Cortisol and ageing: what is proven and what is not

The questions people type into search are short: does cortisol age you, does cortisol increase with age, how do I lower it. Behind that one word sit three different questions, and the evidence behind them is of very different strength. If the topic is new to you, start with the basics — what the stress response is and how it works. This piece is about something else: what has actually been shown about cortisol and ageing, in whom, by what design — and where the shown part ends.

Three questions that usually get merged into one

1. Does cortisol change with age? A question about physiology. Answered by descriptions of daily profiles.

2. Is cortisol related to how people age and what they fall ill with? A question about observation. Answered by cohorts: measure people, then watch what happens to them.

3. Does lowering cortisol change anything? A question about intervention. Answered only by trials in which cortisol was deliberately lowered and an outcome was measured.

Three questions, three levels of evidence — and a "yes" to the first does not transfer to the third. Most of what gets written about cortisol in longevity sections performs exactly that transfer.

"Cortisol level" is not a single number

Before arguing whether it is high or low, you have to say what was measured. Secretion follows a daily rhythm, and time of day is the dominant source of spread: in field saliva sampling it accounts for something like 72% of the variance (Adam & Gunnar, PMID 16500024). The swing within one day in one person is larger than the differences between people sampled at different hours.

• Blood cortisol at a single point. Depends on the hour of the draw, on last night's sleep, and on anxiety about the procedure itself. As a description of chronic exposure it is close to useless. (One caveat: as a screen for adrenal insufficiency, morning serum cortisol remains a valid test — that is a different question and a different disease.)

• Saliva, several samples across a day. This is where the parameters used in research come from: the rise after waking (the cortisol awakening response, CAR) — by international expert consensus, "the marked increase in cortisol secretion over the first 30–45 min after morning awakening" (Stalder et al., 2016, PMID 26563991); the diurnal slope — how steeply the level falls from morning to evening; and total daily exposure.

• 24-hour urinary free cortisol. An integrated daily measure and one of the diagnostic tests in endocrinology.

• Hair cortisol. Integrates exposure over months: the growth rate assumed in this literature is about one centimetre a month (Wennig, 2000, PMID 10978637), so the proximal 3 cm are read as roughly the last three months. 🔴 That is a convention of the field, not a measurement of an individual's hair: personal growth rates range from roughly 0.5 to 2 cm a month, and segments further from the scalp lose part of the signal to washing, water and sunlight.

What is known about the measure itself: in a meta-analysis of 66 studies (N = 10,289) hair cortisol was higher under ongoing chronic load — by about 43% — but not under load in the past, and it consistently did not track perceived-stress questionnaire scores (Stalder et al., 2017, PMID 27965182). Test–retest reliability at two months and at a year sits around r = 0.68–0.79.

The practical consequence: a number from a single morning blood draw does not answer the question you are asking. It answers the question "how much was there that morning at that hour".

Does cortisol change with age

Here the answer is closest to yes, and it is mainly about the shape of the rhythm.

The reference study analysed 24-hour plasma cortisol profiles from 177 profiles, 90 men and 87 women, aged 18–83 (Van Cauter, Leproult, Kupfer, 1996, PMID 8675562). What it found:

• mean daily cortisol between ages 20 and 80 is 20–50% higher, in both sexes;

• the night-time trough rises: in people over 70 it is three to four times higher than in young adults;

• the relative amplitude of the rhythm is reduced, and the circadian rise shifts earlier;

• the morning peak rises with age in women and not in men.

🔴 One design detail changes how to read all of this: it is not a longitudinal study. The work pooled existing profiles from seven laboratories and compared age groups against each other at one slice of time. So this is not "how one person changed over the years" but "how 70-year-olds differ from 25-year-olds". The direction is supported by other datasets, but saying "with age a person's cortisol rises by X" credits the study with a design it did not have.

A flattening rhythm is part of the ageing picture alongside a shrinking temperature rhythm and changing sleep architecture. Which of these is cause and which is consequence, observation does not separate.

What is associated with outcomes — and by what design

The most robust finding here is again not about the level but about the slope: people with a flatter daily curve have worse outcomes.

Whitehall II (Kumari et al., 2011, PMID 21346074) — 4,047 British civil servants, mean age 61, 73.6% men. Six saliva samples across one weekday, followed for a mean of 6.1 years, during which there were 139 deaths, 32 of them cardiovascular. Per standard deviation towards a flatter slope, the hazard ratio for all-cause mortality was 1.30 (95% CI 1.09–1.55) and for cardiovascular death 1.87 (1.32–2.64). For non-cardiovascular death the association was not significant (1.17; 0.96–1.43). And notably, the CAR and morning cortisol were not associated with mortality at all — the measure that is easiest to collect at home predicted nothing in this cohort.

The meta-analysis of slope (Adam et al., 2017, PMID 28578301) covered 80 studies and 179 associations. The average effect was r = 0.147, significant in 10 of 12 outcome subtypes; the largest was with immune and inflammatory measures, r = 0.288.

🔴 The wording has to stay exact: associated is not causes. A flat daily rhythm is, among other things, a marker of disturbed sleep, shift work, depression, inflammation and existing disease. Any of these can sit upstream of both the cortisol pattern and the outcome. The meta-analysis authors themselves discuss flattening both as a marker and as a possible mechanism — and do not choose between them, because observational data cannot.

It is also worth holding in mind who the Whitehall II participants are: employed London civil servants, predominantly male and predominantly white. The authors state plainly that the cohort is not representative even of the UK population.

Telomeres: where the famous picture came from

Everyone knows the line "stress shortens telomeres", and it traces back to one 2004 paper (Epel et al., PMID 15574496) that has been retold thousands of times since. It is worth remembering how it was built.

It is a cross-sectional study of 58 healthy premenopausal women aged 20–50: 39 mothers of a chronically ill child and 19 control women. It measured perceived stress by questionnaire, telomere length in blood mononuclear cells, telomerase activity and markers of oxidative stress. The correlation between stress scores and telomere length was r = −0.31 (−0.27 adjusted for age); the difference between the extreme stress quartiles was about 550 base pairs.

🔴 Three things get lost in the retelling. First: cortisol was not measured in this study at all — it is about perceived stress, not about the hormone. Second: the design shows an association at one slice of time and does not show that stress shortened anything; the authors themselves write that the relationship "may be either causally or correlationally related" and that longitudinal studies are needed. Third: the caregiving mothers and the controls did not differ as groups — what differed was the duration of caregiving.

Meta-analysis of later work does confirm the association, but it is very small: r = −0.06 (95% CI −0.10 to −0.008) across 22 studies and nearly nine thousand participants, described by the authors themselves as "very small"; correcting for publication bias shrinks it to −0.03 and it is no longer significant (Mathur et al., 2016, PMID 26853993).

Where cortisol genuinely decides things — that is a disease, not a shade of normal

There is a condition in which excess cortisol is not a debatable quantity but a diagnosis with consequences: Cushing's syndrome. The recognisable picture is central obesity, purple striae, weakness of the proximal muscles (hips and shoulders), osteoporosis, hypertension and disturbed glucose metabolism.

It is diagnosed by specific tests, and under the Endocrine Society guideline (Nieman et al., 2008, PMID 18334580) one of them is used first and then confirmed by a second:

• 24-hour urinary free cortisol — at least two measurements;

• late-night salivary cortisol — two measurements;

• overnight 1 mg dexamethasone suppression test;

• alternatively, a 48-hour 2 mg/day dexamethasone test.

🔴 A random serum cortisol is not used as a screening test in that guideline — and it is exactly the test people most often order for themselves.

How different an order of magnitude this is shows in the mortality data. In a pooled analysis of 20 cohorts (3,691 patients), the standardised mortality ratio in active Cushing's syndrome was 5.7 against 2.3 in patients in remission (Limumpornpetch et al., 2022, PMID 35464961). Even after biochemical remission of Cushing's disease is achieved, the pooled estimate remains elevated at around 2.5 (van Haalen et al., 2015, PMID 25637072).

The point of this section is not to hunt for a rare disease in yourself. It is the contrast: where cortisol really does determine an outcome, the quantities are of a different order and the diagnosis follows a protocol. A shade of normal is not a smaller version of that disease.

What this field does not have

There are no completed trials in which lowering cortisol in a person without endocrine disease changed a hard outcome. This has to be stated precisely, because "nobody tried" would be false. People did try, and here is what was tried:

• cortisol-lowering drugs (ketoconazole, metyrapone, osilodrostat, mifepristone) were tested in patients with hypercortisolism — that is, not in this population;

• ashwagandha and mindfulness-based programmes were tested in relatively healthy people, but the endpoints were stress-scale scores and cortisol itself, over weeks and months;

• the trials that did move epigenetic age used entirely different interventions and were not aimed at cortisol.

No completed trial could be found with all-cause mortality or a validated biological-age composite as a primary endpoint, where the intervention targeted cortisol in people without disease. This does not mean stress is not worth managing. It means the argument "lower your cortisol and you will live longer" is, today, unsupported.

A related detail that usually surprises people: cortisol appears in none of the common composite estimates of biological age. Not in PhenoAge (albumin, creatinine, glucose, C-reactive protein, lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase, white blood cell count, plus chronological age), and not in standard implementations of the Klemera–Doubal method. More on how biological age is calculated in a separate piece.

What gets sold in the space where evidence is missing

"Adrenal fatigue". A systematic review that selected 58 papers out of 3,470 ends in these words: "This systematic review proves that there is no substantiation that 'adrenal fatigue' is an actual medical condition. Therefore, adrenal fatigue is still a myth" (Cadegiani & Kater, 2016, PMID 27557747). The Endocrine Society puts it the same way: no scientific proof exists to support it as a real medical condition. The label usually bundles fatigue, sleepiness and salt craving — a set compatible with a dozen real causes, including sleep debt, anaemia, hypothyroidism and depression, each of which is diagnosed and managed differently.

Supplements that "lower cortisol". Take the most popular one, ashwagandha. Trials exist and do show reductions in stress scores and serum cortisol: 64 participants over 60 days (PMID 23439798), 60 participants over 60 days (PMID 31517876), 60 participants over 8 weeks (Salve et al., 2019). But these are surrogate endpoints: a questionnaire score moved and the hormone moved, not morbidity and not lifespan. Nor is the picture uniform — in a separate trial in men aged 40–70, cortisol did not differ between groups (PMID 30854916).

🔴 And separately, the part that marketing copy leaves out: ashwagandha has documented idiosyncratic drug-induced liver injury, with published case series including severe outcomes in people whose livers were already diseased; the LiverTox reference assigns it a likelihood category of B. Denmark, after a 2020 assessment, could not establish a safe dose and has not permitted the plant in food and supplements since 2023; warnings have also been issued by Germany's BfR, the Netherlands' RIVM, France's ANSES and Australia's TGA.

Stress-reduction programmes. Here the record is more honest: they reduce perceived-stress scores reproducibly, while the effect on cortisol itself is inconsistent. A meta-analysis of salivary cortisol in healthy adults found only 5 randomised trials and 190 participants (g = 0.41), and with raw measures the effect tended towards zero; larger reviews mix meditation with yoga, which makes the result harder to read.

What actually changes decisions

The practical conclusion is dull, and honest for that reason.

• Do not chase a cortisol number. A single measurement does not answer the question you are asking; a series of saliva samples is a research instrument, not routine diagnostics; and the measure that is easiest to collect in the morning predicted nothing in the largest cohort with mortality data.

• Look at what is measurable and reproducible. Sleep duration and regularity, blood pressure, glucose and HbA1c, waist-to-height ratio, resting heart rate and its variability, inflammation markers. These move with the same lifestyle inputs and, unlike cortisol, come with intelligible reference ranges and intelligible decisions.

• Sleep is the most direct lever of those, and it is also what most distorts any cortisol measurement (a fuller treatment: sleep and healthspan). In a laboratory experiment on 11 healthy young men, six nights of four hours in bed raised evening cortisol and shifted the quiescent period of the axis by about an hour and a half (Spiegel, Leproult, Van Cauter, 1999, PMID 10543671). The necessary caveat: eleven young men in a clinical research centre are not "everyone in ordinary life".

• Do not expect training to "lower cortisol". Acute exercise raises it, and that is normal physiology: for 30 minutes of work at 40, 60 and 80% of maximal oxygen uptake, the change was roughly +6, +40 and +83% respectively (Hill et al., 2008, PMID 18787373). For habitual activity the data are modest and observational: a meta-analysis links it to a slightly steeper diurnal slope, r = −0.043 — not a magnitude anyone should adopt as a target.

• Discuss it with a clinician, not with a calculator. A combination of weight gain concentrated around the abdomen, persistently raised blood pressure, newly disturbed glucose metabolism, weakness in the hip and shoulder muscles, and characteristic skin changes is a reason for an in-person work-up following a protocol — not for buying a supplement, and not for ordering a "blood cortisol" on your own initiative, which as shown above is not part of the diagnostic algorithm.

Stress is a real health factor: in large prospective data, job strain is associated with coronary heart disease (a hazard ratio of about 1.23 across some two hundred thousand workers, Kivimäki et al., 2012, PMID 22981903). Cortisol is the mechanism through which that is convenient to explain. But an explanation and a measurable target are different things, and turning the first into the second is still premature.

This material is informational and does not replace a consultation with a doctor. Decisions about diagnosis and treatment are made in person and on the basis of your own data.

Sources

• Stalder T. et al. (2016). Assessment of the cortisol awakening response: Expert consensus guidelines. Psychoneuroendocrinology, 63, 414–432. PMID 26563991.

• Adam E.K., Gunnar M.R. (2006). Diurnal cortisol in field settings. PMID 16500024.

• Wennig R. (2000). Potential problems with the interpretation of hair analysis results. Forensic Science International, 107(1–3), 5–12. PMID 10978637.

• Stalder T. et al. (2017). Stress-related and basic determinants of hair cortisol in humans: A meta-analysis. Psychoneuroendocrinology, 77, 261–274. PMID 27965182.

• Van Cauter E., Leproult R., Kupfer D.J. (1996). Effects of gender and age on the levels and circadian rhythmicity of plasma cortisol. Journal of Clinical Endocrinology & Metabolism, 81(7), 2468–2473. PMID 8675562.

• Kumari M. et al. (2011). Association of diurnal patterns in salivary cortisol with all-cause and cardiovascular mortality: findings from the Whitehall II study. Journal of Clinical Endocrinology & Metabolism, 96(5), 1478–1485. PMID 21346074.

• Adam E.K. et al. (2017). Diurnal cortisol slopes and mental and physical health outcomes: A systematic review and meta-analysis. Psychoneuroendocrinology, 83, 25–41. PMID 28578301.

• Epel E.S. et al. (2004). Accelerated telomere shortening in response to life stress. PNAS, 101(49), 17312–17315. PMID 15574496.

• Mathur M.B. et al. (2016). Perceived stress and telomere length: A systematic review, meta-analysis, and methodologic considerations. Brain, Behavior, and Immunity, 54, 158–169. PMID 26853993.

• Cadegiani F.A., Kater C.E. (2016). Adrenal fatigue does not exist: a systematic review. BMC Endocrine Disorders, 16(1), 48. PMID 27557747.

• Nieman L.K. et al. (2008). The diagnosis of Cushing's syndrome: an Endocrine Society Clinical Practice Guideline. Journal of Clinical Endocrinology & Metabolism, 93(5), 1526–1540. PMID 18334580.

• Limumpornpetch P. et al. (2022). The effect of endogenous Cushing syndrome on all-cause and cause-specific mortality. Journal of Clinical Endocrinology & Metabolism, 107(8), 2377–2388. PMID 35464961.

• van Haalen F.M. et al. (2015). Mortality remains increased in Cushing's disease despite biochemical remission: a systematic review and meta-analysis. European Journal of Endocrinology, 172(4), R143–R149. PMID 25637072.

• Levine M.E. et al. (2018). An epigenetic biomarker of aging for lifespan and healthspan. Aging, 10(4), 573–591. PMID 29676998.

• Chandrasekhar K., Kapoor J., Anishetty S. (2012). A prospective, randomized double-blind, placebo-controlled study of safety and efficacy of a high-concentration full-spectrum extract of ashwagandha root. Indian Journal of Psychological Medicine, 34(3), 255–262. PMID 23439798.

• Lopresti A.L. et al. (2019). An investigation into the stress-relieving and pharmacological actions of an ashwagandha extract. Medicine, 98(37), e17186. PMID 31517876.

• Lopresti A.L. et al. (2019). A randomized, double-blind, placebo-controlled, crossover study examining the hormonal and vitality effects of ashwagandha in aging, overweight males. American Journal of Men's Health, 13(2). PMID 30854916.

• Spiegel K., Leproult R., Van Cauter E. (1999). Impact of sleep debt on metabolic and endocrine function. The Lancet, 354(9188), 1435–1439. PMID 10543671.

• Hill E.E. et al. (2008). Exercise and circulating cortisol levels: the intensity threshold effect. Journal of Endocrinological Investigation, 31(7), 587–591. PMID 18787373.

• Moyers S.A., Hagger M.S. (2023). Physical activity and cortisol regulation: A meta-analysis. Biological Psychology, 179, 108548. PMID 37001826.

• Kivimäki M. et al. (2012). Job strain as a risk factor for coronary heart disease: a collaborative meta-analysis of individual participant data. The Lancet, 380(9852), 1491–1497. PMID 22981903.