Seven parallel research threads, 2026-08-06. Dose-response trials, transgender men on masculinizing doses, PCOS, mechanism, pellets and guidelines, Mendelian randomization, and an adversarial contrarian pass. Every DOI verified against CrossRef. Existing ResearchLibrary sources harvested before any new search, and only notes carrying status: fulltext or status: pdf were treated as citable.

This is research for discussion with a clinician. It is not medical advice.

The short answer

No, not at any dose a woman is realistically prescribed, and the strong version of the claim collapses under scrutiny. But the question hides three separate exposures that almost every article on the subject blurs together, and the answer differs for each.

Collapsing those into one category called "supraphysiologic" hides a fivefold to tenfold difference in exposure. It is the single most common error in this literature.

What actually holds up

1. The best-designed trial in the right population found nothing.

Huang 2014 randomized 71 hysterectomized women, all on standardized estradiol, to placebo or one of four weekly testosterone doses for 24 weeks, with exposure measured by CDC-harmonized mass spectrometry. Achieved levels ran 19, 78, 102, 128, and 210 ng/dL. The top arm sat at roughly five to six times the female physiologic level.

Lean body mass rose 1.8 kg at the top dose, with a clean gradient of 0.6 kg per 100 ng/dL. Fat mass did not change at any dose. The trial found the effect it was powered for and found nothing on fat, which argues against explaining the null as insensitivity.

The limitation matters though: it measured total fat by DEXA, not visceral fat by CT or MRI. A null on total fat is fully compatible with a visceral gain offset by a subcutaneous loss. So Huang cannot be used to reassure anyone about the visceral compartment specifically, which is how it is most often cited.

2. The only randomized visceral measurement in women points the other way.

STEP-HI randomized older women recovering from hip fracture to testosterone or placebo for six months, all of them exercising. Visceral fat change: testosterone -44 g, placebo +45 g (p = 0.091). Visceral fat as a share of total adipose tissue improved significantly on testosterone (p = 0.003).

Secondary analysis, DEXA-derived, unusual population. Not enough to build a case on. But it is the only randomized visceral-compartment data that exists in women, and it does not support the proposition.

3. At male-range exposure over years, visceral and liver fat do rise.

This is the real signal, and it is worth stating precisely because it is where the worry originates.

• Year one: essentially nothing. The best-powered study (n = 162, DEXA) found visceral fat changed by +3 g, confidence interval -10 to +16. Meanwhile total body fat fell 2.8 kg, leg fat fell 16%, and gynoid fat fell 14%. The waistline masculinizes because the hips and thighs empty out, not because the belly fills.
• Year three and beyond: a real rise. MRI at three years found visceral fat area up 13 cm², a 47% relative increase. A six-year MRI cohort found visceral volume up about 70% and liver fat roughly doubled, while total adipose tissue stayed statistically unchanged.
• The modifier that reframes it: in the three-year data, the visceral increase was concentrated in the participants who gained weight. That shifts the claim from testosterone creates visceral fat to testosterone changes where gained weight goes. Those are different sentences with very different implications for someone whose weight is stable.

Transferability to a postmenopausal woman is low. The dose gap is roughly tenfold, the age gap is thirty years, the route differs (injected esters with large peaks versus steady transdermal), and the estrogen environment differs. Carry across the direction of travel if you like. Do not carry across a percentage.

4. A citation trap worth knowing. A 2026 meta-analysis of 52 studies reports that masculinizing therapy reduces fat mass. True, and it is about total fat. Total fat can fall while visceral fat rises, which is exactly what the six-year imaging shows. Citing that meta-analysis as proof visceral fat does not rise is the same misreading as citing Huang, in the opposite direction.

The causal direction problem

This is where the popular version of the claim falls apart.

In women, body fat drives testosterone more strongly than testosterone drives body fat, by roughly a factor of two and a half. Bidirectional Mendelian randomization in up to 425,097 people, same instruments, both directions:

The forward arrow flipped direction under sensitivity analysis with evidence of pleiotropy. The reverse arrow held under every test thrown at it.

The mechanism is documented at the level of liver cell transcription. Glucose and fructose suppress the liver's production of SHBG, the carrier protein, by way of hepatic fat synthesis. Visceral fat produces exactly that. So:

Visceral fat → liver fat and hyperinsulinemia → SHBG falls → measured free testosterone rises, with no change in androgen production at all.

And free testosterone is arithmetic, not a measurement. The free androgen index is literally total testosterone divided by SHBG. Bioavailable testosterone is calculated from the same two numbers. Hold androgen production perfectly constant, drop SHBG, and both "exposure" variables rise on their own.

Strip SHBG out of the genetic instrument and the testosterone effect largely disappears. The authors of the landmark paper say so themselves: the associations "might be driven by direct effects of SHBG." A second study reproduced the collapse.

There is no visceral specificity in the genetic data either. The one Mendelian randomization study using actual imaged fat volumes found testosterone associated with visceral fat at 0.113 and with abdominal subcutaneous fat at 0.114. Essentially identical. That is a general adiposity signal, not a redistribution into the visceral compartment.

And the dose range disqualifies it from answering the question anyway. Women in the top 5% of the testosterone genetic score differ from the bottom 5% by 0.28 nmol/L. Supraphysiologic dosing moves people by whole nmol/L. Genetics can establish the sign of a lifetime effect at a trivial dose. It cannot establish the magnitude at a pharmacologic one, and nothing licenses assuming the line stays straight across two orders of magnitude.

PCOS is a weaker natural experiment than its reputation

The comprehensive imaging meta-analysis found excess visceral fat at matched BMI overall (SMD 0.41, CI 0.23 to 0.59). But restricted to MRI and CT only, the effect went null: SMD 0.19, CI -0.04 to 0.41. The positive pooled result is carried by DEXA and ultrasound, and DEXA overestimates visceral fat, worse in people who have more of it. The gold-standard studies inside the analysis openly disagree with each other. The imaging evidence base has not been updated since 2019.

The cleanest causal evidence in the whole project points at insulin, not testosterone. Thiazolidinediones lower androgens and raise SHBG in PCOS while causing weight gain. If fat mass drove androgens through fat mass alone, that result would be impossible. Every route to lowering insulin lowers androgens, reliably, across pooled trials in hundreds of women.

But the androgen arm is not zero. Flutamide, a pure androgen-receptor blocker, reduced the visceral-to-subcutaneous ratio in a 12-month randomized CT trial (p = 0.033) and reduced android fat in a smaller double-blind trial. Small, measuring redistribution rather than absolute mass, but genuine experiments pointing the right way.

One methodological warning, because it circulates: GnRH agonist studies are not androgen experiments. They suppress ovarian estrogen and androgen together, and the visceral gain they produce is attributable to estrogen withdrawal. Anyone citing them as evidence that androgens protect against visceral fat has the confound backwards.

Mechanism: there is a real causal arm

The population numbers point backward, but testosterone does have a genuine direct effect on female fat tissue. Both things are true, and they are additive rather than contradictory.

The strongest evidence is an experiment rather than a correlation. In women with PCOS, blocking AKR1C3, the enzyme that makes testosterone inside fat tissue, reduced new fat synthesis in those cells. That puts androgen upstream of the lipid phenotype. Add a randomized trial in 36 postmenopausal women where testosterone measurably suppressed the fat-releasing machinery in tissue biopsies (hormone-sensitive lipase down, PDE-3B up), plus in vivo microdialysis showing suppressed lipolysis after acute androgen exposure, and the mechanism is real.

The sex paradox resolves cleanly. Testosterone has two arms:

• Sex-invariant: it blocks the formation of new fat cells, in both sexes, both depots.
• Sex-reversing: it mobilizes fat out of storage in men, and restrains mobilization in women (β2-adrenoceptor down, hormone-sensitive lipase down, PDE-3B up, new fat synthesis up).

In a male body the mobilizing arm dominates and visceral fat falls. In a female body that arm is absent or inverted, so only the storage-favoring effects remain. Same hormone, opposite result, no hand-waving required.

The counterintuitive part worth keeping: blocking new fat cell formation sounds protective and is not. When storage demand persists and new cells cannot be recruited, existing cells enlarge. Big fat cells are inflamed, insulin-resistant, and leak lipid to the liver and abdomen. That is the PCOS phenotype exactly, and it is a plausible route to worse visceral and liver fat rather than less fat.

One popular theory is dead. The idea that this is all about aromatization does not survive testing. Two prospective controlled studies gave postmenopausal women aromatase inhibitors and imaged them at 12 and 24 months. Visceral fat: no significant change either time (-5.00 cm² and -0.22 cm², both null). If removing aromatization were the engine, that is where it would have shown.

A number not to use: a figure of roughly "+18% visceral fat with aromatase inhibitors" circulates widely and could not be traced to any retrievable primary source. The two well-designed trials say the opposite.

Both the exposure and the outcome are measured badly

The exposure. A single female donor's serum with a reference-method true value of 15.5 ng/dL was sent to 142 clinical laboratories. Results came back at 7.1 to 39.8 ng/dL. Grouped by platform, the group means ran 12.7 to 26.8. Forty-four percent of results exceeded 20 ng/dL. Four widely used immunoassay platforms met none of the study's bias goals.

Same blood. 7.1 to 39.8.

And the error is not random with respect to the outcome. The free androgen index over-estimates free testosterone specifically when SHBG is low, and SHBG is low precisely in the women who have visceral fat. So the exposure measure is systematically inflated in exactly the group with the outcome. That is not noise. That is bias pointing in the direction of the hypothesis.

The outcome. DEXA cannot cleanly separate visceral from subcutaneous abdominal fat. Against MRI in women, mean bias was modest but the individual agreement was poor: 95% limits of -406 to +499 cm³, with error growing as visceral fat grows. That band is larger than any treatment effect under discussion. Waist circumference is worse, and testosterone increases abdominal lean tissue, which inflates a waist measurement without a gram of new fat.

A note on a number that has been circulating in your own material: the "R² = 0.16 below 37.6 ng/dL" figure attributed to the 142-lab study does not appear in that paper. Already corrected on the live article in July; flagging it again here so it does not creep back. The 7.1 to 39.8 spread is the real finding and it is the stronger one anyway.

Effect size reality check

The largest randomized testosterone effect on any fat compartment in women is not statistically distinguishable from zero, the point estimate that exists runs in the protective direction, and the measurement error of a single scan is several times larger than the effect. The GLP-1 effect is one to two orders of magnitude bigger.

After fifteen years, the observational literature has still not produced a number a clinician could put in front of a patient. "Independently associated" is not a dose, a duration, or a magnitude.

Pellets: the real problem, and it is not visceral fat

Pellets do not occasionally overshoot. Overshooting is their pharmacology.

The largest real-world comparison (384 women on pellets versus 155 on approved therapy) found mean peak total testosterone of 194 ng/dL on pellets versus 15.6 ng/dL on approved products, with a pellet range running to 599 ng/dL. Nine women were solidly in male range. Side effects overall: 57.6% versus 14.8%. Abnormal uterine bleeding: 55.3% versus 15.2%. Hysterectomy: 20.3% versus 6.3%.

Individual pellet studies tell the same story: 100 mg implants producing 144 ng/dL sustained for eight months, 75 mg producing a mean of 250, weight-based dosing producing 299 ng/dL at four weeks and still 171 at reinsertion.

Every guideline says not to. The 2019 Global Consensus, endorsed by eleven societies, states that use of any preparation producing supraphysiologic concentrations, "including pellets and injections, is not recommended," and that compounded bioidentical testosterone "cannot be recommended." Nothing has been relaxed since. Two bodies tightened: a 2025 position statement naming pellets and compounded products as not recommended due to supraphysiologic dosing risk, and the 2025/2026 international recommendations stating that available data do not support prescribing testosterone for women for any indication other than postmenopausal low desire.

The harms that actually scale with dose are androgenic and hematologic, not adipose. The largest voice series to date documents 34 women with dysphonia and vocal virilization from testosterone therapy across all formulations; of 19 who completed voice therapy, only 6 reached all goals. Set that against a pellet registry's self-reported "1% transient hoarseness" and the ascertainment problem is obvious. Voice change can be permanent.

And the structural point nobody prices in: a pellet cannot be removed quickly once implanted. If virilization starts, there is no equivalent of taking off a patch.

One study to read carefully: the paper pellet clinics cite for safety measured trough levels, which structurally cannot detect the four-week peak that is the entire problem. Its senior authors carry industry consultancies.

The hole in the evidence, which is the story worth writing

A structured search for testosterone or androgen, in postmenopausal or cisgender women, with visceral fat as an outcome, combined with implant, pellet, supraphysiologic, or high dose, returns zero studies.

Not a thin literature. An empty one.

Tens of thousands of American women are being dosed into the 100 to 300 ng/dL band by pellet clinics. Nobody has ever put one of them in a scanner and looked. Meanwhile the argument rages on both sides using evidence borrowed from 25-year-old transgender men at five times the dose and from women with PCOS whose hormone levels are a readout of their insulin.

That gap is defensible, true, and nobody else is writing it.

What this means for you specifically

The question is close to moot in magnitude while you are on tirzepatide. A 40% reduction in visceral fat is not in the same universe as a hypothetical 90 g, non-significant, direction-uncertain testosterone effect. Whatever testosterone is or is not doing to that compartment is swamped by an order of magnitude or more.

Two things that are not moot:

GLP-1 receptor agonists appear to lower testosterone. So your level may drift below what your dose implies. That argues for measuring rather than assuming, and it points the opposite direction from the worry that started this.

You are the woman the safety trials excluded. The Global Consensus states plainly that its safety conclusions do not generalize to a higher cardiometabolic risk population, because the trials excluded those women, and that safety data do not exist beyond 24 months. With high ApoB and APOE4 homozygosity, that is you. The reassuring lipid finding, that non-oral testosterone at physiologic dose is lipid-neutral, is doing real work in your favor. The levers you control are staying non-oral and staying in range. On a pellet or a compounded injection, that protection is gone.

Also worth separating in your own writing: nothing here supports testosterone for cognitive protection. If it is helping the brain fog, that is symptom management, not neuroprotection, and conflating the two would be exactly the kind of overclaim your citation-checking habits exist to catch.

Questions for your clinician, in priority order:

1. Does the lab run LC-MS/MS for total testosterone, or an immunoassay? Get the lab's own premenopausal upper limit in writing, because there is no universal range across methods.
2. Ask for total testosterone and SHBG as separate numbers, not a calculated free T or free androgen index. If free T is high, those two tell you which of the two stories you are in, and they have opposite implications. A calculated free T genuinely cannot distinguish them.
3. Total testosterone every 4 to 6 months once stable, and within 6 weeks of any dose change.
4. Baseline and interval fasting lipids and liver function. These matter more for you than for a woman with normal ApoB.
5. What is the plan if a level comes back supraphysiologic with no symptoms? Guidance says dose down regardless. Not every prescriber does.

The asymmetry worth sitting with. Underdosing out of fear costs lean mass and stair-climbing power, both documented in randomized trials with real numbers. The visceral risk being avoided is, on the best randomized evidence, smaller than the error bars of the instrument used to look for it. Choosing a well-measured loss to avoid a poorly-measured harm is a category error, not caution.

One honest counterweight, though, because it cuts against the wellness-industry reading too: the lean-mass and power gains in the dose-response trial lived at supraphysiologic exposure. The authors concluded it is unlikely that meaningful gains could be achieved by raising testosterone from low into the normal female range. Anyone selling physiologic-dose testosterone as a body-composition intervention is over-promising.

Confidence, stated honestly

The last row is the important one. Absence of evidence there is not evidence of absence, and it should never be presented as either.

The study that would settle it

Randomized, placebo-controlled, minimum 12 months (24 weeks is too short for a compartment that shifts at 3.8% per year). Postmenopausal women not on a GLP-1, stratified by baseline SHBG, since that is the only way to break the reverse-causation confound inside a trial. At least three active arms spanning physiologic, the pellet band, and clearly supraphysiologic. Total testosterone by mass spectrometry and free testosterone by standardized equilibrium dialysis, never calculated, never as a free androgen index. Visceral fat by MRI or CT, blinded central reading, co-primary with lean mass. Powered on a pre-specified minimal clinically important difference stated up front, so a null is interpretable.

It has never been done. There is no approved testosterone product for women anywhere outside Australia, so no sponsor has an incentive to fund it, which keeps the evidence base thin, which justifies continued caution. That loop is worth naming when you write this up.

Library gaps found along the way

Three of the highest-value sources for this question are not in the ResearchLibrary at all: the 539-woman pellet safety comparison, the 36-RCT safety and efficacy meta-analysis, and the approved-cream pharmacokinetics study. Also missing: the 162-person one-year visceral cohort, the six-year MRI cohort, the randomized MRI study, and the sex-hormone-adiposity Mendelian randomization paper. Worth adding, with PDFs where obtainable, since they are load-bearing for anything published on this topic.

Two housekeeping notes: several library notes carry stale boilerplate claiming they are DOI pointers when their frontmatter correctly says full text is present. And one note worth upgrading from pointer to full text holds a claim that direct immunoassay reads more than 100% higher than mass spectrometry in the same women, which would be a strong addition to the assay argument if it checks out.