Venous Thromboembolism, Myocardial Infarction and Ischaemic Stroke in Feminising Hormone Therapy — Evidence Review (ER-018)

Evidence Review · ER-018

1. In brief

The findings, in one screen. Written so a clinician who reads nothing else is not misled.

3. What this review is, and is not

This review holds a single even-handedness commitment: to treat “the evidence is too weak to show benefit” and “the evidence is too weak to show harm” as the same kind of claim, judged by the same rules. A reassuring finding is held to the same scrutiny as an alarming one; over-claiming safety is the same error as over-claiming risk. Where the evidence is thin, this review grades it thin in both directions.

It is a narrative review, not a systematic review (see Limitations and the methods statement). It does not claim to have captured every relevant study, and its certainty marks are editorial judgements, not formal GRADE ratings.

4. Scope

In scope: venous thromboembolism (deep vein thrombosis and pulmonary embolism), myocardial infarction, and ischaemic (cerebrovascular) stroke, in the setting of feminising (oestrogen-based) gender-affirming hormone therapy. The review is organised on three axes — molecule (ethinylestradiol vs conjugated equine oestrogen vs 17β-oestradiol), route (oral, transdermal, injectable, transmucosal), and population (trans-direct evidence vs cisgender extrapolation).

Out of scope, and why: heart failure, arrhythmia and all-cause or cardiovascular mortality (not graded here — the outcome-specific evidence is treated separately from the composite “cardiovascular disease” endpoints that dominate parts of the literature); bone health, fertility, breast effects and hair (each the subject of a separate planned Evidence Review); and masculinising testosterone therapy, whose thrombotic and erythrocytosis profile is a distinct topic signposted to its own review (ER-023). Where a cohort reports a composite “any cardiovascular disease” outcome, that composite is not treated as a venous-thromboembolism estimate.

5. Limitations

Stated before any finding, so the reader weighs the conclusions accordingly:

• This is a narrative review, not a systematic one: no registered protocol, no exhaustive search, no dual screening, no uniform risk-of-bias instrument. It makes no claim of completeness. (Full method in the companion methods statement.)
• The certainty marks are editorial, using a declared four-axis scheme (causal attribution · observed effect size · clinical importance · population directness) as an exception to the corpus’s house three-axis standard. These are editorial certainty judgements, not formal GRADE ratings.
• Much of the route, molecule and post-VTE-management evidence is extrapolated from cisgender populations and is flagged as such at every use; there is no trans-specific outcome trial isolating route or molecule.
• Some primary sources remain unverified at this draft and are bracketed in text: the exact EMA cyproterone wording; and the perioperative (Kozato), thrombophilia-screening (Ott) and earlier serum-level (Nolan & Cheung 2021) sources, none load-bearing. The historical “45-fold” figure is quarantined pending an unobtainable 1989 primary and is not used in the graded body.
• No cohort has an internal untreated trans comparator, so the separation of treatment effect from era, population and co-medication is inferential throughout.

6. Cohort-independence note

Say this before the reader counts studies. The contemporary venous-thromboembolism evidence rests on two genuinely independent cohort lineages — Kaiser Permanente (US, the STRONG cohort / Getahun) and Amsterdam (Netherlands, the ACOG cohort / van Zijverden). Numerical agreement between these two is real corroboration; the near-twofold estimate is supported by both.

Agreement within a lineage is not. The Amsterdam figures across decades (van Zijverden, which supersedes Nota, and the historical van Kesteren and Asscheman) are one institutional cohort re-analysed, not independent replications; and the widely cited prevalence meta-analysis (Totaro) pools Getahun, Nota and van Kesteren, so it can never serve as independent confirmation of any of them. The coagulation-mechanism evidence (Part A) rests substantially on one Amsterdam clinic’s lineage, with a single independent check (da Cruz). A third national cohort (Glintborg, Denmark) is genuinely independent but reports a broad composite cardiovascular outcome, not a clean venous-thromboembolism estimate, and is used accordingly.

The full mapping — who supersedes whom, who pools whom, and which figures are one dataset wearing several citations — is set out in the companion study-overlap matrix.

Declarations

7.1 Competing interests

The author is a trans woman and supports access to gender-affirming hormone therapy. This interest was declared to the external reviewer before drafting began. The review was written under adversarial review with symmetric bias-checking as its explicit standard, and it reaches several conclusions the author did not want — among them that the venous thromboembolism association is real and not to be explained away, and that the lower observed incidence of myocardial infarction is not demonstrated cardioprotection and must not be presented as a benefit of therapy.

Rationale. The stake is not concealable and was never concealed. Declaring it before the findings is what allows the findings to be read as findings. A reader who knows the bias and can still check every number is in a stronger position than one who has to guess. The mental-health benefit — the finding the author would most prefer to be true — is graded in its own front-matter declaration at least as conservatively as the harms, precisely because this interest bears on it.

7.2 Correction record

This record is not an apology; it is the credential. For an author without institutional affiliation, a visible record of self-correction is the available evidence of method. Each item states what was wrong, where the error came from, who caught it, and what changed.

The standing pattern

Across production, the errors were overwhelmingly in transmission and drafting — secondary-source characterisations read in place of primaries, an ecological finding converted to an individual one, a vivid number reached from memory, an interpretation adopted rather than attributed — and almost never in the primary sources themselves. Sections built directly on primaries read in full held up under adversarial review; the failures clustered where the text summarised, inferred, or reached. This is the corpus’s standing thesis, and this review demonstrates it on its own body.

Substantive corrections

1 · Omission of the contemporary Amsterdam analysis (van Zijverden 2026).
Wrong: Parts B and C were built on Nota 2019 without integrating the superseding 2026 registry analysis. Source: the paper was in the archive and flagged in working notes but left unread — a gap in the author’s own process, not the literature’s. Caught by: external reviewer, round 1. Changed: van Zijverden read in full and made the spine of the venous and arterial estimates; VTE magnitude recalibrated to ≈1.8–2.0×; the arterial section rebuilt (MI SIR 0.50); the unsupported “male-substrate” construct replaced with the study’s own framing.

2 · The “ethinylestradiol-free window” factual error.
Wrong: the text claimed van Zijverden’s 2012–2022 event window meant the cohort was EE-unexposed, and that gate G1 was therefore “discharged twice.” Source: an inference not supported by the paper — the cohort initiated therapy from 1972 and the authors performed no past-EE sensitivity analysis. Caught by: external reviewer, round 2; confirmed against the primary’s own limitation text. Changed: corrected to “not an EE-naïve cohort”; G1 downgraded to “partly addressed.”

3 · Arterial overclaim (the body contradicting its own disclaimer).
Wrong: Part C stated feminising therapy “may lower MI risk” while a provenance note simultaneously disclaimed any causal reading. Source: the author’s principal overclaim migrated from venous to arterial while the venous causal language was being disciplined — the symmetric-bias standard failing on the reassuring side. Caught by: external reviewer, round 3. Changed: full de-causalisation — MI 0.50 reported as observed comparator incidence, the study authors’ interpretation reported as a hypothesis rather than adopted, the grade table reconciled with its disclaimer.

4 · Ecological fallacy in the concentration and stratification sections.
Wrong: Totaro’s study-level age and duration subgroups (< 37.5 years, < 53 months) were converted into individual-level risk statements (“negligible in young patients”). Source: an aggregate meta-regression read as individual-participant data. Caught by: external reviewer, round 1. Changed: reframed as aggregate associations with no individual threshold, in B1 and D.1; the data-derived cut-offs explicitly denied clinical-rule status.

5 · Internal contradiction on the ethinylestradiol mechanism.
Wrong: A.3 stated molecule and dose could not be separated; B2b later called the EE effect “molecular.” Source: drafting inconsistency across tranches. Caught by: external reviewer, round 1. Changed: B2b corrected to “high-dose EE, molecule and dose not separable.”

6 · An unsupported sublingual peak figure (≈ 800 pg/mL).
Wrong: a specific numerical peak was introduced in a wording pass, unverified. Source: a figure reached from memory rather than a primary — precisely the failure the citation-verification rule exists to prevent, committed by the author while running that discipline on others. Caught by: external reviewer, round 4. Changed: the number removed entirely — and not replaced with the reviewer’s alternative figures, which were themselves unverifiable against any held primary; a qualitative statement and a source-gate flag stand in its place.

7 · Exact percentages from an unobtained source (Seal 2012).
Wrong: exact subgroup VTE percentages were retained from an abstract-level source not obtained in full. Source: the same failure mode as item 6 — precise figures resting on an unread source. Caught by: external reviewer, round 5, and recognised by the author as identical in kind to item 6. Changed: the percentages withheld pending the primary; the qualitative ranking retained.

8 · Incorrect hazard-ratio definition.
Wrong: the estimand glossary — built to prevent estimand confusion — defined a hazard ratio as a relative event rate over follow-up, which describes an incidence-rate ratio. Source: author error. Caught by: external reviewer, round 4. Changed: corrected to the instantaneous-hazard definition with the proportional-hazards caveat.

9 · Mental-health benefit overclaim.
Wrong: the text stated therapy “substantially reduces depression and suicidality,” conflating ideation, attempts and mortality and overstating the suicide-specific evidence. Source: a benefit-direction overclaim — the symmetric-bias standard failing toward the author’s own position. Caught by: external reviewer, round 1. Changed: narrowed to dysphoria, well-being and depressive symptoms with an observational-evidence qualifier; the claim is being relocated to a graded front-matter declaration held to the same standard as the harms.

10 · Asymmetric ascertainment scrutiny.
Wrong: detection-bias scrutiny was applied to the reassuring arterial finding but not to the alarming venous one. Source: the author’s own skepticism running harder on the reassuring side than the alarming side. Caught by: external reviewer, round 3. Changed: a symmetric ascertainment caveat added to the venous signal; subsequently (round 4) both were disaggregated into distinct mechanisms — diagnostic ascertainment, preventive care, registry misclassification — rather than lumped as “bias.”

11 · The “natal-male substrate” construct.
Wrong: arterial risk was explained via a “natal-male substrate baseline” — an underdefined construct that implied biological determinism. Source: the author’s compressed-metaphor style tipping into an unsupported explanatory claim. Caught by: both reviewers, round 1. Changed: replaced with the comparator description and an explicit statement that the pattern does not identify its mechanism.

12 · Cyproterone regional-availability error.
Wrong: cyproterone was described as unavailable across “North America.” Source: collapsing US non-approval into a blanket regional claim. Caught by: the author, prompted by a patient’s report of Canadian prescription. Changed: corrected to jurisdiction-specific — approved in Canada and Europe (off-label for feminising therapy), never approved in the US — verified against the Health Canada drug register.

13 · Two unvalidated mechanistic clauses.
Wrong: spironolactone “haemoconcentration” (not an established VTE mechanism) and oestrogen “RAAS-mediated blood-pressure rises” — the latter substituting the author’s own unvalidated narrative for the study’s over-favourable one. Source: mechanism asserted in place of evidence. Caught by: external reviewer, round 4. Changed: both removed or softened to route/dose-dependent.

Corrections logged during earlier production (pre-review)

14 · The “45-fold” figure — analysis revised on further reading. Initially hypothesised to be a probable count-to-fold conflation of van Kesteren’s 45 cases with a fold-change. That hypothesis is now weakened: two independent 2024 secondary reviews (ASH; Nolan & Cheung) both attribute “45-fold (occurring in 6.3%)” specifically to Asscheman 1989 for a 100 µg ethinylestradiol + 100 mg cyproterone regimen, which suggests the figure is a genuine (if extreme, obsolete-regimen) statistic from that primary rather than a transcription artefact. It remains unverifiable — Asscheman 1989 is not obtained — and cannot be reconciled with the same Amsterdam cohort’s later ~20-fold (van Kesteren) without the primary. Current handling: quarantined pending verification of Asscheman 1989, attributed not asserted, and not used in the graded body; the evidence-transmission appendix documents the attribution chain and the revised (non-conflation) reading. This entry is itself a correction of an earlier correction — recorded as such, because revising a hypothesis on new evidence is the method, not a lapse. 15 · APC resistance downgraded from “validated predictor” to “associated phenotype.” 16 · Cunningham comorbidity odds ratios given their confidence intervals and hypothesis-generating framing. 17 · The “twentyfold” van Kesteren figure marked attributed-not-asserted (primary not obtained).

7.3 Verification statement

Read in primary (full text obtained and checked): van Zijverden 2026 · Nota 2019 · Getahun 2018 · Totaro 2021 · Cunningham 2026 · da Cruz 2025 · Toorians 2003 · Schutte 2022 · Canonico/ESTHER 2007 · Vinogradova 2019 (with its published correction) · Kuijpers 2021 · WPATH SOC-8 2022 (Chapter 12, Hormone Therapy — read from the full document; Statements 12.13–12.15 verified and quoted at grade) · Pyra 2020 (concentration–VTE null verified; progestin/MPA signal noted) · Martinelli 2016 (post-VTE continuation, cisgender — verified) · Glintborg 2022 (independent Danish composite-CVD cohort — verified as not a co-equal VTE estimator) · EMA cyproterone restriction (EMA/70255/2020, read from the primary EMA document) · ASH 2024 and Nolan & Cheung 2024 (secondary reviews, read in full but used for framing only, never as primaries for grading). [17 primaries + 2 secondary]

Read in published abstract only (full text not obtained; flagged at every use): Giltay 2000 · Seal 2012 · Manchkanti 2023 · Rothman 2024. [4]

Read in primary (current version) — figure verified: Endocrine Society 2017 guideline (10.1210/jc.2017-01658, Hembree et al.) — the author’s archived copy is the correct published version (the connector’s copy was flagged superseded; the author’s read overrides that). Table 15 verifies the serum-estradiol monitoring range (100–200 pg/mL, testosterone <50 ng/dL), framed as a feminisation/suppression target, not a VTE-safety ceiling. B3 now states the figure as verified rather than attributed.

Not obtained (cited via secondary source or via the pooling meta-analysis; each flagged, none load-bearing without verification): van Kesteren 1997 · Asscheman 1989 · Kozato 2021 (perioperative — cited via ASH/Nolan & Cheung, primary still not obtained; note the N=402-vs-407 discrepancy between ASH’s body text and table) · Ott 2010 · Nolan & Cheung 2021 (the earlier serum-level narrative review; the 2024 general review was obtained instead). [5 not obtained]

Claim-to-reference discipline: every numerical claim in the graded body is traced to a source read in primary, or the limitation is stated in text. Figures resting on unobtained sources have been either removed (items 6, 7) or explicitly bracketed. A full claim-to-reference map — verifying that each citation resolves to a source that supports the claim, not merely that the citation number resolves — remains to be built against the finalised reference list, cross-referenced with the study-overlap matrix so that no figure re-cited across a single lineage is counted as independent corroboration.

AI-preparation note (cross-reference §7.4): this review was prepared with a large language model used for source retrieval, arithmetic and structural editing, and as adversarial reviewer. Several errors above were introduced during AI-assisted drafting (notably items 6 and 8); they were caught by external review and by the author, which is why this log exists. Every claim, number and citation is the author’s responsibility.

Author verification pass (§7.4) — complete. Every figure in the graded body was checked against its primary source by the author using the source-grouped worksheet; all figures matched the primaries as stated. The preparation statement (§7.4) is therefore true as written: every claim, number and citation was verified against the primary source by the author.

7.4 Preparation and authorship statement

Preparation. This review was prepared with the assistance of a large language model (Claude, Anthropic), used for source retrieval, arithmetic verification, structural editing and adversarial review. Every claim, every number and every citation was verified against the primary source by the author, and the author is responsible for all of them. Several errors introduced during AI-assisted drafting are recorded in the correction log; they were identified by the author and by an external reviewer, and this is why the correction log exists.

Rationale, recorded because the alternative was considered and rejected. Omitting this statement does not remove an attack vector — it creates a better one. Silence invites the question “why didn’t she say so?”, which is the story that destroys a corpus. This corpus rests on disclosing what it cannot be trusted to hide: the competing interest, the correction log, the population flags, the “read in abstract; full text not obtained.” A standard that says declare your bias, log your errors, flag your extrapolations — and omit this one thing because it is inconvenient is a standard with a hole in it, and the hole is exactly where the author had something to lose. The statement also does necessary work: it explains why the correction log exists, converting the recorded errors from evidence of carelessness into evidence of process.

This review is a self-published Evidence Review deposited to Zenodo, not a peer-reviewed journal submission; the full-preparation disclosure above therefore applies. For peer-reviewed submission the standing commitment is different — the author writes in her own voice and the AI acts as adversarial reviewer only — and that distinction is declared accurately per venue.

7.5 Correspondence record

Not applicable to this review. No discrepancies in third-party documents were formally reported to any body in the course of this review. Certain observations about secondary sources — a significant finding omitted from one review’s characterisation of a primary (the progestin/VTE signal in Pyra 2020), a sample-size inconsistency within another (Kozato reported as N=402 in body text and N=407 in table of the same review), and the propagation of an unverifiable “45-fold” figure across multiple reviews citing a single unobtained 1989 primary — are recorded in the evidence-transmission appendix as part of this review’s own transparency, not reported to the authors or publishers as formal corrections. Should the author elect to raise any of these with a source’s authors or a correction channel, that would be a separate, deliberate act recorded here at that time.

Mental-health benefit — graded declaration

The benefit that counterweights the venous and arterial risks is graded here as a declaration rather than stated in the body, for two reasons: mental-health outcomes lie outside this review’s VTE / MI / ischaemic-stroke scope and are not its primary subject; and the benefit is load-bearing in any risk–benefit reading, so it is held to the same four-axis evidential standard as the harms rather than asserted narratively.

The author’s competing interest — declared in §7.1: a trans woman who supports access to gender-affirming care — bears directly on this claim. This is the finding the author would most prefer to be true. It is therefore graded at least as conservatively as the harms, and not more generously — which is the entire purpose of the symmetric-bias standard, tested here most directly.

The claim, stated plainly

Feminising gender-affirming hormone therapy is desired and clinically important to many of the people who seek it, and predominantly observational studies associate it with improvements in gender dysphoria, psychological well-being and depressive symptoms. Evidence bearing on suicide attempts and on mortality is weaker, less consistent, and more heavily confounded, and must not be collapsed with symptom-scale measures into a single figure or a single claim.

Four-axis grade — the same scheme applied to the harms

Scope

A full appraisal of the mental-health and psychosocial outcomes of gender-affirming therapy — with its own source base, its own grading, and the adolescent-versus-adult distinctions those outcomes turn on — is a separate review, not this one. This declaration grades only the counterweight needed to read ER-018’s risk findings in context; it does not stand as this platform’s full position on the benefit evidence.

What this declaration does, and does not, license

It does not convert a risk–benefit judgement into a foregone conclusion. A real, clinically important, population-direct benefit with limited-to-moderate causal attribution is precisely the kind of finding that shared decision-making exists to weigh — patient by patient — against a real and modest venous risk. It does not license withholding therapy on thrombotic grounds without that weighing; nor does it license disregarding thrombotic risk because the benefit is valued.

Stated once, plainly, and symmetrically: the harms in this review are graded conservatively, and so is this benefit. Neither is inflated to serve the reading the author would prefer. Withholding or interrupting desired therapy carries its own psychological and, for some patients, physical costs, and those costs belong in the same decision as the thrombotic risk — neither privileged by where it happens to sit in this document.

Methods statement

This states how the review was produced, so a reader can judge its coverage.

This is a narrative review, not a systematic review

ER-018 is a narrative evidence review. It did not follow a registered systematic-review protocol: there was no exhaustive database search across pre-specified terms, no PRISMA flow, no independent dual screening, and no formal risk-of-bias instrument applied uniformly to every study. It therefore makes no claim to have captured all relevant evidence, and its conclusions should be read as a graded appraisal of the principal available sources, not as a complete synthesis. This disclaimer is load-bearing and travels with the deposit; it is not a hedge but an accurate description of method.

The choice is deliberate. The review’s value is in appraising the certainty of a contested literature — separating causal attribution from effect size, direct evidence from cisgender extrapolation, and independent cohorts from re-cited ones — rather than in tallying every publication. Where that appraisal depends on a source, the source is named, its status disclosed, and its independence recorded.

Source discovery

Sources were identified from: the platform’s held source archive; targeted retrieval of primary studies where a claim required them; the reference lists of the principal cohort and review papers (notably the contemporary Amsterdam analysis and the ASH education review); and references surfaced by external adversarial review. The emphasis was on locating the primary outcome cohorts in feminising hormone therapy (venous and arterial), the mechanistic studies in trans populations, and the cisgender route/molecule evidence that the trans literature leans on. Discovery was purposive, not exhaustive.

Search boundary

The review reflects evidence available to the production period ending July 2026, anchored by the most recent contemporary cohort analysis obtained (van Zijverden 2026). Fast-moving external facts — guideline status, litigation, any newer cohort — are to be re-checked on the morning of deposit, per the pre-deposit checklist.

Study selection and appraisal

Every load-bearing numerical claim in the graded body was traced to a source and the source’s status recorded in one of four tiers: read in primary (full text obtained and checked), read in abstract only (flagged at every use), held but retrieval-limited, or not obtained (cited via a secondary source or the pooling meta-analysis, and flagged as an open gate). The verification statement lists every source by tier.

Two rules governed appraisal:

• Secondary sources were never used as primaries for grading. Review articles (e.g. the ASH 2024 education review) were used for framing and to locate primaries, and any claim resting on them was traced — or flagged for tracing — to the underlying primary. Several figures reached ER-018 only via secondary sources; these are recorded as open gates, not asserted.
• Overlapping cohorts were identified before figures were compared. A study-overlap matrix maps which cohorts are independent, which share an institutional lineage, and which meta-analysis pools the others, so that numerical agreement within a single lineage is never counted as independent corroboration.

Certainty grading

ER-018 uses a four-axis editorial certainty scheme — causal attribution, observed effect size, clinical importance, population directness — as a declared exception to the platform’s house three-axis scheme, because this topic turns on distinctions the field routinely collapses. The scheme is editorial and explicitly not formal GRADE; a written decision rule anchors each axis, and the estimand is named beside every effect-size grade. The rationale and the rule are stated in the certainty-scheme note.

Adversarial review and bias handling

The review was produced under iterative external adversarial review, with a symmetric bias standard as its explicit discipline: over-claiming benefit is treated as the same error as over-claiming harm, and reassuring findings are held to the same scrutiny as alarming ones. The author’s competing interest is declared before the findings (declarations block). The correction record documents the substantive errors identified during production and their sources; it is retained as evidence of method, not removed.

Preparation

The review was prepared with the assistance of a large language model used for source retrieval, arithmetic verification, structural editing and adversarial review. Every claim, number and citation is the author’s responsibility; errors introduced during AI-assisted drafting are recorded in the correction log. The full preparation and authorship statement is in the declarations block (§7.4).

What this method does not deliver

A reader should not infer from any conclusion here that the underlying evidence base is complete, systematically searched, or free of publication and ascertainment bias. Where the evidence is thin, the review grades it thin; where a source could not be obtained, it says so; where a figure rests on one re-cited dataset, the overlap matrix shows it. The method’s honesty is in its disclosure, not in a completeness it does not claim.

Part A — Coagulation mechanism

Certainty at a glance (four-axis — see the scheme note in the assembled draft)

Editorial certainty judgements, not formal GRADE ratings.

A.1 — The premise: first pass

Oral oestrogen reaches the liver through the portal circulation before systemic distribution, producing greater first-pass hepatic exposure than non-oral administration. This alters the synthesis and regulation of multiple proteins across the coagulant, anticoagulant and fibrinolytic systems — some rising, some falling — with the pattern and magnitude varying by oestrogen, dose and regimen rather than moving in one uniform direction. Transdermal and injectable routes avoid that gastrointestinal first pass; transmucosal (sublingual or buccal) dosing can partially bypass it through absorption across the oral mucosa, though a swallowed fraction may still undergo gastrointestinal and hepatic first-pass metabolism. In each case the liver still meets circulating oestradiol and its metabolites, but not the concentrated portal bolus — and sublingual dosing produces sharp early peaks that differ from steady-state transdermal or depot delivery, a pharmacokinetic point Part B3 takes up. The route question, stripped to its mechanism, is therefore less about skin versus tablet than about how much oestrogen the liver is made to process — and, as the rest of Part A argues, which oestrogen and at what dose, on evidence unevenly distributed and weighted to one centre.

A.2 — What actually moves: the haemostatic shift, measured in trans people

The shift can be measured directly in this population, on the particular markers each study chose.

Toorians and colleagues measured acquired activated-protein-C resistance — a haemostatic phenotype associated with venous thrombosis in other populations, though not validated as an individual predictor in transfeminine care — with the natural anticoagulants, across four regimens (Toorians 2003; n = 14/14/8/20; VU University Medical Center, Amsterdam; four-month within-person design; markers, not events). On oral ethinylestradiol with cyproterone the normalised APC-sensitivity ratio rose from 1.2 to 4.1 — a resistance that, as a laboratory value, falls in the range the assay records for factor V Leiden carriers, though the comparison is one of laboratory magnitude only and implies nothing about equivalent absolute risk, biology or persistence. Total and free protein S fell about 30%; protein C rose 9%. Cyproterone alone was associated with a small increase in this APC-resistance assay (1.4→1.8) — a measurable haemostatic effect in this sample, whose clinical VTE significance is unknown, and which Part D takes up. That no thrombosis occurred is a statement about size and duration — four months across small arms has insufficient power to detect an event of this rarity — not about safety.

da Cruz and colleagues measured a different panel (da Cruz 2025; n = 40, median 36.5 months, 65% oral oestradiol valerate; matched cisgender women and men; Porto Alegre — independent of Amsterdam; cross-sectional): PAI-1 higher and free protein S lower than in cisgender men, while antithrombin, protein C, prothrombin time, thrombin time, VCAM-1 and fibrinogen did not differ. These findings point in directions conventionally associated with reduced fibrinolysis and reduced anticoagulant activity respectively, while the remainder of the measured panel did not differ. SHBG tracked inversely with free protein S — the expected fingerprint of hepatic exposure, nothing more (SHBG being a marker of exposure, not a validated individual predictor).

A.3 — The ethinylestradiol comparison: large, but molecule and dose are entangled

Oral 17β-oestradiol, which shares the hepatic first pass with oral ethinylestradiol, produced an APC-resistance ratio no different from transdermal oestradiol or cyproterone alone, and no statistically detectable difference on protein C, protein S or prothrombin — while the ethinylestradiol arm moved sharply. High-dose ethinylestradiol thus produced substantially greater haemostatic change than the studied 17β-oestradiol regimens; but the design cannot isolate molecular structure from dose, because the ethinylestradiol was given at 100 µg with a stated dose–response, and molecule and dose are confounded here. The finding’s principal relevance is to historical exposure and to continuing unsupervised ethinylestradiol use — which da Cruz notes still occurs, often at high doses — not to contemporary prescribed oestradiol.

Two limits keep this honest. The route is not wholly inert even where APC resistance is unmoved: oral but not transdermal oestrogen alters the fibrinolytic axis, lowering tissue-plasminogen-activator (Giltay 2000; read in published abstract only, full text not obtained) — though that change is not cleanly prothrombotic and the fibrinolytic markers are inconsistent across the trans studies. And Part A characterises only ethinylestradiol against 17β-oestradiol; it does not characterise conjugated equine oestrogen, whose elevated risk is so far an epidemiological association (Part B) that cannot substitute for direct mechanistic evidence.

A.4 — The contemporary picture: heterogeneous, not “mild”

Studies of haemostatic markers during contemporary feminising therapy do not support a single characterisation such as “mild.” Findings depend on regimen, comparator, exposure duration and endpoints. Studies of coagulation and anticoagulant variables have generally found smaller changes with 17β-oestradiol than with high-dose ethinylestradiol; da Cruz found the parameters measured broadly similar to cisgender women. By contrast, Schutte found increased platelet-activation markers after twelve months of transdermal oestradiol plus cyproterone acetate (PF-4 +17%, β-thromboglobulin +13%), with a transient fibrinogen rise; the study could not separate oestradiol from cyproterone, and PAI-1 did not clearly change. Fibrinolysis findings are limited and inconsistent. These indicate a heterogeneous, endpoint-dependent response rather than a uniformly mild or uniformly procoagulant effect, and no single study measured its named domain comprehensively. da Cruz’s own register is the correct one: not safety, but “some reassurance,” beside a cross-sectional design, small sample, regimen heterogeneity, and no older participants or clinical endpoint.

A.5 — The hinge: heterogeneous mechanism against a larger clinical association

The mechanistic and epidemiological evidence are not directionally discordant in any simple sense. Some contemporary-regimen studies identify modest procoagulant changes — increased platelet-activation markers, changes in selected coagulation or anticoagulant proteins — while other domains appear minimally altered, transiently altered, or shifted toward cisgender-female values. What cannot be said is that the mechanism agrees with the epidemiology: some surrogate findings point in the same prothrombotic direction as the clinical signal, but the mechanistic literature is heterogeneous across domains, regimens, sampling times and designs, and does not speak with one voice.

The central uncertainty is not merely whether a procoagulant signal exists, but whether these heterogeneous surrogate changes are causally related to — and quantitatively sufficient to explain — the relative VTE associations reported in observational cohorts (Part B). That question cannot be answered by setting a 13–17% marker change beside an incidence ratio: the two are not commensurable, and the mismatch is not itself evidence of unexplained residual risk. Nor are these markers validated as mediators of clinical VTE in transfeminine care; they are associations with selected haemostatic variables, not demonstrated causes of a given share of events.

Historical ethinylestradiol exposure, dose and route, antiandrogen co-treatment, population differences and residual confounding all bear on any remaining gap, which Part B adjudicates. The mechanism is not the verdict; it is the set of constraints within which the epidemiology must be read.

Provenance and verification (Part A)

• Read in primary: Toorians 2003 (10.1210/jc.2003-030520); da Cruz 2025 (10.1371/journal.pone.0323606); Schutte 2022 (10.1371/journal.pone.0261312).
• Abstract-level, not obtained: Giltay 2000 (10.1161/01.atv.20.5.1396) — fibrinolytic nuance only, mixed direction stated.
• Centre lineage (G7): Toorians, Giltay, Schutte are the Amsterdam gender-clinic lineage (VUmc/Amsterdam UMC); limited external replication from a single centre, not demonstrated participant overlap. da Cruz (Porto Alegre) is the sole independent mechanism source and is cross-sectional.
• Not claimed: no individual SHBG or hepatic-responder stratification; no clinical-safety claim from markers; no integrated prothrombotic verdict from two markers; the Toorians null treated as underpowering.

Part B — Venous risk: magnitude, route and molecule

Scope of Part B

Part B grades the epidemiological evidence for venous thromboembolism in feminising therapy on the axis molecule × route × population. Five outcomes: (1) VTE overall; (2) route; (3) molecule; (4) dose/level; (5) injectable. Arterial events are Part C (G6); risk modifiers Part D; absolute-versus-relative framing within each outcome.

B1 — Venous thromboembolism, overall

Four-axis grade

Editorial certainty judgements, not formal GRADE ratings.

The absolute picture first (G4)

Lead with the matched, adjusted comparisons rather than a generic background rate. In the Kaiser STRONG cohort, adjusted VTE risk differences against matched cisgender controls were 4.1 per 1000 at two years and 16.7 per 1000 at eight years versus cisgender men (95% CI 1.6–6.7 and 6.4–27.5), and 3.4 and 13.7 versus cisgender women. Crude incidence was 5.5 per 1000 person-years in STRONG and approximately 2.4 per 1000 in the later Amsterdam registry analysis — but these two crude rates are not directly comparable without age and design standardisation (the populations, calendar periods, age structures, outcome capture and treatment histories all differ), and the lower Amsterdam rate is not in itself evidence of a smaller treatment effect. A frequently-quoted young-adult background near 1 per 10 000 per year is not age-matched to these cohorts and should not anchor the contrast. Two pooled/crude proportions circulate — a meta-analytic 2.0% and a contemporary single-centre 1.4% — but prevalence is not incidence, the 2% pools Getahun, Nota and van Kesteren rather than corroborating them, and the 1.4% is a crude proportion over variable follow-up its authors warn against reading as a rate.

The relative picture — numerical concordance, not a settled effect

A 2026 Amsterdam registry-linked analysis (van Zijverden), observing events from 2012 to 2022 and adjusting standardised incidence ratios for age, sex, calendar year and socioeconomic status, estimated a VTE SIR of 1.81 (1.33–2.35) versus general-population men and 1.74 (1.28–2.26) versus women. This is the most recent estimate from this cohort lineage and should take precedence for contemporary event ascertainment; the 2019 Amsterdam analysis (Nota, SIR 4.55/5.52) remains relevant as an earlier analysis using a Norwegian reference population, different outcome capture, and the ethinylestradiol era, not as void. Kaiser’s adjusted hazard ratios were 1.9 (1.4–2.7) versus men and 2.0 (1.4–2.8) versus women.

These strongest contemporary estimates are numerically similar — near 1.8–2.0 — but they are not the same quantity: an individually-adjusted hazard ratio from a matched insured US cohort, and an SES-adjusted standardised incidence ratio against national rates from a Dutch clinical cohort. Nor are they equally adjusted: Kaiser controlled at the individual level for smoking, BMI, blood pressure and cholesterol; van Zijverden adjusted its SIRs for age, sex, calendar year and SES but could not adjust for lifestyle, which it compared only descriptively in a small subset (174 trans women). So the earlier draft’s “twofold-versus-fivefold conflict” is retired — but what replaces it is approximate numerical concordance across different comparators, estimands and adjustment sets, which is informative and should not be presented as a settled pooled effect or a stable biological magnitude.

Two forces explain why older estimates ran higher. Ethinylestradiol (G1, partly addressed): Nota’s exclusion of pre-2001 starters dropped its VTE SIR from 5.52 to 3.92 (women) and 4.55 to 3.39 (men) — a treatment-era sensitivity analysis, imperfect but direct. van Zijverden’s 2012–2022 event window post-dates routine Dutch EE prescribing, so current EE exposure is unlikely; but the cohort initiated GAHT from 1972, van Zijverden performed no past-EE sensitivity analysis (its authors note almost all pre-2001 starters had used EE), and it is therefore not an EE-naïve cohort — prior EE exposure, treatment switching and survivor selection are not excluded. G1 is thus partly addressed by Nota’s era analysis and contextually by the later window, not “discharged.” Adjustment (different sets, above) explains part of the rest.

The honest summary: venous thromboembolism is consistently observed at higher incidence among cohorts receiving feminising hormone therapy than among several cisgender reference populations, numerically near twofold in the strongest contemporary analyses. How much of that association is caused by contemporary prescribed oestradiol — rather than era, formulation, co-medication or residual confounding — remains uncertain, and no cohort has an internal untreated trans comparator to settle it. And the same detection scrutiny applied to the arterial findings in Part C applies here, though the mechanisms must be kept distinct rather than lumped: differential healthcare contact could affect diagnostic ascertainment (more frequent contact may raise the chance a symptomatic event is investigated and coded), which is separate from preventive-care differences (which would change true incidence, not its detection) and from registry under-ascertainment (which could affect either group). Their direction and magnitude here are unknown, and the concern is weaker where exposed and reference outcomes come from the same national registry — as van Zijverden notes of its own design. The point retained is only the symmetric one: whatever ascertainment scrutiny is applied to the reassuring arterial finding must be applied to the venous signal too, not reserved for the finding that does not flatter the stance.

The excess appears concentrated — aggregate-level, not an individual threshold

The observed events cluster among older and more comorbid patients, consistent with general VTE epidemiology — but this is not the same as establishing how the treatment-associated relative or attributable risk varies across strata. In the prevalence meta-analysis, series with mean age under 37.5 years or mean therapy under 53 months reported few or no events (pooled point estimates of 0%, wide intervals, low power, no individual-participant data); 37.5 years and 53 months are data-derived splits, not clinical cut-offs, and do not establish that a patient under 37.5 has near-zero risk. A contemporary cohort’s sixteen VTE cases were older (median 50 vs 29) and more often carried hypertension (OR 10.7, 95% CI 3.40–33.30), hyperlipidaemia (6.9, 2.12–21.03) and diabetes (7.2, 1.64–24.85) — estimates from sixteen events, wide and hypothesis-generating, not validated predictors. Most cases had at least one reported concurrent or pre-existing risk factor, though the study’s categories should not be equated with the standard classification of a VTE as provoked by a major transient factor.

Both disciplines, opposite in direction. The clustering is useful for counselling — it makes “numerically twofold on average” more nuanced than a flat figure. But the reading that the excess is “largely driven by established risk factors rather than GAHT” points reassuring and is held at arm’s length: it rests on sparse events without a comparator, and some of those cardiometabolic conditions may change during long-term therapy, so the cohort cannot determine whether they preceded treatment, arose independently, or partly mediated risk. The clustering pattern is plausible; the partition of the incremental GAHT-associated risk across those strata is not established.

A time-course that appears not to front-load

Several analyses show events continuing to occur after many years of treatment: the Kaiser initiation subgroup’s hazard was near-null for two years then rose; therapy duration predicted prevalence in the meta-analysis; the contemporary US cohort’s mean latency was 7.8 years and the 2026 Amsterdam median was 13.3 years to VTE. This is consistent with a pattern less front-loaded than the acute cisgender one — but each strand is confounded with age and cumulative exposure, and latency conditioned on having an event is not an incidence-time analysis. It does not yet establish that the relative hazard function differs from menopausal or contraceptive oestrogen exposure; it cautions against reading acute cisgender first-pass data straight across, without asserting a distinct biology.

Feeding the A.5 magnitude question

Part A asked whether modest surrogate changes can account for the clinical association. Part B narrows it: the strongest contemporary estimate is numerically near twofold, older estimates were attenuated in a treatment-era sensitivity analysis and also differed in adjustment, comparator selection and event ascertainment, the events occur disproportionately in older comorbid samples, and the latency is long. The gap between marker and outcome is therefore smaller and more explicable than the alarming end of the literature implied — without vanishing, and without becoming a mediation estimate, since (per A.5) percentage marker changes and incidence ratios are not commensurable.

Historical anchor

Secondary accounts of the historical Amsterdam cohort document a substantially higher VTE burden during high-dose ethinylestradiol use. Exact comparator-specific estimates remain insufficiently verified here (the primary was not reviewed) and are addressed in the evidence-transmission appendix rather than the graded body, where the circulating “45-fold” figure is examined and set aside.

Provenance and verification (B1)

• Read in primary: Getahun 2018 (10.7326/M17-2785); Nota 2019 (10.1161/CIRCULATIONAHA.118.038584); van Zijverden 2026 (10.1093/eurheartj/ehaf837); Totaro 2021 (10.3389/fendo.2021.741866); Cunningham 2026 (10.3390/jcm15114166).
• Independence (G7): Getahun (Kaiser) is the independent outcomes anchor. van Zijverden takes precedence over Nota but is the same Amsterdam lineage; Totaro pools the cohorts and is never cited as independent corroboration.
• Read in primary — an independent but composite cohort: Glintborg 2022 (10.1530/EJE-22-0306) — a Danish national register cohort (2671 trans persons), genuinely independent of the Amsterdam and Kaiser lineages. Its primary outcome is a broad composite “any CVD” including medication prescriptions and milder outcomes, not a clean VTE analysis, so it is not treated as a co-equal VTE estimator. Its pure venous-diagnosis signal in trans women is a non-significant trend (incidence rate 2.31 vs 1.20 per 1000py vs control men, p=0.09; underpowered), consistent with but not corroborating the ~2× estimate. Two findings from it are used directly: gender-affirming hormone therapy did not statistically mediate the CVD excess in trans women (it did in trans men) — a caution against over-attributing the venous association to GAHT; and its authors explicitly flag surveillance bias from routine annual screening at gender clinics, which supports the ascertainment caveat above from an independent source that names it in its own cohort.
• Ascertainment: code- and registry-based throughout; imperfect, misclassification possible.
• Not claimed: no mediation estimate; no individual age/duration threshold; no route inference from case distributions; no EE-naïve characterisation of any cohort.

B2 — Route and molecule

B1 established that the excess is real, numerically near twofold when adjusted, disproportionately in older comorbid samples, and slow — and the route and molecule evidence is overwhelmingly cisgender, where the effect is the acute first-pass one the trans time-course does not clearly match.

B2a — Route: transdermal versus oral 17β-oestradiol

Four-axis grade

The recommendation for transdermal therapy rests primarily on extrapolation from menopausal studies together with hepatic first-pass biology, rather than on direct transfeminine outcome evidence. The cisgender association is directionally consistent but not “firm” in the sense of trial evidence. ESTHER (effectively a 17β-oestradiol study — only two CEE cases) gave adjusted VTE odds ratios of 4.2 (1.5–11.6) for oral and 0.9 (0.4–2.1) for transdermal; the QResearch/CPRD analysis gave 1.27 (1.16–1.39) oral and 0.96 (0.88–1.04) transdermal. Same direction, very different magnitude. Two caveats: a confidence interval crossing one does not establish equivalence, so transdermal is better described as showing no statistically detectable increase than as “neutral”; and transdermal is preferentially prescribed to higher-risk patients, so confounding by indication could bias either way.

For trans populations the outcome evidence is close to absent — no route subgroup in the meta-analysis, sparse uninterpreted strata in Kaiser, a contemporary cohort’s case route-split that is a numerator without denominators. What supports the lever is mechanism, carried forward honestly from Part A: Schutte found transdermal oestradiol plus cyproterone raised platelet-activation markers — so the regimen is not haemostatically inert, though platelet activation is not the hepatic first-pass pathway the route is meant to spare, and estradiol cannot be separated from cyproterone. The defensible position: transdermal is a reasonable preference for higher-risk patients, graded an inference from cisgender data and mechanism, not a trans-validated reduction in events.

B2b — Molecule: ethinylestradiol vs conjugated equine oestrogen vs 17β-oestradiol

Four-axis grade

Ethinylestradiol is strongly implicated in the larger historical associations — Part A found substantially greater haemostatic effects with the high-dose EE regimen than with the studied 17β-oestradiol regimens, although molecule and dose could not be separated — and it is deprecated, so its relevance is legacy risk and unsupervised use. Conjugated equine oestrogen hides inside “estradiol”: cisgender data put oestrogen-only CEE at 1.49 (1.39–1.60) against 1.27 for oral estradiol, and the one trans dataset that separated it (Seal) reported a higher proportion of VTE among CEE users than among those on estradiol valerate or ethinylestradiol — a small single-centre audit with VTE secondary and unstable subgroups, whose ranking of EE below CEE is a reason to lean on it lightly (its exact subgroup percentages are withheld here pending the primary). 17β-oestradiol and esters hydrolysed to estradiol (e.g. estradiol valerate — a prodrug, not chemically identical) are the contemporary standard and the low end. High-dose EE is of substantially greater concern than contemporary estradiol; CEE is reasonably avoided where estradiol is available, based predominantly on indirect menopausal evidence rather than a reliable trans-specific comparison — and the review does not present EE and CEE as equally established.

B2c — The progestogen and anti-androgen bridge to Part D

ESTHER found progestogen class mattered in menopausal therapy: norpregnane derivatives OR 3.9 (1.5–10.0), oral oestrogen with nortestosterone 6.7 (2.1–21.9), versus micronized progesterone 0.7 and pregnane derivatives 0.9. Cyproterone sits inside that non-significant pregnane class — but the class is lumped (it holds medroxyprogesterone and dydrogesterone), cisgender, menopausal, and at doses unlike antiandrogen dosing. ESTHER therefore shows that accompanying progestogen class affects VTE risk in menopausal HT; it does not provide a cyproterone-specific estimate applicable to transfeminine care.

A trans-direct signal now complements that cisgender class data. Pyra 2020 (US cohort, N=2509 trans women, read in primary) found no association between venous thromboembolism and estradiol or its route, but recent progestin prescription was associated with roughly threefold odds of thromboembolism (adjusted OR 2.95, 95% CI 1.02–8.57), with a much larger but far more imprecise estimate for medroxyprogesterone acetate specifically. The finding is fragile and must be held at arm’s length: five events among progestin users, wide intervals, an observational cohort where the progestin was MPA and the anti-androgen spironolactone (not cyproterone), and possible confounding by indication. It yields nothing cyproterone-specific. But it is trans-direct evidence, pointing the same direction as ESTHER’s class effect, that an accompanying progestogen should not be assumed haemostatically inert — and it is absent from the secondary review that characterised Pyra, surfacing only in the primary.

Cyproterone is therefore best described as a plausible haemostatic co-exposure that should not be assumed inert, whose independent contribution to clinical VTE risk has not been quantified. Part D takes it up on trans-relevant grounds.

Provenance and verification (B2)

• Read in primary: ESTHER 2007 (10.1161/CIRCULATIONAHA.106.642280); Vinogradova, Toorians, Schutte (earlier tranches).
• Abstract-level: Seal 2012 (10.1210/jc.2012-2030) — used with weaknesses stated; a candidate to read in primary or drop its exact molecule-separated percentages before deposit. Smith 2004 (JAMA) a candidate to firm the CEE strand.
• Population directness (G3): the load-bearing route and CEE evidence is cisgender; flagged throughout; the trans time-course (B1) is the standing caution against straight-across reading.

B3 — Dose, serum level, and injectable oestradiol

Two outcomes graded not established — in both directions.

Four-axis grade

B3a — Dose and serum oestradiol level

No trans study establishes a dose- or level-response for VTE. In Kaiser the maximum daily oral estradiol dose in those with a VTE or stroke rose from 3.6 mg in the first two years to 5.6 mg after, flat in the event-free — untested for want of events, and entangled with duration. The contemporary cohort had serum estradiol for none of its cases. The Endocrine Society’s 2017 guideline recommends monitoring estradiol and maintaining it within the premenopausal physiological range — serum estradiol should not exceed the peak physiological range of 100–200 pg/mL, with serum testosterone below 50 ng/dL (Hembree 2017, Table 15) — but this is framed as a feminisation and testosterone-suppression target, not a validated VTE threshold, and not a mandatory therapeutic window. A “physiological range” is therefore only meaningful with the pharmacokinetic timing specified, which matters most for injectable and sublingual regimens: sublingual dosing produces a sharp early peak followed by rapid decline, and injectable regimens vary by ester and interval, so peak, midpoint and trough answer different questions and a trough alone can conceal wide peak-to-trough variability. The sampling time relative to the dose must be recorded; a pre-dose trough can help assess minimum exposure, while peak or midpoint sampling may be needed where excessive fluctuation is suspected. No sampling point is validated as a VTE-safety threshold, and the peak magnitude of sublingual dosing depends strongly on dose, assay and timing, so no single numerical peak is stated here — it would require a trans-specific sublingual pharmacokinetic study this review does not rest on. Pyra 2020 (N=4402, read in primary) found no association between venous thromboembolism and estradiol or testosterone by blood concentration in any model (estradiol odds ratio 0.99 per 100-unit change). Keeping levels physiological is sound practice and mechanistically plausible for VTE, but the venous dose-response in trans people is unestablished.

B3b — Injectable oestradiol

There is no comparative VTE data for injectable oestradiol. The scoping literature (Rothman 2024, abstract-level) concerns dose and levels, not thrombosis. Injectable regimens can produce high peak concentrations — particularly with larger doses or longer intervals — but a transient measured peak is not cumulative exposure, no peak-to-VTE threshold is established, and the concern inherits the unestablished level–VTE link of outcome 4. The only trans signal is two of a contemporary cohort’s six unprovoked events on intramuscular oestradiol: three cases, no denominator. The grade is symmetric — not shown to raise VTE risk, not shown not to.

Provenance and verification (B3)

• Read in primary: ASH 2024 (10.1182/hematology.2024000592, secondary review — framing/management, not a primary for grading). Getahun/Cunningham dose observations from B1.
• Via ASH, primaries not obtained: Pyra 2020, Nolan & Cheung 2021 — candidates to read before B3 closes. Rothman 2024 abstract-level.
• Not claimed: no dose-response or level threshold; no injectable estimate either way; the physiological range is a surrogate, not a VTE ceiling, and only interpretable with sampling timing specified.

Part C — Arterial risk

Part C grades the two arterial outcomes separately from venous thromboembolism (G6). The separation is substantive: arterial events have different dominant mechanisms — atherosclerosis, plaque disruption, platelet-rich arterial thrombosis, cardioembolism and small-vessel disease — and although haemostatic pathways overlap between venous and arterial disease, the biomarker findings of Part A cannot be assumed to predict stroke or myocardial infarction. The informative comparator also shifts toward cisgender men.

The interpretive core — observed incidence, not a demonstrated effect

On the most recent registry data, arterial-event incidence in transfeminine people is not higher, and for myocardial infarction is lower, than age-, sex-, calendar-year- and SES-standardised general-population male reference rates. The 2026 Amsterdam analysis (van Zijverden), using sex-specific general-population reference rates stratified by age and calendar year and adjusted for socioeconomic status, found — in trans women versus general-population men — a myocardial-infarction SIR of 0.50 (0.32–0.71) and a cerebrovascular-accident SIR of 0.94 (0.72–1.19). These are comparator findings: observed events against expected general-population male rates. They do not establish that gender-affirming hormone therapy prevents myocardial infarction. The study did not randomise oestrogen, had no internal untreated trans comparator, and could not adjust individual lifestyle (its lifestyle data were descriptive, in 174 of 2714 trans women); its authors themselves flag residual confounding and several alternative explanations.

The study’s authors hypothesised that the pattern reflects hormone-associated shifts in cardiovascular risk factors — feminising therapy aligning risk with the phenotypic hormone profile rather than reversing sex differences. That is biologically plausible and worth stating, but it is a hypothesis, not a demonstrated mediation, and ER-018 reports it as the authors’ interpretation rather than adopting it. Two candidate drivers sit behind any real shift and the data separate neither: the profound suppression of endogenous testosterone (at least an equal candidate to any oestrogen effect), and exogenous oestradiol’s mixed cardiometabolic changes — some favourable lipid movement, possible triglyceride and body-fat increases, and inconsistent blood-pressure effects that may vary by route, dose, baseline physiology and co-medication. Whether any of these mediates the observed MI pattern is unknown.

Four-axis grade

Editorial certainty judgements, not formal GRADE ratings. The incidence column reports comparator associations; it is not a treatment-effect estimate.

C.1 — Ischaemic stroke

The adjusted estimates show no detectable stroke excess against cisgender men: 0.94 (0.72–1.19) in the SES-adjusted 2026 Amsterdam analysis and 1.2 (0.9–1.7, not significant) in Kaiser. The earlier Amsterdam figure of 1.80 against men (Nota) takes second place — it was unadjusted for smoking in a cohort with 43–46% ever-smokers, used a Norwegian comparator, and spanned the ethinylestradiol era. Rates lie closer to the cisgender-male than the cisgender-female comparator in some analyses (van Zijverden 1.20 vs women, not significant; Kaiser 1.9); these are between-population comparisons and do not demonstrate an individual longitudinal shift caused by treatment.

Two qualifiers. The Kaiser oestrogen-initiation subgroup showed a stroke elevation appearing only after roughly six years — imprecise, but a reason for long-term vigilance rather than complacency. And a large hospital database (Manchkanti 2023, conference abstract via the ASH review, primary not obtained) reported lower cerebral-infarct rates in oestrogen-treated trans women than in cisgender women — directionally concordant, held at arm’s length as unadjusted administrative data. The stroke evidence does not establish an oestrogen-driven excess, while leaving a modest long-term contribution unexcluded.

C.2 — Myocardial infarction

Among the arterial outcomes reviewed, myocardial infarction shows the strongest comparator difference in the latest analysis. Observed incidence versus cisgender men was lower in the SES-adjusted 2026 data (0.50, 0.32–0.71) and showed no detectable difference in Kaiser (0.9, 0.6–1.5). Against cisgender women it was elevated (van Zijverden 1.30, not significant; Kaiser 1.8) — the direction expected of a between-population comparison where the cohort’s arterial risk sits nearer the male reference. There is no venous-style mechanism to invoke and no cross-comparator excess to explain — but the reassuring reading demands the same scrutiny as any convenient finding, and the candidate explanations for a lower observed SIR must be kept distinct rather than lumped as “bias”: residual confounding (including differential smoking); selection and survivor effects; healthy-user or healthcare-engagement effects; greater preventive treatment of cardiovascular risk factors (which would lower true incidence, not merely its detection); and outcome misclassification (which would explain a lower SIR only if silent events or their capture differed systematically between the cohort and the reference population). The defensible statement is narrow: the latest analysis found lower observed MI incidence than standardised general-population male reference rates, Kaiser found no detectable difference, and a cardioprotective causal effect of feminising therapy has not been established. The clinical task is management of conventional cardiovascular risk factors — hypertension, diabetes, smoking, migraine with aura, individual vascular history — alongside review of the hormone regimen.

C.3 — The transmasculine mirror (signpost to ER-023)

The same contemporary Amsterdam analysis reported stronger arterial associations in testosterone-treated trans men: a myocardial-infarction SIR of 4.20 (2.72–6.01) versus cisgender women, tending higher even against cisgender men (1.51, 0.98–2.16), with a raised cerebrovascular SIR (1.55). This warrants separate assessment in ER-023, and van Zijverden’s own conclusion is that cardiovascular risk management should be prioritised in trans men. It does not resolve the long-term arterial uncertainties that remain in transfeminine care — duration, ageing, regimen, and the out-of-scope outcomes below.

Feeding the corpus argument

Part C’s contribution is a boundary. The venous story (Parts A–B) is a real, modest association, with several potentially modifiable levers whose preventable fraction is unquantified. The arterial story is not its extension: on current cohort comparisons, feminising therapy does not show higher arterial-event incidence than cisgender-male reference populations, and one recent analysis found lower MI incidence — but whether therapy itself causes any reduction remains unknown, and the finding must not be read as demonstrated cardioprotection. Collapsing venous and arterial into one elevated “cardiovascular risk” — the commonest error in this field’s public framing — is what the separate grading exists to prevent; the reciprocal error, reading a lower comparator SIR as proof of benefit, is the one Part C now guards against explicitly.

Scope note

ER-018 grades venous thromboembolism, myocardial infarction and ischaemic stroke. Other cardiovascular outcomes — heart failure, arrhythmia, hypertension as an endpoint, cardiovascular mortality — are outside its scope and are not assessed here. The title names these three outcomes for that reason.

Provenance and verification (Part C)

• Read in primary: van Zijverden 2026 (10.1093/eurheartj/ehaf837) — the load-bearing arterial source, read for what it does and does not establish (comparator associations, not treatment effects; no lifestyle adjustment in the main model); Getahun 2018 (10.7326/M17-2785); Nota 2019 (10.1161/CIRCULATIONAHA.118.038584).
• Abstract-level, via ASH: Manchkanti 2023 (Blood 2023;142(suppl 1):1282) — directional counterpoint only. Connelly 2019 (Hypertension) a candidate for the arterial risk-factor mechanism.
• Independence (G7): van Zijverden takes precedence over Nota but is the same Amsterdam lineage; Getahun (Kaiser) is the independent anchor and agrees directionally.
• Not claimed: no extension of Part A’s mechanism to arterial events; no protective causal claim for oestradiol; no adoption of the authors’ mediational interpretation as fact; no transmasculine estimate beyond the ER-023 signpost.

Part D — Management

Part D translates the graded evidence of Parts A–C into two questions: who carries the venous risk, and what can be changed. It is not a prescribing protocol and not a policy position; it should not be used to start, stop or withhold treatment, and no decision should rest on it alone. It carries a standing frame: the venous association is modest, and several potentially modifiable treatment- and patient-level factors can be addressed — though the degree to which addressing them reduces clinical events in transfeminine populations is not quantified. That sits opposite a benefit that matters to patients: gender-affirming hormone therapy is desired and clinically important to many, and predominantly observational studies associate it with improvements in gender dysphoria, psychological well-being and depressive symptoms — though selection, concurrent care and the absence of randomised comparison limit causal attribution, and evidence on suicide attempts and mortality is weaker still and should not be collapsed with symptom measures into one figure. This benefit is graded on the same four axes as the harms in the front-matter mental-health declaration, and held to the same standard — not upgraded because it is the finding the author would prefer. Withholding or interrupting desired therapy can cause psychological and physical harms for some patients, and those consequences belong in shared decision-making alongside thrombotic risk.

D.1 — Who: risk stratification

The events appear to cluster (B1), but the evidence is aggregate-level and is not a validated prediction tool. No transfeminine-specific VTE risk score exists.

Associated with fewer observed events (cohort level): younger average age and shorter average duration — study series with mean age under ~37.5 years or mean therapy under ~53 months reported few events. These are aggregate associations, not individual thresholds; 37.5 years and 53 months are data-derived splits, not clinical cut-offs, and none identifies a particular patient as being at population-baseline risk.

Established VTE risk factors (general venous medicine), which should be prioritised: personal or family history of VTE, high-risk thrombophilia, active malignancy, recent major surgery, immobilisation, and obesity, along with older age.

Treatment-related considerations, graded by evidential status (not equal in weight): strong historical concern — ethinylestradiol; indirect (cisgender/menopausal) concern — conjugated equine oestrogen and oral route; biologically plausible but unvalidated for trans VTE outcomes — higher oestradiol dose and cyproterone co-treatment.

A note on the cardiometabolic factors. Hypertension, dyslipidaemia and diabetes appeared in a small cohort’s cases, but they are stronger arterial than direct venous predictors, and that cohort had sparse events; they warrant management as part of overall vascular care, not treatment as established venous predictors. And some may change during long-term therapy, so the cohorts cannot determine whether they preceded treatment, arose independently, or partly mediated risk.

D.2 — What: the modifiable levers, graded by evidence strength

Molecule is the clearest estrogen-formulation intervention: avoiding ethinylestradiol is well-supported; preferring estradiol to CEE is reasonable but rests on indirect evidence. WPATH SOC-8 aligns, and its grading matches the evidence weight — it recommends (strong) against ethinylestradiol (Statement 12.13) and suggests (conditional) against conjugated estrogens where estradiol is available (12.15). In a patient already on contemporary 17β-oestradiol, individual and transient risk factors matter more than fine formulation distinctions.

Route is a defensible but inferred preference for higher-risk patients — B2a graded it an extrapolation, and Part D does not upgrade it. WPATH SOC-8 addresses this directly but at conditional strength: Statement 12.14 suggests (not the stronger “recommends”) transdermal oestrogen for those at higher VTE risk “based on age > 45 years or a previous history of venous thromboembolism”. Its supporting text is candid that transdermal use with concomitant anticoagulation may lower risk but that data to guide such decisions are limited — expert risk-management guidance consistent with B2a’s grade, not trans-specific outcome validation.

Anti-androgen. Cyproterone is a plausible haemostatic co-exposure (ESTHER’s progestogen-class signal, Schutte’s platelet/fibrinogen changes, Toorians’ APC rise), but none isolates a cyproterone-dose–VTE relationship, and the study built to compare doses (Kuijpers, 882 trans women) had too few events to test thrombosis. What Kuijpers establishes is that 10 mg suppresses testosterone as completely as 100 mg, with less prolactin and HDL-cholesterol effect — so use the lowest effective dose. Cyproterone’s principal established cumulative-dose concern is meningioma, not VTE: the European Medicines Agency restricted higher-dose cyproterone (daily doses of 10 mg or more) to its authorised androgen-dependent indications only once other options, including lower doses, have failed, advising gradual reduction to the lowest effective dose after response — because meningioma risk rises with cumulative dose. Meningiomas have been reported primarily at doses of 25 mg/day and above, after prolonged (several-year) exposure, with the tumour rare overall (an estimated 1 to 10 per 10,000 people, depending on dose and duration). Cyproterone-containing products are contraindicated in patients with a current or previous meningioma, and if a meningioma is diagnosed during treatment, all cyproterone-containing products must be permanently discontinued (EMA/70255/2020, 14 February 2020). This is not a transgender-specific ruling, but a cumulative-dose concern directly relevant to off-label transfeminine use — and it reinforces the dose-minimisation point above. Cyproterone’s regulatory status is jurisdiction-specific: it is an approved product in Canada and across Europe — used off-label for feminising therapy — but has never been approved in the United States, so only in a US context does the practical question become spironolactone versus a GnRH agonist rather than cyproterone dose. Where appropriate and accessible, a GnRH agonist or spironolactone may avoid cyproterone; selection should weigh efficacy, cost, renal function, potassium, blood pressure, volume-depletion symptoms and urinary frequency (spironolactone), administration burden, bone health and flare (GnRH agonist), and patient preference — not a universal hierarchy.

Dose and conventional factors follow B3 and general medicine: keep levels physiological with the sampling timing specified (peak versus trough diverge on injectable and sublingual regimens), and address smoking, obesity, mobility, blood pressure, lipids and glycaemia as overall vascular care — while prioritising the established VTE factors of D.1.

D.3 — Thrombophilia and screening

Routine population screening before therapy is not recommended, because thrombophilia testing has low yield and uncertain utility in unselected patients; testing should be targeted where a personal VTE history or a strong family history suggestive of high-risk inherited thrombophilia means the result would change management. A history of VTE, high-risk thrombophilia, antiphospholipid syndrome, active malignancy or myeloproliferative disease requires individualised, haematology-informed assessment — these are not equivalent (antiphospholipid syndrome is not heterozygous factor V Leiden), are not automatically permanent contraindications, but may materially alter formulation, timing and anticoagulation decisions.

D.4 — After a venous thromboembolism

Treat acute VTE according to standard thrombosis guidelines; suspected acute VTE is an urgent diagnostic matter and hormone decisions should not delay assessment. A direct oral anticoagulant is appropriate for many patients, but selection must account for renal function, malignancy, drug interactions and bleeding risk — and in high-risk antiphospholipid syndrome, particularly triple-positive APS or prior arterial thrombosis, a vitamin K antagonist is generally preferred over a DOAC. The ASH review is an expert synthesis, not a transfeminine-specific guideline, and the underlying anticoagulation guidance is general. Gender-affirming hormone therapy use alone is not an indication for routine long-term primary anticoagulant prophylaxis; standard prophylaxis remains indicated when established surgical, medical or immobilisation criteria are met.

On whether to continue the hormones: continuation or resumption during therapeutic anticoagulation may be reasonable after individualised assessment, usually with review of formulation, dose and route — a switch from oral to transdermal oestradiol, and dose reduction, are common. The evidence that anticoagulation mitigates recurrence during continued oestrogen is largely extrapolated from cisgender contraceptive and menopausal populations, not transfeminine care directly: the largest analysis (Martinelli 2016, from the EINSTEIN DVT/PE trials, read in primary) found recurrent VTE incidence of 3.7%/year on hormonal therapy versus 4.7%/year off during therapeutic anticoagulation, adjusted hazard ratio 0.56 (95% CI 0.23–1.39) — reassuring in direction, but the wide interval crosses one, so it shows no detected increase, not proven safety, and the study explicitly excluded its two transgender participants. The independent contribution of ongoing therapy to recurrence in transfeminine care has not been quantified. Ongoing oestrogen should be considered in recurrence-risk assessment — while recognising that the recurrence implications of contemporary transdermal estradiol are not directly quantified and may differ from oral EE or CEE — and anticoagulation duration and intensity should follow standard provoked/unprovoked, recurrence- and bleeding-risk assessment. Continuation is often reasonable; it is not automatic for cancer-associated, antiphospholipid-associated or recurrent events, and it exchanges thrombotic risk for a bleeding burden that belongs in the shared decision.

D.5 — Perioperative management — signpost only

ER-018 does not adjudicate perioperative hormone management. Available evidence has not demonstrated that routine perioperative oestrogen cessation reduces VTE — one retrospective cohort of 402 gender-affirming surgical cases reported one venous event among those who suspended oestrogen and none among those who continued — but comparative data are sparse and insufficient to establish safety across procedures and risk strata, and this is not the same as evidence against withholding in every setting. Perioperative risk depends on procedure, duration, prophylaxis, mobility, age and prior VTE; older patients carry higher baseline risk. The question is unsettled and consequential enough to deserve dedicated treatment, flagged here as a future build and not resolved.

Provenance and verification (Part D)

• Read in primary: Kuijpers 2021 (10.1210/clinem/dgab427) — cyproterone dose-minimisation (efficacy/side-effects by dose: yes; dose–VTE: no; meningioma cumulative-dose concern). EMA cyproterone restriction (EMA/70255/2020, 14 February 2020) — meningioma cumulative-dose restriction, contraindication and permanent-discontinuation rule, read from the primary EMA document. WPATH SOC-8 Chapter 12 — Statements 12.13/12.14/12.15 verified and quoted at grade. Health Canada drug register — cyproterone (Androcur) approved product, off-label for GAHT. ESTHER, Schutte, Toorians, Cunningham, Totaro, Getahun, Nota, van Zijverden from earlier tranches.
• Not obtained (not load-bearing, flagged): Ott 2010 and Kozato 2021 (perioperative) — the no-screening recommendation rests on general thrombosis consensus, not on the unread Ott series; the perioperative signpost does not rest on the exact Kozato figure.
• Not claimed: no cyproterone dose–VTE response; no primary-prophylaxis recommendation; no perioperative protocol; no universal anti-androgen hierarchy; no claim that the preventable fraction of the venous association is known.

References

Author-year in-text; this list alphabetical by first author.

Asscheman H, Gooren LJ, Eklund PL. Mortality and morbidity in transsexual patients with cross-gender hormone treatment. Metabolism. 1989;38(9):869–873.

Canonico M, Oger E, Plu-Bureau G, et al; ESTHER Study Group. Hormone therapy and venous thromboembolism among postmenopausal women: impact of the route of estrogen administration and progestogens. Circulation. 2007;115(7):840–845. doi:10.1161/CIRCULATIONAHA.106.642280

Coleman E, Radix AE, Bouman WP, et al. Standards of care for the health of transgender and gender diverse people, version 8. Int J Transgend Health. 2022;23(suppl 1):S1–S259. doi:10.1080/26895269.2022.2100644

Cunningham TJ, et al. Venous thromboembolism in transgender women receiving gender-affirming hormone therapy. J Clin Med. 2026;15(11):4166. doi:10.3390/jcm15114166

da Cruz KLdO, et al. Haemostatic effects of gender-affirming hormone therapy. PLoS One. 2025;20:e0323606. doi:10.1371/journal.pone.0323606

European Medicines Agency. Restrictions in use of cyproterone due to meningioma risk. EMA/70255/2020. 14 February 2020.

Getahun D, Nash R, Flanders WD, et al. Cross-sex hormones and acute cardiovascular events in transgender persons: a cohort study. Ann Intern Med. 2018;169(4):205–213. doi:10.7326/M17-2785

Giltay EJ, Gooren LJG, Emeis JJ, Kooistra T, Stehouwer CDA. Oestrogens and fibrinolysis in transsexual men and women. Arterioscler Thromb Vasc Biol. 2000;20(5):1396.

Glintborg D, Rubin KH, Petersen TG, et al. Cardiovascular risk in Danish transgender persons: a matched historical cohort study. Eur J Endocrinol. 2022;187(3):463–477. doi:10.1530/EJE-22-0306

Hembree WC, Cohen-Kettenis PT, Gooren L, et al. Endocrine treatment of gender-dysphoric/gender-incongruent persons: an Endocrine Society clinical practice guideline. J Clin Endocrinol Metab. 2017;102(11):3869–3903. doi:10.1210/jc.2017-01658

King H, Padilla Kelley T, Shatzel JJ. The hematologic management of transgender and gender-diverse patients receiving gender-affirming hormone therapy. Hematology Am Soc Hematol Educ Program. 2024;2024:652–663. doi:10.1182/hematology.2024000592

Kozato A, Fox GWC, Yong PC, et al. No venous thromboembolism increase among transgender female patients remaining on estrogen for gender-affirming surgery. J Clin Endocrinol Metab. 2021;106(4):e1586–e1590. doi:10.1210/clinem/dgaa966

Kuijpers SME, Wiepjes CM, Conemans EB, Fisher AD, T’Sjoen G, den Heijer M. Toward a lowest effective dose of cyproterone acetate in trans women: results from the ENIGI study. J Clin Endocrinol Metab. 2021;106(10):e3936–e3945. doi:10.1210/clinem/dgab427

Manchkanti S, et al. Thrombotic and arterial events among transgender women receiving estrogen: a nationwide analysis. Blood. 2023;142(suppl 1):1282.

Martinelli I, Lensing AWA, Middeldorp S, et al. Recurrent venous thromboembolism and abnormal uterine bleeding with anticoagulant and hormone therapy use. Blood. 2016;127(11):1417–1425. doi:10.1182/blood-2015-08-665927

Nolan BJ, Cheung AS. Gender-affirming hormone therapy for transgender and gender-diverse adults in Australia. Intern Med J. 2024;54(9):1450–1457. doi:10.1111/imj.16413

Nolan BJ, Cheung AS. Relationship between serum estradiol concentrations and clinical outcomes in transgender individuals undergoing feminizing hormone therapy: a narrative review. Transgend Health. 2021;6(3):125–131.

Nota NM, Wiepjes CM, de Blok CJM, Gooren LJG, Kreukels BPC, den Heijer M. Occurrence of acute cardiovascular events in transgender individuals receiving hormone therapy. Circulation. 2019;139(11):1461–1462. doi:10.1161/CIRCULATIONAHA.118.038584

Ott J, Kaufmann U, Bentz E-K, Huber JC, Tempfer CB. Incidence of thrombophilia and venous thrombosis in transsexuals under cross-sex hormone therapy. Fertil Steril. 2010;93(4):1267–1272.

Pyra M, Casimiro I, Rusie L, et al. An observational study of hypertension and thromboembolism among transgender patients using gender-affirming hormone therapy. Transgend Health. 2020;5(1):1–9. doi:10.1089/trgh.2019.0061

Rothman MS, et al. Injectable estradiol in gender-affirming hormone therapy. Endocr Pract. 2024. doi:10.1016/j.eprac.2024.05.008

Schutte MH, Kleemann R, Nota NM, et al. The effect of transdermal gender-affirming hormone therapy on markers of inflammation and haemostasis. PLoS One. 2022;17(1):e0261312. doi:10.1371/journal.pone.0261312

Seal LJ, Franklin S, Richards C, Shishkareva A, Sinclair C, Barrett J. Predictive markers for mammoplasty and a comparison of side effect profiles in transwomen taking various hormonal regimens. J Clin Endocrinol Metab. 2012;97(12):4422–4428. doi:10.1210/jc.2012-2030

Toorians AWFT, Thomassen MCLGD, Zweegman S, et al. Venous thrombosis and changes of haemostatic variables during cross-sex hormone treatment in transsexual people. J Clin Endocrinol Metab. 2003;88(12):5723–5729. doi:10.1210/jc.2003-030520

Totaro M, Palazzi S, Castellini C, et al. Risk of venous thromboembolism in transgender people undergoing hormone feminizing therapy: a systematic review and meta-analysis. Front Endocrinol. 2021;12:741866. doi:10.3389/fendo.2021.741866

van Kesteren PJM, Asscheman H, Megens JAJ, Gooren LJG. Mortality and morbidity in transsexual subjects treated with cross-sex hormones. Clin Endocrinol (Oxf). 1997;47(3):337–342.

van Zijverden LM, Wiepjes CM, van Diemen JJK, et al. Cardiovascular disease in transgender people receiving gender-affirming hormone therapy. Eur Heart J. 2026:ehaf837. doi:10.1093/eurheartj/ehaf837

Vinogradova Y, Coupland C, Hippisley-Cox J. Use of hormone replacement therapy and risk of venous thromboembolism: nested case-control studies using the QResearch and CPRD databases. BMJ. 2019;364:k4810. doi:10.1136/bmj.k4810

Claim-to-reference map

For each load-bearing claim: the source(s) cited, what the source actually shows, and a support check — verifying that each citation resolves to a source that supports the claim, not merely that the citation resolves. Cross-referenced against the study-overlap matrix so that no figure re-cited across a single cohort lineage is counted as independent corroboration, with a population-directness flag on every claim that borrows from a cisgender source.

Support-check key: ✓ direct support · ⚠ supported with population/scope caveat (flagged in text) · ⟳ two-lineage corroboration verified · ◐ observed-comparator, not causal · ○ rests on unobtained/secondary source (flagged)

Part A — Mechanism

Part B1 — VTE Magnitude

Part B2 — Route and Molecule

Part B3 — Level and Injectable

Part C — Arterial

Part D — Management

Independence audit (cross-referenced to the study-overlap matrix)

• The only genuine two-lineage corroboration is the ~2× VTE magnitude (Getahun ⟳ van Zijverden). Every other cohort claim rests on one lineage or one source, and is marked accordingly.
• No claim in this map cites Totaro as independent confirmation of Getahun, Nota or van Kesteren — Totaro pools them.
• No Amsterdam-lineage figure is counted twice: van Zijverden supersedes Nota (not additive); van Kesteren and Asscheman feed only the quarantined transmission appendix.
• Every cisgender-sourced claim carries a population flag (⚠): route (ESTHER, Vinogradova), progestogen class (ESTHER), post-VTE continuation (Martinelli). None is applied to trans populations without that flag.
• Glintborg is not cited for a VTE magnitude claim — only for independence, GAHT-non-mediation, and surveillance-bias corroboration, consistent with its composite-outcome design.

Pre-deposit action: once the reference list is locked to first-appearance numbering, re-run this map against the numbered citations to confirm every in-text marker resolves to the intended entry — the check a numeric integrity scan cannot perform.

Study-overlap matrix

Which cohorts are independent, which share a lineage, which pool the others

This map prevents counting one dataset’s finding as independent corroboration when it appears under several citations — distinguishing genuinely independent facts from the same fact re-cited. Every relationship below is drawn from sources read in full; items not read in full are marked.

The core principle for this review

For venous thromboembolism, there are effectively two independent cohort lineages measuring outcomes directly, plus one meta-analysis that pools them and several single-source or cisgender inputs. Numerical agreement between the two lineages is corroboration; numerical agreement within a lineage (or a meta-analysis quoting its members) is not.

Cluster 1 — The Amsterdam lineage (ACOG / Centre of Expertise on Gender Dysphoria, VUmc / Amsterdam UMC)

One institutional cohort, sampled and re-analysed across decades. Treat all of these as NON-independent of one another.

Mechanism sub-lineage (same institution, VUmc/Amsterdam gender clinic):

Consequence: the entire mechanism argument (Part A) rests substantially on one clinic’s lineage, with da Cruz (Cluster 3) the sole independent mechanistic check. Flagged in Part A provenance.

Cluster 2 — Kaiser STRONG (US, independent)

Cluster 3 — Other trans-direct cohorts

Cluster 4 — The pooling meta-analysis (NOT independent corroboration)

Cluster 5 — Cisgender / menopausal inputs (route & molecule; indirect population)

Cluster 6 — Secondary syntheses (framing only; never a primary for grading)

Independence summary (the one-line answer for any figure)

• VTE magnitude ≈ 2×: corroborated by two independent lineages — Kaiser (Getahun) and Amsterdam (van Zijverden). Genuine.
• Arterial (MI lower / CVA similar): Amsterdam (van Zijverden) primary, Kaiser (Getahun) concordant, Manchkanti (abstract) directional. Two-lineage support.
• Anything from Totaro: contains Getahun + Nota + van Kesteren — not independent of them.
• “45-fold” / “20-fold”: single Amsterdam-lineage population (Asscheman = van Kesteren); appears independent only because it is re-cited (ASH → Asscheman). Quarantined.
• Mechanism (Part A): one Amsterdam clinic lineage + da Cruz as the sole independent check.
• Route/molecule: two independent cisgender datasets (ESTHER, Vinogradova); no independent trans outcome data.

Not obtained (integrate/verify at sweep, with independence status above): van Kesteren, Asscheman, Kozato, Ott, Nolan & Cheung 2021. Now obtained this session and struck from the list: Glintborg, Pyra, Martinelli, SOC-8 Ch.12.

Appendix — Evidence transmission

This appendix documents cases where a figure or finding changed shape as it moved from a primary source into secondary characterisations — the corpus’s standing thesis that the errors in this field are overwhelmingly in transmission, not in the primaries. Each case is recorded because catching it is what the method is for; none is asserted as a finding of this review, and where a primary could not be obtained the item is quarantined, not used.

The standing pattern

Across sources checked for this review and its predecessors, a recurring share propagated with corruption: a significant finding dropped from a secondary’s one-line summary; a sample size inconsistent between a review’s body text and its own table; a striking figure re-cited across multiple reviews, all tracing to one primary that none of the citing authors appears to have re-read. The mechanism is consistent — secondary sources compress, and compression strips qualifiers, drops the inconvenient subgroup, and hardens an estimate into a headline. The defence is equally consistent: read the primary, and read it before quoting the secondary.

Three cases from this review, in order of how much they matter.

Case 1 — The “45-fold” figure: one datum, four citations, no obtainable primary

What circulates. A “45-fold increased risk of venous thromboembolism (occurring in 6.3%)” in trans people treated with ethinylestradiol 100 µg plus cyproterone acetate 100 mg daily.

The attribution chain. The figure is attributed, consistently, to Asscheman 1989 (Metabolism) — by the ASH 2024 education review and, independently, by the Nolan & Cheung 2024 narrative review. Both cite the same 1989 primary. That same Amsterdam population also appears in van Kesteren 1997, which reports 45 VTE cases in its cohort and a fold-increase nearer twentyfold — and the coincidence of “45 cases” (van Kesteren) with “45-fold” (Asscheman) is what first raised the possibility of a count-to-fold conflation.

What the reading revised. The conflation hypothesis is weakened, not confirmed. Two independent 2024 reviews attributing “45-fold (6.3%)” specifically to Asscheman 1989 — with a consistent 6.3% incidence attached — makes it more likely the figure is a genuine (if extreme, obsolete-regimen) statistic from that primary than a transcription artefact of van Kesteren’s case count. The two Amsterdam figures (Asscheman ~45-fold, van Kesteren ~20-fold) cannot be reconciled without the primaries, and both are the same institutional lineage (see the study-overlap matrix), so neither independently corroborates the other in any case.

Handling. Asscheman 1989 and van Kesteren 1997 are not obtained. The “45-fold” figure is therefore quarantined: attributed, not asserted, and not used in the graded body. It describes a 100 µg ethinylestradiol regimen that is deprecated and bears no relation to contemporary 17β-oestradiol dosing. Its only role here is as a documented example of a figure that four sources carry and none can currently verify.

Case 2 — A significant finding a secondary dropped: the Pyra progestin signal

What the secondary said. The ASH 2024 review characterised Pyra 2020 as finding “no association between thrombosis and gender-affirming hormone therapy as assessed by blood concentrations” — accurate, as far as it goes.

What the primary also said. Reading Pyra 2020 in full recovered a finding the secondary omitted: recent progestin prescription was associated with roughly threefold odds of thromboembolism (adjusted OR 2.95, 95% CI 1.02–8.57), with a much larger but far more imprecise estimate for medroxyprogesterone acetate specifically. This is a significant, trans-direct association — and it is exactly the kind of finding that matters for the progestogen-co-exposure question this review takes up (B2c).

Why it matters. The concentration-null and the progestin-positive are both true; the secondary carried the first and dropped the second. Had this review cited Pyra through the secondary, as it initially did, it would have carried a partial reading of a paper it treated as evidence. The finding now appears in B2c, fully caveated (small event count, MPA not cyproterone, wide interval, US spironolactone-based cohort). This is the transmission thesis in miniature: the primary carried more than the summary of it.

Case 3 — An internal inconsistency in a secondary: the Kozato sample size

The discrepancy. The ASH 2024 review reports the Kozato perioperative cohort as N=402 in its body text and N=407 in its own summary table, for the same study. A five-patient discrepancy within a single review, unresolved on its face.

Handling. The Kozato primary (10.1210/clinem/dgaa966) is not obtained, so the discrepancy cannot be adjudicated here, and the perioperative signpost (D.5) does not rest on the exact figure. Recorded so that any future citation of Kozato resolves the count against the primary rather than either secondary value.

Case 4 — A superseded copy in the retrieval layer, resolved by the correct primary

The version hazard. The source-retrieval connector held a copy of the Endocrine Society 2017 guideline flagged superseded, carrying a warning not to quote figures from it. The guideline is the common source for the serum-estradiol monitoring range (100–200 pg/mL) cited across the field.

How it resolved — the archive gate working as designed. Rather than quote the figure from a flagged copy, the review initially held it as attributed-not-asserted. The author’s own archived copy was then confirmed to be the correct published version (Hembree 2017, Table 15), which verifies the figure directly. B3 now states it as verified. This is the transmission discipline running in the retrieval layer: a superseded copy in an automated source is exactly the kind of version hazard the field propagates unknowingly, and the safeguard is not to quote from a flagged source until the correct primary is in hand. Here the correct primary was in hand, and the figure holds. Recorded as a clean resolution, not an open issue.

A note on citation precision (not an error, a caution)

Getahun 2018 is variously cited with N=102,417 (the full matched cohort, including cisgender comparators) and with the ~2,800 transfeminine figure (the exposed subgroup). Both are correct for what they describe; a claim about trans women that quotes the 102,417 figure has quietly borrowed the whole cohort’s N for a subgroup finding. This review uses the transfeminine figures throughout, and flags the distinction because the larger number circulates in secondary citations where the subgroup number is meant.