# Fact-check of conversation.md claims, Sections A–D

Verdict legend:
- Confirmed (C): primary source backs it up
- Approximately right (≈): order-of-magnitude correct but oversimplified or context-dependent
- Wrong (W): contradicted by literature
- Genuinely uncertain (?): literature disagrees or no good data

Sources consulted: Wikipedia "Pharmacokinetics of estradiol" and "Estrone sulfate" (heavily cited compilations), Kuhl 2005 Climacteric, Stanczyk 2013/2024, Longcope et al. 1968 JCI, Yager et al. 1989 Cancer Res, Canonico ESTHER 2007, Scarabin/Mohammed 2015 meta-analysis, transfemscience.org reviews, Bar 2024 abstract.

---

## A. Basic enzyme biochemistry

These are mostly textbook claims and largely correct. I checked each briefly.

**A1. CA II catalyzes CO2 + H2O ⇌ HCO3- + H+, accelerates ~10⁶–10⁷×.**
Verdict: **Confirmed.** Carbonic anhydrase II kcat ≈ 10⁶ s⁻¹, uncatalyzed reaction rate is ~10⁻¹ s⁻¹ → rate enhancement ~10⁷. Textbook (Lindskog 1997; Berg, Stryer Biochemistry).

**A2. 17β-HSD type 2 dominates in adult liver, NAD⁺-dependent, oxidative (E2 → E1).**
Verdict: **Confirmed.** HSD17B2 prefers NAD⁺ and catalyzes oxidative inactivation of E2 to E1 (and also testosterone → androstenedione). It is highly expressed in liver, gut, endometrium, and placenta. (Wu et al. 1993; Mindnich/Adamski reviews; UniProt P37059.)

**A3. 17β-HSD type 1 is NADPH-dependent, reductive (E1 → E2), high in placenta/ovary.**
Verdict: **Confirmed.** HSD17B1 strongly prefers NADPH and catalyzes E1 → E2 reduction. Highly expressed in placental syncytiotrophoblast and ovarian granulosa cells. (Peltoketo et al. 1999; Lin et al. 2006.) Side note: HSD17B1 also has modest activity for E2 → E1, but the equilibrium in vivo with high NADPH/NADP⁺ strongly favors reduction (PMC2736091).

**A4. SULT1E1 specifically sulfates estrogens at C3-phenol; Km ~ nM range.**
Verdict: **Confirmed.** SULT1E1 has a Km for E2 of ~5–20 nM, the lowest Km of any human SULT for any substrate, hence its selectivity for estrogens at physiological concentrations. (Falany 1997.) Above ~100 nM substrate inhibition kicks in — this is the well-known "bell-shaped" SULT1E1 curve and probably explains why the v3 model's "substrate-inhibition" sulfation term is needed.

**A5. UGT1A1, 1A3, 1A8, 1A10 glucuronidate E2 at C3; UGT2B7 at C17.**
Verdict: **Confirmed.** This is the standard regiospecificity reported by Lépine et al. 2004 (JCEM) and reviewed in Itäaho et al. 2008 (Drug Metab Dispos). UGT1A10 has highest E2-3-G activity; UGT2B7 dominates E2-17-G. Other UGTs (1A4, 2B4) contribute less.

**A6. UGT1A8/1A10 are gut-specific (gut wall first-pass).**
Verdict: **Confirmed.** UGT1A7, 1A8, and 1A10 are extra-hepatic, with 1A10 predominantly intestinal and 1A8 also intestinal/colonic. (Tukey & Strassburg 2000; Strassburg et al. 1998.) This is the molecular basis for gut-wall first-pass conjugation of oral E2.

**A7. STS reverses sulfation; expressed in many tissues including target/breast.**
Verdict: **Confirmed.** Steroid sulfatase (STS) hydrolyzes E1S → E1 and DHEAS → DHEA. Expressed widely; the breast/endometrial expression is the basis for the STS-inhibitor therapeutic concept. (Reed et al. 2005 Endocr Rev.)

**Section A overall: solid. No corrections needed.**

---

## B. PK constants & ratios (load-bearing)

### B1. Oral E2 bioavailability ~5% [C1]

Verdict: **≈ Approximately right but the number is fragile and the *definition* matters.**

What the literature actually says:
- Wikipedia/Kuhl 2005 (the most-cited summary): "absolute bioavailability of oral micronized estradiol is approximately **5%**, with a possible range of **0.1% to 12%**" (with another sometimes-quoted figure of "28 to 127% mean AUC variability across individuals," i.e. ~4.6-fold).
- Stanczyk reviews echo ~5%.

**Important nuance that the previous AI papered over: "bioavailability of what?"**
- For *unconjugated free E2 plasma AUC*, F ≈ 2–5%. This is the relevant number for plasma free-E2 effects (uterus, breast, brain, bone, cardiovascular targets at peripheral concentrations).
- For *total estrogen exposure* (E2 + E1 + E1S + glucuronides), F is much higher — perhaps 20–40%, because the first-pass losses are mostly *conjugation* to E1S and E1G, which form a circulating reservoir and partly recycle. This is why oral E2 produces such high E1 and E1S levels and why oral E2 has a clinical half-life of 13–20 h vs. 0.5–2 h IV.
- For *hepatic estrogenic effect* (SHBG induction etc.), the relevant exposure is the high portal concentration during absorption — neither plasma F nor total-estrogen F describes it well. The clinical effect of 1 mg oral on SHBG is comparable to ~100 μg/d transdermal, which is roughly the "hepatic equivalence" of 1 mg PO.

So "5%" is fine for free-E2 AUC, but the previous AI's free-pass use of "5%" without specifying which compartment is sloppy. (Source: Kuhl 2005, Climacteric 8 Suppl 1; transfemscience.org "Approximate Comparable Dosages of Estradiol by Different Routes.")

### B2. EE bioavailability ~40–50% [C1]

Verdict: **Confirmed.** Multiple sources: "oral bioavailability of EE is approximately 45%, range 38–48%" (Wikipedia/Stanczyk 2013 *Contraception*; Goldzieher 1990). Higher than E2 because the 17α-ethinyl substitution sterically blocks 17β-HSD oxidation, so EE survives first-pass without being inactivated to a less-potent metabolite.

### B3. Plasma free E2 fraction ~2% [C1]

Verdict: **Confirmed at baseline; partial caveat on the "falls to ~1% in pregnancy" claim if that appeared in the source AI's chain.**

Standard distribution (Dunn 1981, classic Anderson 1974 study; non-pregnant women, follicular):
- ~2% free
- ~38% SHBG-bound
- ~60% albumin-bound (loosely bound, "bioavailable")

In pregnancy: SHBG rises 5–10×; total E2 rises ~100×; molar binding equation predicts free fraction drops to ~0.5–1%. So "~1% in pregnancy" is correct in direction. With oral HT raising SHBG +67–171%, free fraction drops modestly (perhaps from 2% → ~1.5%), with corresponding compensatory rises in total measured E2 if dose is fixed. (Source: Wikipedia "Estradiol"; PMC discussion of free hormone hypothesis; Lindberg 2005 JCEM for SHBG induction magnitudes.)

### B4. Hepatic ER-α Kd = 0.25 nM (Yager 1989) [C1]

Verdict: **≈ Number is in the paper, but the assay and interpretation are weak load-bearing.**

The Yager et al. 1989 paper (*Cancer Res* 49: 6605, PMID 2573415) is a **rat hepatocyte study** in a two-stage hepatocarcinogenesis model using ethinylestradiol as promoter. It reports a Kd of 0.25 nM for [³H]-E2 binding to rat hepatocyte nuclear/cytosolic ER using an exchange assay. The value is reasonable: across tissues, ER-α Kd for E2 is typically reported as 0.05–1 nM depending on assay (cell-free vs whole-cell, competitive vs saturation, ligand). 0.25 nM is in the textbook ballpark.

**Caveats the previous AI's chain glossed over:**
1. This is rat liver, not human liver. Rodent ER affinity is usually within ~2× of human, but not identical.
2. *In vivo*, ER-α responsiveness on SHBG (the supposed downstream effect) is not driven purely by Kd. SHBG promoter activation in HepG2 cells responds to E2 with EC50 around 1–10 nM (HepG2 is a particular hepatoma line; Selva & Hammond 2009, PMID 19386766). So the "convolve free E2 with a Hill curve with Kd = 0.25 nM" model the AI built is *too sensitive* — at 30–60 pg/mL free E2 (~0.1–0.2 nM), it would already saturate hepatic ER, contradicting the dose-response observed clinically.
3. There is no single "Lindberg 2003 paper" that builds a Hill curve from 0.25 nM. The clinical SHBG-vs-oral-dose data (Lindberg et al. — but the right citation is likely 2005, JCEM 90:3431, "Effects of Oral and Transdermal Estradiol Administration on Levels of SHBG in Postmenopausal Women") reports SHBG +67–171% with oral E2 but does *not* publish a fitted Hill curve with that Kd. The AI invented a chain ("Yager Kd → Lindberg Hill curve") that isn't actually in any single paper.

**Correct framing:** "Hepatic ER-α has Kd for E2 in the sub-nM range (Yager 1989 reports 0.25 nM in rat). Empirically, clinical SHBG induction in postmenopausal women on oral E2 is dose-dependent in the 0.5–2 mg/d range, suggesting the *effective* EC50 for SHBG induction sits around the portal E2 concentration achieved with 0.5 mg PO — much higher than the receptor's nominal Kd because of integrated transcription/turnover dynamics."

### B5. Hill coefficient n ≈ 1.0–1.2 [C1]

Verdict: **? Genuinely uncertain / probably reasonable default.**

There is no single experimental measurement of *the* Hill coefficient for hepatic ER-α → SHBG. A Hill coefficient near 1 is the canonical assumption for monomeric ligand binding to a receptor without cooperativity. ER-α functions as a dimer but the ligand-binding step is monomeric per subunit; in reporter assays, n is usually 1.0–1.5 (Anstead et al. 1997). Using ~1.0–1.2 is fine as a default, but flag it as a modeling assumption, not an experimentally validated parameter.

### B6. MCR_E2 ≈ 1400 L/day, MCR_E1 ≈ 2200 L/day, MCR_E1S ≈ 150 L/day [C1]

Verdict: **Confirmed (with the standard caveat that classic MCR studies normalize to BSA).**

From Longcope, Layne, Tait 1968 (*J Clin Invest* 47:93, PMC297151):
- Whole-blood MCR of E2 in women: **1,360 ± 40 L/d/m²** → ~2,300 L/d for a 1.7 m² woman, but the "absolute" 1400 L/d figure used by the AI is what gets quoted in textbooks when you don't normalize, and is roughly OK.
- MCR of E1 in women: **1,910 ± 100 L/d/m²** → ~3,200 L/d for 1.7 m². The AI's "2200 L/d" is on the low side. Stanczyk lists similar.
- MCR of E1S: **~80 L/d/m²** (Ruder et al. 1972, JCI; PMC302214) → ~140 L/d absolute. AI's 150 is fine.

So 1400/2200/150 are reasonable order-of-magnitude figures with the right ordering and ratios. Strictly, the women's MCR_E1 is closer to 2700–3200 L/d once you de-normalize, and the AI's 2200 underestimates by ~30%.

### B7. Free E2 half-life: 1–2 h; E1S half-life: 10–12 h [C1]

Verdict: **Confirmed.**
- IV E2 half-life: 0.5–2 h (Düsterberg 1985; Wikipedia). The very rapid component is distribution; terminal is 1–2 h.
- E1S half-life: 10–12 h (Ruder et al. 1972; Wikipedia "Estrone sulfate").
- Note that *oral* E2 has an apparent half-life of 13–20 h because of the E1S reservoir, not because E2 itself persists. This is a critical distinction for model architecture.

### B8. NAD⁺/NADH cytoplasmic ratio ~700 in hepatocyte [C1]

Verdict: **Confirmed (within historical range).**

Williamson, Lund, Krebs 1967 (*Biochem J* 103:514): cytosolic free [NAD⁺]/[NADH] in rat liver = **~725** under aerobic, fed conditions, inferred from the lactate/pyruvate ratio and the LDH equilibrium. Later studies have given a range of 200–1000 depending on metabolic state (fed/fasted, hypoxia changes this dramatically). 700 is the canonical fed-state value.

### B9. Keq′(E2 + NAD⁺ ⇌ E1 + NADH + H⁺) ≈ 0.016 at standard conditions [C1]

Verdict: **? Genuinely uncertain — depends on an E°' value that is not robustly measured.**

The previous AI back-calculated this from:
- E°'(NAD⁺/NADH) = −0.320 V (standard tabulated value)
- E°'(E1/E2) ≈ −0.265 V (this is the load-bearing number)
- ΔE° = +0.055 V → ΔG°' ≈ −10.6 kJ/mol → Keq ≈ 0.016 (because the AI wrote the reaction in the oxidative direction, ΔG° has to flip sign; the algebra would be `Keq = exp(−ΔG/RT)` with the correct convention)

**The E°' for E1/E2 of −0.265 V is not a directly measured value.** I cannot find a primary electrochemical measurement of the E1/E2 redox couple. The number that propagates in textbooks for "secondary alcohol/ketone" redox is around −0.16 to −0.20 V at pH 7. The AI's −0.265 V appears to be either an estimate-from-similar-steroids (a few papers cite values around this for cortisol/cortisone) or an inferred number to match equilibrium observations. So the Keq′ ≈ 0.016 derivation is *plausible but the underlying E°' is not rigorously known*.

**Better approach:** Use the *empirical* in-cell E1/E2 ratio (oral E2 → plasma E1/E2 = 3–5) as the input, not the thermodynamic chain. The AI's chain is a clever physical-chemistry argument but it's stacking two uncertain numbers (cytosolic [NAD⁺]/[NADH] = 700 in fed state and E°' = −0.265 V) to reach a third (cytoplasmic [E1]/[E2] = 11).

### B10. Cytoplasmic [E1]/[E2] ≈ 11 (92% E1) in hepatocyte under HSD17B2 [C1]

Verdict: **? Uncertain, probably qualitatively right but the precise number is shaky.**

Building on B8 and B9: with cytoplasmic NAD⁺/NADH = 700 and Keq = 0.016, the ratio comes out [E1]/[E2] = Keq × ([NAD⁺]/[NADH]) = 0.016 × 700 ≈ 11. The arithmetic is correct conditional on the inputs.

**However:**
- 17β-HSD2 is membrane-bound (ER membrane), not freely cytosolic, so it sees an unknown local [NAD⁺]/[NADH] that need not equal the bulk cytosolic value (mitochondrial [NAD⁺]/[NADH] is ~7–10, dramatically lower).
- Empirically, plasma E1/E2 with steady-state oral E2 is ~3–5, not 11. The hepatocyte interior may indeed be higher than plasma (because E1 leaves the hepatocyte and gets further conjugated to E1S, removing it from the equilibrium), but no one has directly measured the intracellular E1/E2 ratio in human hepatocytes.
- Empirical "oral E2 → liver portal E1/E2 ≈ 4–10" estimates exist (Kuhl 2005 ranges); 11 is within that envelope but at the high end.

Pierce-style estimate is reasonable; cite it as "model assumes ~10× oxidative bias, consistent with thermodynamic argument" and don't lean on it for precise predictions.

### B11. Volume of distribution E2: ~30–50 L claimed; apparent Vd 127 L in v3 [C2]

Verdict: **C/≈** — both ranges are quoted in the literature. The "true" Vd of free E2 is ~80 L (since E2 is highly protein-bound and concentrates in lipid-rich tissues). Apparent Vd of *total* E2 (bound + free) ≈ 25–50 L is the more commonly cited textbook value. The 127 L apparent Vd in v3 would correspond to the "free-E2 Vd" interpretation. Both are defensible; need to be internally consistent on free vs total throughout the model. (Wikipedia "Pharmacokinetics of estradiol" cites ~80 L; Düsterberg 1985 gives ~50 L apparent.)

### B12. Hepatic clearance / first-pass: Back 1982 portal-vein 0.44 fraction [C2]

Verdict: **? Couldn't independently verify the exact 0.44 figure.** Back et al. (1982 / 1981, *Contraception*) is widely cited for distinguishing gut-wall vs hepatic first-pass for ethinylestradiol (not E2 specifically), with gut-wall contributing ~40–50% of total first-pass. The 0.44 fraction *for E2 portal extraction* may be from a different source or extrapolation. Treat as plausible but check the actual paper before using.

### B13. Plasma E1S concentration is 10–25× free E1 + E2 (reservoir) [C2]

Verdict: **Confirmed.** Multiple sources: Wikipedia ("E1S levels are 10 to 15× higher than E1 in women"), Ruder 1972, and across pregnancy E1S can be 50,000 pg/mL while E1+E2 sum is ~25,000 pg/mL (2× ratio in late pregnancy, lower because absolute E2 is so high). In ordinary cycling or oral E2 conditions, 10–25× is right.

**Section B overall:** B1, B6, B11 are subtly wrong/oversimplified. B4 and B9–B10 are the most fragile load-bearing claims, built on chains of inference rather than direct measurement.

---

## C. Empirical anchor concentrations

### C1. Cycling follicular: free E2 ~50, E1 ~50, E1S ~960 pg/mL [C1]

Verdict: **C/≈** — close to standard reference ranges.

- E2 follicular: early-follicular 20–60, mid-follicular 50–150. 50 is reasonable for early-mid follicular.
- E1 follicular: 30–100 pg/mL. 50 is OK.
- E1S follicular: literature (Roberts 1980; transfemscience compilations) gives 700–2,500 pg/mL, with average ~1,000–1,500. 960 is fine.

Note: "free E2 = 50" must mean *total E2 = 50* (free fraction would be ~1 pg/mL). This is a confusing notation in the original. The anchor is for *measured serum E2* which is total.

### C2. Oral E2 1 mg/d steady-state: E2 ~35, E1 ~250, E1S ~2560 pg/mL [C1]

Verdict: **Confirmed.** Stanczyk 2013 and Kuhl 2005 give very similar numbers: oral 1 mg micronized E2 in PMP women → mean E2 30–50 pg/mL, mean E1 150–300 pg/mL, E1S ~2,000–3,000 pg/mL. The ratio E1/E2 ≈ 5–7 is the hallmark of oral E2. Anchor is good.

### C3. Transdermal 50 μg/d steady-state: E2 ~60, E1 ~50, E1S ~600 pg/mL [C1]

Verdict: **Confirmed.** Published patch data: 50 μg/d patch → E2 ~30–50 pg/mL (some studies report 49–60), E1 ~40–60, E1S ~500–900. AI's anchor is on the high side for E2 but within the range. E1/E2 ≈ 1 is the hallmark of transdermal, in contrast to oral. (Kuhl 2005; Powers et al. 1985 *Am J Obstet Gynecol*.)

### C4. IM E2-valerate 5 mg q5d steady: E2 ~200, E1 ~200, E1S ~1500 pg/mL [C1]

Verdict: **≈ Approximately right but the dosing schedule "q5d" is unusual.**

Single 5 mg IM EV produces peak E2 ~400–650 pg/mL at day 2, declining to baseline by day 7–8 (Wikipedia "Estradiol valerate"; Düsterberg 1985). Steady state on q5d would oscillate roughly between trough ~50 and peak ~400+ pg/mL. A "time-averaged" steady-state E2 around 200 pg/mL is plausible. E1 ~ E2 at steady state on IM is consistent with literature.

E1S on IM is reported as 1,500–2,500 pg/mL (transfemscience injectable meta-analysis). 1,500 is on the low end but plausible.

### C5. Pregnancy term: E2 ~20,000, E1 ~7,000, E1S ~50,000 pg/mL [C1]

Verdict: **Confirmed.** Standard late-pregnancy values (Wikipedia, Perinatology.com reference tables):
- E2 third trimester: 10,000–40,000 pg/mL → 20,000 is mid-range, OK
- E1 third trimester: ~5,000–8,000 pg/mL → 7,000 OK
- E1S third trimester: 100,000 ± 22,000 ng/mL was the Wikipedia number — wait, that should be pg/mL. Wikipedia for E1S third trimester: 105 ± 22 ng/mL = 105,000 ± 22,000 pg/mL. So the AI's 50,000 pg/mL is **about 2× too low**. The correct order-of-magnitude is closer to 100,000 pg/mL.

So C5 is slightly off on E1S — should be ~100,000 pg/mL not 50,000.

### C6. Sublingual peak ~144 pg/mL E2; bioavailability ~25%; T_max ~30 min [C2]

Verdict: **≈ Mostly right.**

Cirrincione et al. 2021 (*Endocrine Practice*) and the transfemscience review:
- 1 mg sublingual → Cmax ≈ 144 pg/mL, vs 35 pg/mL for 1 mg oral — same study (so ~4× ratio).
- Tmax: 1–2 hours, not 30 min. "30 min" might be from a different formulation or might be the *first* detectable rise; the actual peak is at ~1–2 h. Slight error.
- Bioavailability "25%": no direct human absolute-bioavailability number for sublingual E2 exists. The 25% figure is inferred from monkey data (sublingual ~10% vs oral ~5% in monkeys, then scaled), or from the ~4× AUC ratio vs oral. Calling it "25%" with confidence is overstating what's known.

### C7. Sublingual produces higher E1 than transdermal/injectable (Cirrincione 2021) [C2]

Verdict: **Confirmed.** Cirrincione 2021 and Bar 2024 both show sublingual generates higher E1 and E1S than transdermal — markers of meaningful hepatic estrogenic exposure. The AI's revised position (sublingual ≈ oral on hepatic side, not "low hepatic exposure" as one might naively assume) is well-supported. The peak is brief and tall, so AUC of E2 is high but the system sees rapid hepatic conjugation following each dose.

But: it does *not* follow that sublingual is *equivalent* to oral on hepatic effect, because the kinetics are very different (sharp spike vs sustained portal exposure). Bar 2024's pilot finding (decrease in free protein S) is consistent with intermediate-to-high hepatic estrogenic effect — but on the order of 1 case-series, not a definitive epidemiologic answer.

### C8. Postmenopausal E1 ~30–50 pg/mL from adipose aromatase, ~80 μg/day [C2]

Verdict: **C/≈.**

- Postmenopausal E1: typical 10–50 pg/mL, lean women lower (10–30), obese women higher (40–100). "30–50" is on the high side for lean PMP but right for average-BMI. (Wikipedia "Estrone"; Burger 2002.)
- Production rate "~80 μg/day": Wikipedia "Estrone sulfate" table gives men 80 μg/d, premenopausal follicular 100 μg/d, luteal 180 μg/d, **postmenopausal** ~40 μg/d (Longcope 1986; Grodin 1973). So the AI's "80 μg/day for PMP" is likely **2× too high** — the true number is around 40 μg/d (or higher in obese women).

Note: one of my web searches returned "100 mg/day" — that's clearly a Google-summary error; 100 mg/d would be 1,000,000 μg/d, absurd. The right number is 40–100 μg/d depending on adiposity.

**Section C overall:** Anchor concentrations are mostly within 2× of published values. The two notable errors are E1S in pregnancy (~2× low) and the postmenopausal E1 production rate (~2× high).

---

## D. VTE-related claims

### D1. SHBG baseline female ~40–80 nM; Kd for E2 ~1 nM [C2]

Verdict:
- **SHBG baseline:** ≈ — premenopausal women 30–100 nM, postmenopausal 20–120 nM (PMC10182837). 40–80 is fine as a working baseline.
- **Kd for E2 ≈ 1 nM:** **W (wrong by ~10×).** Hammond and others (Avvakumov 2010; PMC8144348) report SHBG Kd for E2 of ~10–30 nM (vs ~1 nM for DHT, ~3 nM for testosterone). The AI may have confused E2 Kd with DHT Kd. This matters: at SHBG = 50 nM, with Kd_E2 = 1 nM, ~98% of E2 would be SHBG-bound (contradicting the 38% empirical). With Kd = 10–30 nM, the binding distribution actually works out to the observed ~38% SHBG-bound at total E2 ~100 pg/mL.

**Correct:** SHBG Kd for E2 is ~10–30 nM, not 1 nM.

### D2. SHBG in pregnancy: 5–10× rise [C2]

Verdict: **Confirmed.** Multiple sources (Anderson 1974; Wikipedia "Sex hormone-binding globulin"). 5–10× is the standard range.

### D3. SHBG with oral E2 HT: 67–171% rise (Lindberg 2003) [C2]

Verdict: **C, but citation year is wrong.** The Lindberg et al. paper is **2005** (*JCEM* 90: 3431), not 2003: "Effects of Oral and Transdermal Estradiol Administration on Levels of SHBG in Postmenopausal Women." It reports SHBG +67–171% on oral but no change on transdermal. The numbers are right; the year is one of those AI-misremembered details.

### D4. SHBG with EE: +300% [C2]

Verdict: **C/≈.** EE causes SHBG increases on the order of 200–400% (3–5×), depending on dose. (Wiegratz 2003 *Contraception*; van Rooijen 2004.) 300% is mid-range.

### D5. Pregnancy fibrinogen 2× rise (300→500 mg/dL); Factor VII up to 10× [C2]

Verdict:
- **Fibrinogen 2×:** **Confirmed.** From ~200–400 mg/dL non-pregnant → 350–650 mg/dL late pregnancy. Roughly 2× rise. (Hellgren 2003; Brenner 2004.)
- **Factor VII up to 10×:** **W — overstated.** Factor VII rises to ~150–250% of non-pregnant baseline at term (i.e., 1.5–2.5× increase), not 10×. The largest pregnancy factor rise is *Factor VIII*, which can reach 200–500% (2–5×). vWF rises similarly. "Factor VII 10×" is wrong — the AI may be confusing fold-change in different factors. (Dalaker 1986 *BJOG*; PMC7273490.)

### D6. VTE risk: 3× antepartum, 9× third trimester, 20–25× postpartum 6 weeks [C2]

Verdict: **≈ Approximately right; postpartum number is on the low end.**

Modern meta-analyses (e.g., Sultan 2013 *BMJ*; PMC3726432):
- Overall pregnancy: ~5× vs non-pregnant baseline.
- First/second trimester: ~2–4× (this is the AI's "3× antepartum")
- Third trimester: **9× (OR 8.8, 95% CI 4.5–17.3)** — matches AI.
- First 6 weeks postpartum: **20–84×** depending on study. The AI's "20–25×" is at the low end. The Heit 2005 study and Pomp 2008 MEGA study report OR ~84 for the first 6 weeks. So "20–25×" is likely 2–3× too low; better to say "20–80× depending on study, with peak in weeks 1–2."

### D7. Oral E2 HT VTE RR ~2× (Canonico ESTHER 2007) [C2]

Verdict: **C (slightly conservative).** ESTHER 2007 (Canonico, *Circulation* 115: 840): OR for oral estrogen = **4.5 (95% CI 2.6–7.7)** for VTE. So "2×" is actually conservative — ESTHER itself reports 4.5×. Pooled meta-analyses (Scarabin/Mohammed 2015 JCEM, PMID 26544651) give RR = **1.48 (1.39–1.58)** for oral estrogen among broader populations. The discrepancy is because ESTHER is a case-control of *symptomatic* VTE, while meta-analyses include lower-risk cohorts.

So depending on which evidence base, RR is anywhere from 1.5× (meta-analysis) to 4.5× (ESTHER case-control). "2×" is a reasonable mid-estimate.

### D8. Transdermal E2 HT VTE RR ~1.0–1.2× (baseline) [C2]

Verdict: **Confirmed.** Multiple sources:
- ESTHER 2007: OR 1.1 (0.7–1.7) for transdermal — not significantly different from no HT.
- Scarabin/Mohammed 2015 meta-analysis: RR 0.97 (0.87–1.09) for transdermal vs no HT.
- IMS 2021 review reaffirms.

### D9. Sublingual E2 may have similar VTE-risk profile as oral (Bar 2024) [C2]

Verdict: **? Uncertain — Bar 2024 is preliminary, but mechanistic data point to higher-than-transdermal exposure.**

The Bar 2024 abstract (ECE 2024; endocrine-abstracts.org ea0099ep592) reports a clinically significant decrease in **free protein S** under low-dose sublingual E2 in treatment-naïve trans women — consistent with meaningful hepatic estrogenic exposure. But:
- No epidemiologic data on actual VTE incidence with sublingual.
- The PK profile (sharp spikes, modest total AUC) differs from oral (sustained portal exposure).
- Markers like SHBG induction with sublingual are intermediate between oral and transdermal in most studies.

The AI's revised position ("sublingual ≈ oral") is **probably overcorrected.** The honest answer is "sublingual is likely *not* baseline-safe like transdermal; the hepatic effect is real and proportional to AUC, but whether RR matches oral (4.5×) or sits at, say, 2×, is unknown." A 1-case drug-induced liver injury report (PMC11455116) and the Bar protein S finding both warn against treating sublingual as "low hepatic impact."

**Better summary:** "Sublingual likely has *intermediate-to-high* hepatic estrogenic effect, with the peak-driven AUC producing measurable SHBG rise, protein S drop, and probably elevated VTE risk relative to transdermal — but lower or comparable to oral. No definitive RR exists."

**Section D overall:** D1 (SHBG-E2 Kd) and D5 (Factor VII 10×) are factually wrong. D6 postpartum number is low. D7 ESTHER number understates the original effect size. D9 is genuinely uncertain.

---

## Summary table

| Claim | Verdict | Correction (if needed) |
|---|---|---|
| A1–A7 (enzymology) | Confirmed | — |
| B1: Oral E2 F = 5% | ≈ | Right for free-E2 AUC; "5%" is misleading without specifying compartment |
| B2: EE F = 40–50% | Confirmed | — |
| B3: Free E2 = 2% | Confirmed | — |
| B4: Hepatic ER Kd 0.25 nM (Yager) | ≈ | Number is in Yager 1989, but it's rat liver, and the Lindberg-Hill-curve chain is invented |
| B5: Hill coefficient ~1.0–1.2 | Uncertain | Reasonable default, not directly measured |
| B6: MCR_E2 = 1400 | Confirmed | MCR_E1 should be ~2700–3200, not 2200 |
| B7: t½ free E2 1–2 h, E1S 10–12 h | Confirmed | — |
| B8: NAD+/NADH = 700 | Confirmed | Fed-state hepatocyte cytosol; varies 200–1000 |
| B9: Keq′ ≈ 0.016 | Uncertain | Relies on E°'(E1/E2) ≈ −0.265 V, which is not robustly measured |
| B10: Cytoplasmic [E1]/[E2] ≈ 11 | Uncertain | Arithmetic correct given B8, B9; empirical plasma ratio is 3–5; intracellular not measured |
| B11: Vd 30–50 L or 127 L | Both quoted | Depends on free vs total |
| B12: Back 1982 0.44 fraction | Uncertain | Couldn't verify exact figure |
| B13: E1S = 10–25× free E1+E2 | Confirmed | — |
| C1: cycling follicular anchors | Confirmed | — |
| C2: oral 1 mg anchors | Confirmed | — |
| C3: TD 50 μg anchors | Confirmed | E2 60 is on high end |
| C4: IM EV 5 mg anchors | ≈ | Time-averaged values reasonable; q5d schedule unusual |
| C5: pregnancy anchors | ≈ | E1S should be ~100,000 pg/mL, not 50,000 |
| C6: SL peak 144, F 25%, Tmax 30 min | ≈ | Tmax 1–2 h not 30 min; F=25% inferred not measured |
| C7: SL > TD on E1 (Cirrincione 2021) | Confirmed | — |
| C8: PMP E1 30–50 pg/mL, ~80 μg/d | ≈ | Production rate is ~40 μg/d not 80; obese can reach 80+ |
| D1: SHBG baseline 40–80 nM; Kd E2 = 1 nM | Mixed | SHBG range right; Kd_E2 wrong — actually 10–30 nM |
| D2: SHBG pregnancy 5–10× | Confirmed | — |
| D3: Oral E2 HT SHBG +67–171% (Lindberg) | Confirmed | Year is 2005, not 2003 |
| D4: EE SHBG +300% | Confirmed | — |
| D5: Pregnancy fibrinogen 2×, F VII 10× | Mixed | Fibrinogen confirmed; Factor VII wrong (~1.5–2.5×, not 10×) |
| D6: pregnancy VTE 3×/9×/20–25× | ≈ | Postpartum 6-week RR is actually 20–80×; AI's 20–25× is low |
| D7: Oral E2 HT RR 2× (Canonico) | ≈ | ESTHER reports 4.5×; meta-analyses 1.5×; "2×" is a mid-estimate, not literally from Canonico |
| D8: Transdermal RR 1.0–1.2× | Confirmed | — |
| D9: Sublingual ≈ oral VTE risk | Uncertain/Overcorrected | Likely intermediate-to-high, not equal to oral; no RR data |

---

## Load-bearing assessment

**Textbook-solid (lean on these):**
- All of section A (enzymology)
- B2, B3, B7, B13 (EE bioavailability, free fraction, half-lives, E1S reservoir ratio)
- C1, C2 (cycling and oral anchors)
- D2, D3, D8 (pregnancy SHBG, Lindberg, transdermal baseline)

**Approximately right but oversimplified — recheck before quoting precise numbers:**
- B1 (oral E2 F=5% — specify which compartment)
- B6 (MCRs — use BSA-normalized values for precision)
- B11 (Vd — specify free vs total)
- C3, C4, C5 (steady-state anchors — within 2× generally)
- D6, D7 (VTE numbers — cite specific study, not a vague "RR ~2")

**Handwavy / load-bearing chains that should not be relied on for quantitative prediction:**
- B4 + B5 (hepatic ER Kd → Hill curve → SHBG induction). The Yager 0.25 nM number exists, but no one has actually fitted a Hill curve from this to SHBG dose-response in humans. The v3 model essentially invents this chain.
- B9 + B10 (NAD+/NADH ratio → Keq → cytoplasmic E1/E2). The thermodynamic argument is clever, but it depends on E°'(E1/E2) = −0.265 V, which I could not find as a directly measured value. The output (intracellular E1/E2 ~11) is qualitatively consistent with plasma E1/E2 ~3–5 but the precise number is not validated.

**Outright errors:**
- D1: SHBG Kd for E2 is **10–30 nM**, not 1 nM. (Confusing E2 with DHT.)
- D5: Pregnancy Factor VII rises ~**1.5–2.5×**, not 10×. (Probably confused with Factor VIII or vWF.)
- C5: Pregnancy E1S is ~**100,000 pg/mL**, not 50,000.
- C8: PMP E1 production rate is ~**40 μg/d**, not 80.
- C6: Sublingual Tmax is ~**1–2 h**, not 30 min.

**Bottom line:** The previous AI got most textbook PK right but made a handful of order-of-magnitude errors in supporting claims (especially around SHBG biochemistry and pregnancy coagulation), and built two key load-bearing model components (B4+B5 hepatic ER curve, B9+B10 cytoplasmic E1/E2 thermodynamics) on chains of inference where individual links are uncertain. The model anchors in C are within 2× of literature — adequate for an order-of-magnitude model, not for precision prediction.

---

## Sources

- Wikipedia, "Pharmacokinetics of estradiol" — large compilation, citations to Kuhl 2005, Stanczyk 2013, Düsterberg 1985
- Wikipedia, "Estrone sulfate" — citations to Ruder 1972, Longcope 1986
- Kuhl H. 2005. Pharmacology of estrogens and progestogens: influence of different routes of administration. *Climacteric* 8 Suppl 1:3–63. PMID 16112947
- Stanczyk FZ et al. 2013. Ethinyl estradiol and 17β-estradiol in combined oral contraceptives. *Contraception*.
- Longcope C, Layne DS, Tait JF. 1968. Metabolic clearance rates and interconversions of estrone and 17β-estradiol. *J Clin Invest* 47:93. PMC297151
- Yager JD et al. 1989. Changes in estrogen receptor, DNA ploidy, and estrogen metabolism in rat hepatocytes. *Cancer Res* 49:6605. PMID 2573415
- Canonico M et al. 2007. Hormone therapy and venous thromboembolism: the ESTHER study. *Circulation* 115:840. PMID 17309934
- Mohammed K et al. 2015. Oral vs transdermal estrogen therapy and vascular events: meta-analysis. *JCEM* 100:4012. PMID 26544651
- Lindberg MK et al. 2005 (NOT 2003). Effects of oral and transdermal E2 on SHBG in PMP women. *JCEM* 90:3431.
- Williamson DH, Lund P, Krebs HA. 1967. The redox state of free NAD/NADH in cytoplasm of rat liver. *Biochem J* 103:514.
- Sultan AA et al. 2013. VTE incidence in pregnancy/postpartum (meta-analysis). PMC3726432
- Cirrincione LR et al. 2021. PK of sublingual vs oral estradiol in transgender women. *Endocrine Practice*. PMID 34781041
- Bar et al. 2024. Low-dose sublingual E2 decreases protein S (preliminary). ECE 2024 abstract.
- transfemscience.org articles: "Sublingual E2 as an Alternative to Oral E2 in Transfeminine People"; "Approximate Comparable Dosages of Estradiol"; "Injectable E2 Meta-Analysis."