My total testosterone fell 27% but my free androgen index fell 64%. Which is real?

Baseline, before starting treatment: total testosterone 2.6 nmol/L, SHBG 22 nmol/L. After eight months and 16 kg of weight loss: total testosterone 1.9 nmol/L, SHBG 45 nmol/L. My clinic report gives a free androgen index for each and the second is dramatically lower, and I do not understand how a 27% fall in one input produces a 64% fall in the output.

I would like to work the arithmetic myself and understand which number describes what is happening to my body. And separately: my hirsutism is unchanged. Eight months, apparently large biochemical improvement, no visible difference at all. Either the biochemistry is not the thing that drives the hair, or the timescales are very different, and I do not know which.

Practical questions attached to that: which androgen measurements are actually worth ordering, does the assay matter, and does it matter which day of my cycle the sample is taken — given that my cycles are irregular so "day 3" is not always a well-defined concept.

Both numbers are real; they measure different things, and the free index is the one that corresponds to biological activity. The reason a 27% fall becomes a 64% fall is that the index is a ratio, and your denominator more than doubled.

The arithmetic

The free androgen index is total testosterone divided by SHBG, both in nmol/L, expressed as a percentage:

FAI = (total testosterone ÷ SHBG) × 100

Baseline:

• 2.6 ÷ 22 = 0.1182.
• × 100 = 11.8.

Follow-up:

• 1.9 ÷ 45 = 0.04222.
• × 100 = 4.22.

Change: (11.8 − 4.22) ÷ 11.8 = 0.642, a 64% fall, exactly as your report says.

Now decompose it, because the decomposition is the insight. The index is a quotient, so proportional changes multiply:

• Testosterone ratio: 1.9 ÷ 2.6 = 0.731, so a factor of 0.731 — a 27% fall.
• SHBG ratio: 45 ÷ 22 = 2.045, and since it is in the denominator it contributes a factor of 1 ÷ 2.045 = 0.489 — a 51% fall in the index.
• Combined: 0.731 × 0.489 = 0.357, so the index falls to 35.7% of baseline, a 64.3% fall.

So of the total effect, roughly a third comes from less testosterone being made and two thirds from more of it being bound and inactive. The SHBG change is the bigger lever, and that is not a quirk of your numbers — it is the general pattern, because SHBG is more responsive to insulin than testosterone production is.

For reference, an FAI above about 5 is generally taken as raised in women, with method-specific variation. Your 11.8 was clearly elevated and your 4.22 is not. That is a substantial biochemical change by any reading.

Why SHBG moves so much

SHBG is synthesised by the liver and its production is suppressed by insulin. It is one of the more sensitive available readouts of hepatic insulin exposure — arguably a better one than fasting insulin itself, since it integrates over time and has none of the assay-harmonisation problems. Doubling your SHBG is strong evidence that hepatic hyperinsulinaemia has substantially resolved.

Other things that move SHBG, worth knowing because they will confound future readings:

• Up: oestrogen, particularly oral oestrogen; hyperthyroidism; hepatic disease; substantial weight loss; pregnancy.
• Down: insulin; androgens; glucocorticoids; hypothyroidism; growth hormone.

The oral oestrogen point is the practically important one: combined oral contraceptives can raise SHBG two- to threefold on their own. If you start or stop one, your FAI will change dramatically with no change in androgen production, and any series spanning that change is uninterpretable. Record contraceptive status alongside every androgen result.

A better index if you can get it

FAI is a crude approximation. It ignores albumin, which binds testosterone weakly but is present in vast excess and carries a meaningful fraction of the total. The better calculation is calculated free testosterone by the Vermeulen method, which takes total testosterone, SHBG and albumin and solves the binding equilibrium properly. Most laboratories and several reputable online calculators will do it. It matters most when albumin is abnormal — nephrotic syndrome, liver disease, or the low-albumin states of significant illness — and it is otherwise a modest refinement.

Directly measured free testosterone by equilibrium dialysis is the reference method and is expensive and rarely available. Direct free testosterone by analogue immunoassay is unreliable at female concentrations and should be avoided; if a report gives a "free testosterone" without stating the method, assume it is the unreliable kind.

Why your hirsutism has not changed

Timescale, not mechanism. Androgens drive the conversion of vellus to terminal hair in androgen-sensitive follicles, and that conversion is not reversible on the timescale of the biochemistry.

Two separate lags:

• Existing terminal hairs continue their growth cycle. The anagen phase for facial and body terminal hair runs to several months, and follicles are asynchronous. Hairs already committed to terminal growth will complete their cycle regardless of what your androgens are now doing.
• Reversion of a terminal follicle to vellus is slow and incomplete. Some follicles do not revert at all. This is why the standard advice for any anti-androgen therapy is to expect a minimum of six months before assessing response, with twelve months being more realistic, and why mechanical or laser hair removal is usually part of the plan rather than an admission of pharmacological failure.

So eight months with no visible change is disappointing and not evidence that the biochemistry is irrelevant. The honest expectation is: reduced rate of new terminal hair conversion first, which you would notice as needing less frequent removal rather than as fewer hairs; then, over a year or more, some reduction in density in a proportion of women; and never a return to a pre-PCOS state from metabolic improvement alone.

Worth stating plainly because it is often oversold: no amount of weight loss reliably clears established hirsutism. Interventions that target the follicle directly, and pharmacological anti-androgens where appropriate, are a separate conversation with a clinician and are not made redundant by good metabolic numbers.

edited 12 Nov 2025 by sian_llewellyn — reworded for clarity after a comment

On which measurements to order and how to take them, since the pre-analytical decisions matter more than the choice of index.

The assay problem

This is the single biggest issue in female androgen measurement. Direct immunoassays for total testosterone were designed and validated at male concentrations, roughly ten to twenty times higher than female ones. At female concentrations they suffer from cross-reactivity with other steroids and from poor precision near the lower limit, and different platforms disagree with each other by amounts comparable to the difference you are trying to detect.

The reference method is liquid chromatography with tandem mass spectrometry, and it is worth asking which method your result came from. If your baseline and follow-up were run by different methods, the comparison is compromised. If both were run by the same immunoassay on the same platform, the comparison is defensible even if the absolute values are questionable — the same logic that applies to serial creatinine or serial A1c.

SHBG is measured by immunoassay and is comparatively well behaved, which is another reason the SHBG-driven part of your change is the part to trust most.

Timing

• Time of day. Testosterone has a diurnal rhythm, higher in the morning, though the amplitude in women is smaller than in men. Draw before 10:00 if practical, and keep it consistent.
• Cycle day. The conventional advice is the early follicular phase, days 2 to 5. Your objection is fair: if cycles are irregular or absent, the phase is undefined and the advice is unactionable. The practical resolution is that androgens and SHBG vary comparatively little across the cycle — unlike progesterone, LH and FSH, which vary enormously — so an androgen panel can be drawn on any day in an anovulatory woman. What you must not do is compare a follicular-phase sample against a luteal-phase one for the gonadotrophins.
• Record what you were taking. Combined oral contraceptives suppress ovarian androgen production and raise SHBG, so androgens measured on a COC tell you nothing about your underlying state. Most guidance suggests three months off hormonal contraception before assessing androgens, which is often impractical and is the reason a great many PCOS diagnoses rest on measurements taken under suppression.

What is worth ordering

• Total testosterone by LC-MS/MS, plus SHBG, plus albumin. From these you can compute calculated free testosterone, which is the best available index. Order them together on the same sample.
• DHEA-S, once. It indexes adrenal androgen production, identifies the minority with a predominantly adrenal source, and does not respond to insulin sensitisation, so there is no point repeating it serially.
• 17-hydroxyprogesterone, once, early follicular, to exclude non-classical congenital adrenal hyperplasia, which mimics PCOS and is managed differently. This is a genuine diagnostic test rather than a monitoring one.
• TSH and prolactin, once, to exclude the other common causes of oligomenorrhoea.
• A metabolic panel — HbA1c or an oral glucose tolerance test, and a lipid panel — because the metabolic risk in PCOS is the part with the longest-term consequences and it is frequently neglected in favour of the cosmetic and reproductive symptoms.

What is not worth ordering serially

• LH and FSH, or the LH:FSH ratio. Historically emphasised, now considered neither necessary nor sufficient for diagnosis, and both fluctuate substantially through the cycle. A raised ratio is supportive at most.
• Anti-Müllerian hormone. Elevated in PCOS and increasingly used as a proxy for antral follicle count where ultrasound is impractical. It is not a monitoring tool: it changes slowly, reflects the follicle pool rather than the metabolic state, and no threshold guides treatment.
• Fasting insulin and HOMA-IR. Unstandardised assays, large within-person variation, no decision threshold. If you want an integrated readout of hepatic insulin exposure, your SHBG is already giving you one for free.
• Repeat DHEA-S, repeat 17-OHP, repeat prolactin. Diagnostic tests, not monitoring tests.

The general shape: a wide diagnostic panel once, then a narrow monitoring panel — total testosterone, SHBG, HbA1c, lipids — at long intervals. This is the same baseline-versus-monitoring distinction that applies everywhere else, and PCOS panels violate it more often than most.

A short addition on menstrual outcomes, since they are the thing most women notice first and the reporting around them is loose.

What the small studies and the wider weight-loss literature support is an increase in menstrual frequency. What matters clinically is ovulation, and the two are not the same thing. Anovulatory bleeding occurs, and a woman can bleed on a schedule without ovulating, which matters both for fertility and for endometrial protection.

Confirming ovulation requires evidence, and the practical options in descending order of reliability:

• Mid-luteal serum progesterone, drawn about seven days before the expected next period. A value comfortably above the laboratory's ovulatory threshold confirms that a corpus luteum formed. The difficulty with irregular cycles is knowing when seven days before the next period is, which in practice means repeating the draw weekly across the second half of a cycle.
• Urinary LH testing detects the pre-ovulatory surge. In PCOS it produces frequent false positives because baseline LH is already elevated, which is a well-recognised and frustrating limitation.
• Basal body temperature charting is free, retrospective and noisy.

The endometrial point deserves stating because it is a real consequence of prolonged anovulation rather than a theoretical one: unopposed oestrogen without cyclical progesterone exposure is associated with endometrial hyperplasia and, over long periods, endometrial carcinoma. Restoring ovulatory cycles addresses that, and where cycles do not resolve, progestin exposure of some form — cyclical progestin, a combined contraceptive, or a levonorgestrel intrauterine system — is the usual protective strategy. Which of those applies is a clinical decision and one of the more consequential ones in PCOS care, and it is easy to lose sight of while focusing on weight and hair.

So if your cycles do become regular, that is genuinely good news on two axes and worth confirming rather than assuming. And if they do not, the endometrial question is the one to raise, ahead of the cosmetic ones.