brazilian pilot in 32 patients reporting a dose-dependent fall in viral load (Pott-Junior 2021); and a multidrug case series co-authored by Peter McCullough that called ivermectin “a safe, inexpensive and effective early COVID-19 treatment validated in 20+ random, controlled trials” (Hazan 2022). And the real test of a show's honesty: a randomized trial in Egypt that had reported the opposite, ivermectin cutting survival time from 29 days to 18.3 (Elshafie 2022). A paper saying ivermectin killed people is retracted too.

Now the human cancer record, which begins with an absence that has to be stated as a finding, not a shrug. As of this recording, 9 October 2026, no randomized controlled trial of ivermectin in any malignancy, at any phase, has reported a result in the indexed literature: a PubMed search returns zero records, and no prospective trial with imaging-confirmed tumor response exists either. The version that circulates is weaker — that there is no phase III trial. True, and it understates it. A statement about what exists has to be dated to the day it is made.

Two reviews say the same in print. Patel, Chawla and Parmar (2025) found that preclinical work demonstrates inhibition of proliferation, apoptosis, and modulation of pathways including Wnt/beta-catenin and Akt/mTOR. Their next sentence: “However, clinical evidence in humans is limited, with no large-scale randomized controlled trials (RCTs) confirming therapeutic benefits.” That is a narrative review, the weakest form, because it states no method for choosing what it includes. A 2026 review of the whole field reaches the same place, naming “pharmacokinetic limitations, heterogeneous dosing regimens, and the paucity of adequately powered randomized clinical trials” as what stands in the way (Al-Zoubi 2026) — this hour's thesis, printed by people who are for these drugs. Drug repurposing has its own new Atlas page, with metformin, curcumin and amygdalin as its neighbors; none of them is this argument.

There is one registered trial, and I will tell you exactly how much I know about it: NCT05318469. I could not open the registry record. ClinicalTrials.gov is unreachable from the machine this research was done on, refused by policy rather than by the site, so I will not read you a sponsor, a phase, an enrollment number, a design, a status or a completion date. Where this show has elsewhere called it single-arm with conference-abstract results, that came from our own unverified note, not the registry, and it is struck. PubMed returns zero records for the identifier. The version that circulates names a phase I/II trial with no registry number, no sponsor and no named drug — the shape a trial takes when it does not exist. This one appears to exist. It is a filed plan, and nobody, including me, has read a result from it.

And the National Cancer Institute, at its true strength. The claim in circulation is that it began new preclinical efficacy studies in early 2026. What I can source is one sentence in one peer-reviewed research letter: Rockwell and colleagues (2026) state that “the National Cancer Institute recently announced plans to study ivermectin for the treatment of cancer” — and say in the next breath that rigorous clinical trial evidence does not presently support its use, nor fenbendazole or other benzimidazoles. An announcement of intent is not a result, and I will not date a research program from a news site.

Which brings us to the paper everybody sends me. Hulscher and colleagues (2026), in Anticancer Research, reported a prospective observational cohort — followed forward, no comparison group — of 197 cancer patients prescribed ivermectin and mebendazole off-label through a US telemedicine platform. They took compounded capsules — mixed to order by a pharmacy, not manufactured as an approved product — of 25 milligrams of ivermectin and 250 of mebendazole. Data came from digital surveys at baseline and six months; 122, 61.9 percent, completed follow-up. The headline, in the abstract's words: the “Clinical Benefit Ratio (CBR) was 84.4% (95% confidence interval=77.0-89.8%).” At follow-up, 32.8 percent reported no evidence of disease, 15.6 regression, 36.1 stable disease, 15.6 progression. And in the same abstract: concurrent chemotherapy 27.9 percent, radiation 21.3, surgery 19.7.

Four things about that — three checkable from the free abstract, the fourth from the journal's own notice. First, the outcome. Clinical Benefit Ratio is not a standard endpoint; it is a composite the authors built from patients' own survey answers, counting no evidence of disease, regression and stable disease all as benefit. The rulebook for deciding whether a tumor has shrunk is RECIST, which measures named lesions on imaging against their own earlier measurements. This cohort did not use it. Nobody measured a tumor, and the cancer-pathology page carries how a diagnosis and its extent are established. Second, more than a quarter were on chemotherapy at the same time, a fifth on radiation, a fifth had surgery, any of which shrinks tumors on its own. Third, 75 of the 197 never answered the follow-up survey; if the people doing badly are the ones who stop replying, the remainder looks better than the truth. And fourth, from the notice rather than the abstract: the journal is asking when baseline was assessed relative to treatment exposure, which the abstract does not say.

Then the disclosure, with an arithmetic correction of the show's own. An earlier version said eleven of the twelve authors listed the company that sold the treatment. That is wrong, in the generous direction: on the PubMed record, all twelve list The Wellness Company, Boca Raton — the telemedicine business that prescribed the treatment and supplied the capsules. Hulscher and McCullough also list the McCullough Foundation, Risch the Yale School of Public Health. The seller supplied the patients, the drug, the data and the authorship. That is not an accusation of dishonesty. It describes a design in which no independent party touched anything, which is why a result like this needs a randomized trial rather than standing in for one.

The editorial trail is two notices long, and both have to be read. On 9 June 2026 the Editorial Board of Anticancer Research issued a formal Expression of Concern recording serious concerns about the verifiability, statistical reliability and ethical oversight of the dataset, and opening a “formal Post-Publication Data Integrity and Ethical Oversight Audit” directed at the mandatory ethics-committee documentation, source-verified records confirming the baseline diagnoses of the 197 participants, and objective medical documentation of the reported regressions (Expression of Concern 2026).

The previous edition closed on one line from that notice, and I have re-read it to confirm it is there: “Disclosing limitations does not exempt a clinical dataset from the foundational scientific requirements of empirical verifiability and independent ethical oversight.” I am keeping it, and saying in the same breath what became of the half about ethical oversight. In September 2026 the same board issued an update. “The audit described in the original notice is concluded.” “Those requests have been withdrawn.” “The question of ethical oversight is not pursued and forms no part of the ongoing review.” “The Board draws no adverse inference from the non-production of any material.” And: “No allegation of fabrication, falsification, plagiarism, image manipulation, or other research misconduct is made against any author” (Update to the Expression of Concern 2026).

The weight moves to what the board is still asking, almost exactly the four things I listed: the construction of the Clinical Benefit Ratio composite and the self-reported measure under it; the timing of baseline assessment relative to exposure; the basis for the reported dose and adverse-effect relationship; the representativeness of the follow-up sample; and the handling of participants without follow-up and of concurrent therapy. And the sentence that keeps the other half of the balance, quoted whole this time: “The appeal is under active consideration. No final decision has been taken on whether the article should be corrected, retracted, or otherwise updated.” Retraction is still on the table; so is correction. One last pair of facts, stated and not interpreted: the PubMed record carries the article type “Retracted Publication” and the title prefix “RETRACTED:”, and repeated searches found no separate retraction notice indexed as of 5 October 2026. Both are true at once.

The rest of the record, and the two dewormers beside it

That is the cohort. What is left of the human record is two papers, neither a trial. Guilford and Yu describe six patients with stage IV cancer evaluated between 2011 and 2022 and given combinations of ivermectin, mebendazole, praziquantel, niclosamide and antifungals. All six, they report, outlived their prognosis, with no serious adverse events (Guilford & Yu 2026).

Now how the drugs were chosen. Each regimen was selected by Acupuncture Meridian Assessment, which the authors call a modification of electroacupuncture according to Voll: it measures electrical conductance at points on the skin, and the reading picks the drug. An allocation method is how a study decides who gets what, and one with no demonstrated relation to the patient's disease is not a flaw in the study. It is the study. Five drugs plus antifungals plus dental treatment means nothing can be assigned to ivermectin, and six patients from an unknown denominator over eleven years, with no control, are survivors selected by being written up. Both authors are Science Officer at one private company, and their own word for the results is "preliminary."

The second paper is the honest version of the single-patient story. A 74-year-old man with nodular melanoma had nodal and liver metastases at staging. He declined guideline-directed therapy and pursued lifestyle changes alongside ivermectin and fenbendazole, all on his own. His disease shrank, and his circulating tumor DNA — tumor genetic material shed into the blood — fell from 2.04 to 0.18 mean tumor molecules per milliliter. Then both reversed: it rose to 0.93 and the dominant node grew. His tumor carried at least 50 mutations per million bases of DNA, the extreme upper end of what makes a tumor visible to the immune system, and his own authors conclude that spontaneous immune-mediated regression is at least as plausible as anything he took (Cheng & Araujo 2026). Had he stopped being followed at the low point, this would have been published as a cure.

Two laboratory papers belong beside them. Asano and colleagues treated A549 human lung cancer cells and normal fibroblasts with recombinant methioninase — an enzyme that strips an amino acid cancer cells are unusually dependent on — or cisplatin, or ivermectin, and found for each the concentration that alone cuts viability by 30 percent. For ivermectin it was 2.08 micromolar. Given together at those three figures the drugs essentially eradicated the cancer cells within 24 hours and did close to nothing to the fibroblasts, which needed 6.4 micromolar (Asano 2025). In glassware, one cancer line. And 2.08 micromolar is about 1,800 nanograms per milliliter: thirty-five times the 52 measured in people. Eight of the seventeen authors, including the senior and corresponding author, list AntiCancer Inc., which develops the partner enzyme; the other nine list a university orthopedic department.

The last concedes the argument in its own opening. Sheir and colleagues in Alexandria built ivermectin nanocrystals wrapped in platelet membrane. Their stated reason: repurposing ivermectin for cancer "is hindered by poor solubility and high toxicity, restricting its parenteral administration" — parenteral meaning by injection rather than by mouth. In mice bearing a mouse breast tumor the particles homed to the tumor and slowed its growth (Sheir 2026). Mice. And the number the last edition left out: the coated formulation's own half-maximal concentration is 2.89 micrograms per milliliter on a triple-negative breast cancer line, about 3.3 micromolar — fifty-six times the measured human peak. The people most invested in this drug's cancer future are building delivery systems to avoid giving it the way people are buying it.

Mebendazole belongs here: the drug most people mean by dewormer, and tested in humans further than ivermectin ever has been. The benzimidazoles — mebendazole, albendazole, fenbendazole — bind beta-tubulin and block assembly of the microtubules a cell builds to pull its chromosomes apart when it divides. Lacey set that out in 1988 (Lacey 1988) and stated the paradox two years later: these are the only truly broad-spectrum dewormers, and they act on fungi and mammalian cells too, so why do they not poison us (Lacey 1990). There is no unique worm target. Tubulin is everywhere, and the difference is one amino acid deep — worm beta-tubulin carrying phenylalanine at position 200 is susceptible, tyrosine resistant (Kwa 1995) — plus poor absorption from the gut. That is how thin the margin is.

Now the human ladder, rung by rung; the full record is on the new mebendazole page. Gallia and colleagues ran a phase 1 trial: everyone on the drug, nothing to compare against. Twenty-four patients with newly diagnosed high-grade glioma took mebendazole with temozolomide after radiation; four of the fifteen at the top dose developed severe liver-enzyme elevations, reversing on dose reduction. Median overall survival was 21 months (Gallia 2020), which gets quoted as efficacy. It is a single arm, and its other number is progression-free survival of 13.1 months in the 17 patients who took the drug more than a month against 9.2 in the 7 who took less — a split of one cohort by time on treatment, and patients who stay on treatment longer are already doing better. It is the comparison that flatters a drug.

Krystal and colleagues gave mebendazole to ten children and young adults with high-grade glioma and reported a 33 percent response rate (Krystal 2024) — no control, and given with bevacizumab and irinotecan, which produces responses here by itself. Then the hardest paper in the set. Mansoori and colleagues ran a phase 2a study in treatment-refractory gastrointestinal cancer, dose pushed as far as 4 grams a day, aiming at a serum concentration of 300 nanograms per milliliter. Eleven were enrolled and ten started treatment; only five reached it — the paper gives the five with no denominator, so take the ten. Every patient progressed, and four met criteria suggesting hyperprogression (Mansoori 2021). That group had published a favorable single case seven years earlier, then tested its own anecdote and printed the answer (Nygren & Larsson 2014).

Then the one randomized trial that set a success bar in advance. Patil and colleagues at Tata Memorial Center in Mumbai randomized 88 adults with recurrent glioblastoma, 44 per arm, to lomustine plus mebendazole or temozolomide plus mebendazole. The bar: survival at nine months of 55 percent or more would justify going further. The result was 36.6 percent in the temozolomide arm and 45 in the lomustine arm (Patil 2022). It missed. And read the funding line, because it answers the claim this hour is about: two Indian cancer foundations and the Indian Council of Medical Research, no pharmaceutical sponsor. People with nothing to sell ran the trial.

And one outlier, with my suspicion attached. Hegazy and colleagues in Egypt randomized 40 patients with metastatic colorectal cancer, double-blind, to FOLFOX — the standard chemotherapy backbone for this cancer — with bevacizumab plus either placebo or mebendazole. Response at twelve weeks, 10 percent against 65; median progression-free survival, 3 months against 9.25; both p values printed as "0.000" (Hegazy 2022). No p value is zero; that is a rounding, and printing it that way is a tell about the analysis. It was registered retrospectively, which removes the protection registration exists to give. And a three-month median in a control arm on FOLFOX with bevacizumab is low enough that the arm to question is the control arm. Nothing this size has been replicated.

Fenbendazole is the veterinary one, licensed for animals and never for people, and the whole molecular foundation under the protocol circulating online is one laboratory's work, of which a 2018 paper is the main account. Dogra and colleagues report moderate affinity for mammalian tubulin — moderate is their word — and cytotoxicity to human cancer cells at micromolar concentrations. It moved p53 into mitochondria and inhibited glucose uptake and the sugar transporters, which is the Atlas page on the Warburg effect. Fed orally, it blocked growth of human tumors grafted into mice. Then the authors' own conclusion: fenbendazole "may be evaluated as a potential therapeutic agent" (Dogra 2018). May be evaluated. Not is one.

No human cancer trial of fenbendazole exists in PubMed. What exists is a case literature about the liver; the new fenbendazole page carries all six. An 80-year-old woman with advanced lung cancer on pembrolizumab developed severe liver injury after a month of fenbendazole taken on the strength of social media; it resolved on stopping, and her authors wrote plainly that "she did not experience tumor shrinkage" (Yamaguchi 2021). A Norwegian man was admitted two weeks after immunotherapy with a tenfold rise in cholestatic markers — blocked bile flow rather than dying liver cells. He was given no immunosuppression; the fenbendazole history, 2 grams daily the week before, emerged after that; the markers fell on their own and immunotherapy resumed uneventfully (Skaara 2025). Three more: biopsy-proven liver-cell injury (Thakurdesai 2024), one that would have been called immune-mediated hepatitis until a causality score identified fenbendazole (Krishnan 2026), and one in a patient already taking a liver protectant (Satinsky 2026).

The worst is the 65-year-old man with prostate cancer who took veterinary fenbendazole and ivermectin on alternate days for three months at a dose he described as "one squirt." He presented with jaundice and an ALT of 1,764 units per liter. An R-value of 37.50 — the ratio of liver-cell enzymes to bile enzymes, which tells the two patterns apart — confirmed the liver-cell pattern, other causes were excluded, and his causality score of 9 reads as highly probable (Powderly 2026).

Then the paper people send as proof. Three patients with advanced cancers, two reported in complete remission and one near-complete, after adding fenbendazole to other active therapies — which means that even as published it could not attribute the remissions to fenbendazole. Read its affiliations as printed, which is this hour's rule: two of its three authors list advocacy organizations, the International Society for Orthomolecular Medicine and Association Cancer et Métabolisme; the first author lists Alberta Health Services' Cross Cancer Institute, a public cancer center (Makis 2025). It has been retracted, the statement published 21 January 2026 (Retraction Statement 2026). Anyone sending you that PDF is sending a withdrawn paper. Mebendazole has been tested in people and it went badly. Fenbendazole has not been tested at all, and what stands in its place is a liver.

The control, the harm, and the pattern

A dish against a bloodstream: that has been the hour. Here is the control — one antiparasitic that did something measurable in people, at an ordinary licensed dose. Atovaquone is an antimalarial that inhibits mitochondria. Skwarski and colleagues gave it to patients with non-small cell lung cancer awaiting surgery, at the standard 750 milligrams twice daily for twelve days: a window-of-opportunity trial: the drug goes in the gap between diagnosis and surgery, and the tumor that comes out is measured. Thirty patients were evaluable by hypoxia PET-CT, a scan of how much of a tumor is starved of oxygen: fifteen on the drug, fifteen not. Eleven of the fifteen treated had a meaningful reduction in hypoxic volume, median minus 28 percent against plus 15.5 untreated; adjusted, expected hypoxic volume was 55 percent lower, p equals 0.004 (Skwarski 2021). Now the design, out loud: sequential cohorts, not randomized. Still the first clinical evidence that targeting a tumor's mitochondria reduces hypoxia in a person.

Then the cleanest failure in the field. Niclosamide is a tapeworm drug with real preclinical credentials in prostate cancer. Schweizer and colleagues tested it with standard enzalutamide in men with metastatic castration-resistant prostate cancer. Twenty screened, five enrolled; both patients at the higher dose had a dose-limiting toxicity, the side effect bad enough to stop the dose rising. Maximum plasma concentration ran 35.7 to 182 nanograms per milliliter and was, in their words, "not consistently above the minimum effective concentration in preclinical studies." No declines in PSA, and the board that watches a running trial closed it for futility. Their verdict on their own drug: "Oral niclosamide is not a viable compound for repurposing as a CRPC treatment" (Schweizer 2018; Correction 2018). A real mechanism, a real trial, and a drug that could not be absorbed to the concentration its own laboratory work demanded.

The same lesson, at molecular resolution, running the other way. Imatinib is a leukemia drug. In culture it has, in their words, negative effects on diverse physiological processes in the adult blood fluke Schistosoma mansoni, survival among them. Taken into rodents infected with that parasite, it did nothing to worm burden or egg production. They found why: serum albumin and alpha-1 acid glycoprotein, two blood proteins that bind drugs, soak imatinib up and abolish the killing in culture — and on infection mice carry six to eight times more acid glycoprotein than humans. Their recommendation is the rule this hour runs on: find out whether plasma proteins block an effect in glassware before committing to animal work (Beckmann 2014). Three sentences in circulation fail against it. Anthelmintics in oncology are not "heavily investigated." They do not "starve the tumor" — mebendazole is a microtubule drug, niclosamide a receptor degrader. And leukemia kinase inhibitors are not "proving highly effective against parasitic infections" — the one tested is the paper above, and it proved the opposite.

Meanwhile something is happening to patients, and it can be counted. Rockwell and colleagues took an electronic health record network of 68,373,949 patients at 67 organizations and counted same-day outpatient prescriptions of ivermectin with a benzimidazole against the matched months of 2024. After a January 2025 celebrity podcast endorsement that reached more than sixty million people, that prescribing roughly doubled: rate ratio 1.97, interval 1.70 to 2.29 — a rate ratio being how many times more often something happened than in the period compared against. Among patients with a cancer diagnosis in the previous year, 2.63; within that group, 3.91 in the South against 1.04 in the Northeast, 3.05 in White patients against 1.28 in others, and 2.79 in men against 1.93 in women (Rockwell 2026). I am using this rather than the conference abstract that circulates for the same claim, which was never peer-reviewed. Their own limits: these are orders, not doses swallowed; the design cannot show the podcast caused the rise; and nobody measured whether treatment was delayed, which is what worries me.

Beside the population number, one patient. A 73-year-old woman with metastatic breast cancer self-administered high doses of ivermectin on the strength of information she found online, developed acute altered mental status, generalized seizures and respiratory failure, needed intubation and ventilation, and recovered fully within forty-eight hours. Her authors call it an unintentional high dose and give no figure for it. They note that the Food and Drug Administration has explicitly discouraged ivermectin for cancer outside clinical trials, and that such cases "remain underreported despite growing off-label use driven by online misinformation" (Saperstein 2026). Set that beside the manufacturer's study: sixty-eight healthy adults, three in four of them on ivermectin, took up to ten times the highest approved dose with no indication of nervous-system toxicity (Guzzo 2002) — healthy, screened, supervised. An ill woman on other drugs, taking an unknown quantity for an unknown time with nobody watching, is not the same exposure, and tolerability in the first setting is not permission in the second.

Now the error that organized this literature. It is one sentence. In 2018 Juarez, Schcolnik-Cabrera and Dueñas-González published a review gathering the mechanisms: the resistance pump, Akt and mTOR, WNT-TCF, the purinergic receptors, PAK1 — nearly every result this hour walked through. Then it says that ivermectin's antitumor activities "are achieved at concentrations that can be clinically reachable based on the human pharmacokinetic studies done in healthy and parasited patients," and concludes that this could allow a rapid move into cancer trials (Juarez 2018). Two of that review's authors, including its senior author, also produced the breast cancer paper this hour credits with its lowest concentration: the same group, the same institution in Mexico City. That belongs on the record beside both. The claim was still being restated in 2022, by a liver cancer paper reporting activity at "clinically relevant concentrations," retracted in February 2026 (Lu 2022; Retraction 2026), alongside a retracted 2019 glioma paper (Liu 2019). Measured against the published human peaks, the reachability sentence is wrong, and almost every downstream claim inherits it.

Now that inheritance in a patient's hands. One chatbot conversation told this show's owner that oncology screens had "debunked" the view of ivermectin as merely neurotoxic, and that it is a "potent direct YAP1/TAZ inhibitor" — YAP being a growth-signaling protein a cell uses to decide whether to keep dividing, TAZ its partner — "now being rapidly repurposed and tested in clinical oncology." And it supplied a citation whose every field matches the record: Xuan and colleagues, 2025, PubMed identifier 41220567 — the number the index files it under — and digital object identifier 10.3389/fcimb.2025.1678067, the paper's permanent address. That is what makes it persuasive. Now its index terms: liver cirrhosis, schistosomiasis, hepatic stellate cells. No neoplasm term anywhere. It is mice with schistosomiasis, with lower YAP in the liver's scar-making cells (Xuan 2025). No cancer, no patient, no trial. The error has a name, and it is not the chatbot's: citing a record nobody opened. The real root of the claim is a 2015 mouse study of liver cancer whose drug screen found that macrocyclic lactones inhibit YAP1 activation, and suggested combining them with an inhibitor of TGF-beta, the signal telling cells to stop dividing and lay down scar (Nishio 2015). Suggested. In mice.

The pattern was named before this wave. In 2023 Wolfgang Löscher, a pharmacologist, reviewed ivermectin as a candidate for epilepsy and concluded, for the reasons already given here, that it is not one. He closed hoping his analysis would "avert the unjustified hype associated with the recent use of ivermectin to control COVID-19 from recurring in neurological diseases such as epilepsy" (Löscher 2023). It recurred. It changed disease. And part two of this season left a thought to pick up here. Praziquantel cures liver fluke infection at a predicted 93.8 percent for the two-dose day (Qian 2022) — and a year later, of 52 infected people with periductal fibrosis, 34.6 percent still had it (Wangboon 2018). Removing a cause is not treating its consequence, and a drug that kills a worm is not thereby a drug that reaches a tumor. Of the twelve infectious agents established as causes of cancer — 2.3 million cases in 2024, about 12 percent of all cancer — only three are parasites, and all three are flukes; roundworms, which is what ivermectin treats, are not among them (Rumgay 2026).

Three refusals. This part is not telling anyone to take ivermectin, mebendazole or fenbendazole. It is not saying the laboratory work is fake, because it is not, and one of its results — keeping the resistance pump off alongside real chemotherapy rather than in place of it — deserves a trial. And it is not saying the question is closed: one trial is registered, and the decisive experiment has never been done. Nobody has measured ivermectin's concentration inside human tumor tissue. Every ratio here is computed against plasma — the conservative choice, and not the complete answer.

Sources

Every paper referenced on air, in the order it comes up. Links go to the publisher via DOI.