Fact-check · 25 July 2026

The race to build a continuous hormone monitor

Line-by-line verification of a ChatGPT-written article about wearable patches that measure estradiol and testosterone directly. Checked against PubMed, the EU CORDIS database, and the FDA device classification and 510(k) databases.

Verdict: zero fabricated sources, but the technology is consistently presented as more mature than it is.

Every paper, grant number and quotation checks out. The three load-bearing scientific claims are correct. That is unusual for an AI-written piece and worth saying plainly.

The failure mode is different: rat data reads as human data, one section claims uncertainty about facts that are published, a company's fundraising copy is treated as a specification, and the most clinically important point for women on hormone therapy is buried in a single clause.

What it got right

The five corrections

Materially wrong

1. Rat data is presented as human data

"A newly published microneedle platform expands the concept toward multiplexed measurement that includes testosterone."

The paper is real and the four-hormone panel (estradiol, progesterone, LH, testosterone) is real (Zhou et al. 2026). But it has three separate arms, and only one involved a human body:

  • Continuous four-hormone in vivo tracking: rats
  • Clinical validation: 45 banked human serum samples, ex vivo
  • Human on-body wear, 5 days: testosterone only

"A four-hormone continuous wearable has been demonstrated in humans" is not supported.

Materially wrong

2. The EstroPatch section claims uncertainty about published facts

"Low confidence: EstroPatch will use any specific sensing chemistry, microneedle design, hormone panel, wireless protocol, or commercialization pathway."

The grant details are correct. But the technical design is public. The CORDIS fact sheet states the project will perform "real-time, continuous monitoring of estrone, estradiol, and estriol levels in dermal interstitial fluid by integrating microneedles, electrochemical sensing, microfluidics, and wireless communication."

So three cells in the article's comparison table are wrong. The fluid is interstitial fluid, not "not publicly established." Skin penetration is minimally invasive microneedles, not "unknown." The panel is all three estrogens, not "exact analyte unclear." Grouping it with noninvasive sweat approaches is a category error. It is also a single postdoctoral fellowship of about €321,000, not a program.

Materially wrong

3. "Interstitial fluid may be more clinically interpretable than sweat"

Presented as a reasoned advantage. It is an untested assumption, and the direct evidence cuts against it. Where interstitial-fluid steroid data exists, it is cortisol measured by microdialysis, and it tracks plasma poorly:

  • In severe burns, there was no significant correlation between plasma cortisol (total or free) and interstitial fluid cortisol (Cohen 2009).
  • In 35 septic patients, correlation was only moderate with "considerable overlap" (Vassiliadi 2013).

Both are critical-illness populations, which is a real limitation. But the first direct sex-steroid partitioning data is one month old and in rats. The honest version: interstitial fluid is what glucose monitors already read, but for steroid hormones the blood-to-tissue relationship has barely begun to be measured.

Overstated

4. The 30-day wear claim is vendor copy, not a specification

The published on-body record for a hormone-sensing aptamer wearable is 5 days. The typical in vivo record for this sensor class is under 6 hours, and one human microneedle study ran 5 hours with "high rates of device failures" (Friedel 2023).

So 30 days is roughly six times the best published result and about a hundred times the typical one, for a molecule millions of times more dilute than glucose. It should be labeled as an unvalidated company claim.

There is a mechanism behind the skepticism the article missed entirely. Drift in these sensors has two phases: biofouling, and the act of measuring itself desorbing the sensing layer (Downs & Plaxco 2022). Sampling rate trades directly against sensor lifetime. A monitor that samples aggressively kills itself faster. That constraint has no equivalent in glucose monitoring.

Omission

5. The reproductive-hormone sensing was in artificial sweat

UCLA's own technology listing says the multiplexed estradiol, progesterone and LH sensor was demonstrated "in artificial human sweat." The article never mentions this.

The verified picture is narrower than the piece implies. The strongest confirmed on-body result is one hormone, cortisol, from 2022 (Wang et al., Science Advances). The three reproductive hormones exist so far in artificial sweat on a bench.

The arithmetic, worked

The article says estradiol is "millions of times" less concentrated than glucose. That is correct, and it leaves the better number on the table. Against fasting glucose at 5.0 mmol/L:

TargetConcentrationRatio to glucose
Estradiol, ovulatory peak~1,500 pmol/L3.3 million to 1
Estradiol, mid-follicular~300 pmol/L17 million to 1
Estradiol, postmenopausal~20 pmol/L250 million to 1
Estradiol in sweat~10 pmol/L500 million to 1
Total testosterone, women~1.0 nmol/L5 million to 1
Free testosterone, women~12 pmol/L420 million to 1

For the menopause use case the gap is a quarter of a billion. "Hundreds of millions of times more dilute" is both more accurate and more striking.

The point the article buried

Both sweat and interstitial fluid measure the free, unbound hormone fraction. The article mentions this in a passing clause. It deserves a section, because it is the most useful thing in the piece for anyone on hormone therapy.

Protein-bound steroid cannot cross into sweat, and the capillary wall largely excludes SHBG from interstitial fluid. Concrete numbers: interstitial cortisol ran 0.80 µg/dL against free plasma 1.7 and total 8.8, about 9% of total (Cohen 2009).

Why that matters: SHBG has a nearly five-fold reference range in women, 18 to 86 nmol/L (Braunstein 2011), and oral estrogen raises it while transdermal largely does not. A woman whose SHBG doubles shows a steady or rising total testosterone while her free testosterone falls, and free is what binds the receptor. Total is the number nearly every clinician orders.

So a wearable reading the free fraction is not a worse version of the standard blood test. It is a different and arguably better analyte. The catch is that there is no reference interval, no traceable standard and no decision threshold for continuously measured free hormone. Nobody knows what normal looks like for free estradiol sampled every fifteen minutes. That is a bigger blocker than sensor chemistry.

Other things worth knowing

The honest version

Two quotations also need repair. The UCLA technology-transfer "quote" is a clipped fragment with the subject clause silently dropped and mid-sentence capitalization added, and the EU funding "quotation" is a flattened spreadsheet row rather than anything a person wrote.

None of which means the underlying story is weak. The strongest sentence the evidence actually supports is this: a finger-worn Caltech patch has measured estradiol in women's sweat across real menstrual cycles and matched blood levels, and a separate microneedle platform has begun four-hormone work with human on-body testosterone data. That is genuinely impressive, and it does not need the inflation.

Three parallel verification threads run 25 July 2026 against PubMed, scite, Consensus, the EU CORDIS database, and the openFDA device classification and 510(k) endpoints. Every DOI was retrieved by tool call rather than recalled. Nothing here is medical advice, and no wearable discussed should replace clinician-directed hormone testing.