The Evidence Moved. The Recommendation Should Move With It.

The Short Version

• The largest study yet of testosterone in healthy women, 5,323 women aged 19 to 59, finds a steady decline of about 1.4 percent a year and a spread that widens with age. The middle 99 percent of healthy women fall between roughly 3 and 92 ng/dL. A single flat reference range fits older women worst.
• Women with hirsutism ran higher, but the two groups overlapped at every age. One testosterone value cannot tell you whether a woman has androgen excess. The authors want labs to report where she sits among healthy women her age, not an H or an L.
• In telehealth with clinical decision support, a paper like this does not have to wait years to reach practice. We read it against our own rules and are updating them. The updated rule reaches every clinician at once, with the citation attached. A person reads the paper first.

The reference range printed next to a woman's testosterone result often comes from a very small group of people.

Roche is one of the leading immunoassay suppliers in the UK. Its female testosterone range came from 149 women, split into two age groups. One group covered everyone from 20 to 49.

Sit with that for a second. A 21-year-old and a 48-year-old get measured against the same line.

The US is not doing better. A review of American lab practice, cited in the paper, found that only 10 percent of labs running reproductive hormones had done their own reference range study. The ones that did used between 20 and 200 people.

In June, the Journal of Endocrinological Investigation published a study that takes a real swing at this. Researchers from Hertility Health, University College London and Imperial College London built an age-continuous reference range for total testosterone from 5,323 women. By the authors' count, it is the largest analysis of its kind.

I read it twice. Once as a pharmacist. Once as the person responsible for the clinical rules inside our decision support platform at Nimbus Healthcare. The second read is the reason for this post.

What they did

The starting pool was 30,453 UK women who bought at-home hormone testing between September 2020 and August 2025. Then the exclusions. Normal BMI. Regular cycles. No PCOS or other reproductive conditions. No hormonal contraception. No self-reported hirsutism. No diabetes. That left 5,323 women aged 19 to 59.

Every sample was collected the same way. Finger-prick capillary blood, day 3 of the cycle, after at least an 8-hour fast. Every sample ran on the same immunoassay, the Roche Cobas e801, in one accredited lab.

Then the statistics. The team fit six candidate models and picked the best one by Bayesian Information Criterion. The winner says log testosterone follows a slightly skewed normal curve at any given age. The average falls in a straight line with age. The spread grows in a straight line with age.

In plain English: women lose testosterone steadily across adult life, and they get more different from one another as they go.

Four findings worth your time

It is a slope, not a cliff. The model puts the decline at 1.4 percent a year in total testosterone. Run that rate forward and the typical woman at 50 sits about a third below the typical woman at 18. There is no sudden drop at a birthday or at a last period. The authors note other work showing the decline is gradual through the menopausal transition, too.

The spread widens with age. This one matters more than it sounds. A single range drawn across all ages is least accurate for older women, because that is where women differ most from one another. The flat range is weakest exactly where many hormone conversations happen.

The healthy range is wide. The middle 99 percent of the reference cohort ran from 0.1 to 3.2 nmol/L, about 3 to 92 ng/dL. The most common value at age 18 was 1.45 nmol/L, about 42 ng/dL.

Then there were five women. Out of 5,323, five had values between 13 and 34.5 nmol/L. That is roughly 375 to 1,000 ng/dL, which is male territory. None reported a reproductive condition, and their gonadotropins showed no sign of suppression. The authors list unreported testosterone use, a rare androgen-secreting tumor, or an assay anomaly as possible explanations. They kept them in the model because five women out of 5,323 did not move it.

Hirsutism ran higher, and it separated less than you would think. The team ran the same model on 2,338 women who reported hirsutism but met every other criterion. At 18, their most common value was 1.80 nmol/L against 1.45 in the reference group. That is about 24 percent higher, or roughly 52 versus 42 ng/dL. The gap narrows with age. And the authors say plainly that the two groups overlapped significantly at every age.

That overlap is the clinical point. A woman at 52 ng/dL might look like the typical hirsute 18-year-old. She might also be a perfectly healthy one. The number alone cannot tell you which. Her history, her exam and the rest of her panel can.

Report the percentile, not the flag

The best idea in this paper is not a number. It is a reporting format.

The authors make an argument I agree with completely. Any cutoff takes two nearly identical results that happen to land on either side of a line and sorts them into different boxes. Their alternative is simple. Tell the clinician where this woman sits among healthy women her age. "65th percentile for a 45-year-old" tells you something "within normal limits" never will.

I made the same argument about men in August, when I wrote that low testosterone is a flag, not a prescriptionLow Testosterone Is a Flag, Not a PrescriptionA pharmacist's read of the new Lancet Healthy Longevity meta-analysis: what 24,510 men actually tell you about testosterone and cancer, and the two findings the headlines are getting backwards.. An H or an L next to a hormone result throws away most of what the lab knows. This paper hands labs a model to do better for women. The formula is published. The labs have to adopt it.

What this paper cannot tell you

One more line in the paper is aimed at people like me. The introduction notes that testosterone creams and gels are now widely available online and through private companies, especially for women in perimenopause and menopause. It also notes that outside sexual symptoms, efficacy data are thin, and long-term safety data are thin too.

Nimbus is one of those private companies. I would rather say that here than have you find it in the disclosure.

The part I care about most

A 2011 review in the Journal of the Royal Society of Medicine took its title from the most quoted number in translational research: 17 years from research to routine practice. The authors were honest that the estimate is rough. Nobody argues that the lag is short.

Think about how a paper like this one reaches a patient today. The lab range changes when the manufacturer revises its insert. The textbook changes in the next edition. You change when you happen to read the paper. One clinician at a time.

Here is the beauty of running clinical decision support in telehealth. The recommendations our clinicians see come from a rule set, not from anyone's memory. Every rule has a version, an owner and a source. When the evidence moves, we change the rule once. Every recommendation after that carries the new logic and the citation behind it, to every clinician, the next time they open a chart.

That is what real time means here. It does not mean a model read a paper and changed its mind. A person read the paper. A clinician signed off on the change. Distribution took minutes instead of years.

What this paper changes in our rules

When this paper crossed my desk, we did what we do with any paper that touches a protocol. We read it against the rules in IntelliHealth™Building IntelliHealth™: AI Clinical Decision Engine at ScaleTechnical deep-dive into our clinical decision engine serving 120,000+ patients a month., our clinical decision support engine, line by line.

The target held up. Our rule keeps women on testosterone therapy between 20 and 50 ng/dL. That sits inside the healthy range in this paper, and the ceiling is a little above the typical healthy 18-year-old. It also matches the 2019 Global Consensus Position Statement on testosterone therapy for women, endorsed by eleven societies including the Endocrine Society and the North American Menopause Society. That statement says treatment should aim for concentrations in the premenopausal physiological range.

What the paper sharpened is how the rules around that target should work. Here is what came out of the review.

1. Women's testosterone gets its own profile. Results are read against female values. The direction is "stay within range," not "higher is better." For a man on treatment, rising is usually the goal. For a woman, rising past her ceiling is the thing you want to catch, and this paper ties higher values to hirsutism. The profile also leaves room for the expected rise during titration.

2. The ceiling is a check, not a sentence. A ceiling written in prose is easy to miss. A measured serum value above 50 ng/dL means the system will not suggest a dose increase, and it tells the clinician why. Code runs the check. The language model only explains it.

3. Serum drives the ceiling. The target is a serum number. A saliva result can inform the conversation. It should not move the ceiling.

4. The recommendation shows its source. When IntelliHealth advises holding a dose, the clinician sees the rule, the lab value and the papers behind it, with links. If they disagree, they can read exactly what we read.

Every IntelliHealth recommendation still goes to a clinician who makes the call. The rules decide what the clinician sees first. They do not decide for them.

What we are not changing yet

The tempting move is an age-adjusted ceiling. The paper practically asks for one. By my arithmetic from the published model, the typical healthy woman at 50 sits somewhere around 25 to 30 ng/dL. That is about half our ceiling.

We are not doing that yet, on purpose.

The paper's numbers come from an immunoassay in women who still have regular cycles. Most of the women in our hormone programs are in perimenopause or menopause, and their serum testosterone is measured by LC-MS/MS. Porting this curve straight across would mean comparing our patients to a population they are not in, on an assay we do not use.

So the percentile idea goes on the roadmap with a condition attached: a reference model matched to our assay and our patients. Fast distribution does not excuse a sloppy rule. It makes a careful one more important, because a bad rule reaches everyone at once too.

What I would do with this on Monday

1. Read a woman's testosterone against her age. A result of 26 ng/dL means something different at 25 than it does at 55.

2. Ask your lab where its female range came from. How many women, what ages, which assay. If the answer is a manufacturer insert built on a small group, treat the flag as a rough guide.

3. Do not diagnose androgen excess from one number. The hirsute and healthy groups overlapped at every age. History, exam and the rest of the panel do the separating.

4. If you treat, keep her inside the premenopausal physiological range, measured in serum. That is the consensus position, and nothing in this paper argues with it. Recheck after you start and after every dose change.

5. If you use any decision support, ask whether its testosterone logic knows the patient is a woman. Men and women need different floors, different directions and different ceilings.

A good paper used to take years to reach the chart. It does not have to anymore.

When the next one lands, how long will it take to reach the recommendations your clinicians see?

Key sources: Wainwright E, Getreu N, Crawford L, et al. "Redefining testosterone reference ranges for adult females." J Endocrinol Invest. 2026;49(9):2339-2349. doi:10.1007/s40618-026-02929-w (PMID 42340622, open access, PMC13498515). Davis SR, Baber R, Panay N, et al. "Global Consensus Position Statement on the Use of Testosterone Therapy for Women." J Clin Endocrinol Metab. 2019;104(10):4660-4666. doi:10.1210/jc.2019-01603 (PMID 31498871). Morris ZS, Wooding S, Grant J. "The answer is 17 years, what is the question: understanding time lags in translational research." J R Soc Med. 2011;104(12):510-520. doi:10.1258/jrsm.2011.110180 (PMID 22179294). Unit conversions use 1 nmol/L = 28.84 ng/dL. The age-based typical values for women over 18 are my arithmetic from the published model, not figures from the paper's tables.

Disclosure: I am the founder and CEO of Nimbus Healthcare, which operates telehealth services and a compounding pharmacy, including women's hormone programs that may prescribe testosterone. That is a commercial interest in this topic, and you should weigh this piece accordingly. This is a reading of the published evidence, not medical advice, and it is not a substitute for a clinician who knows your history.