The short answer: Finnish studies indeed show a strong association between frequent sauna use and lower mortality. However, due to the design of the statistical models, we still cannot cleanly separate the effect of the sauna itself from the healthy lifestyle of its users.

  • Research note: This article is based on a structured review of the Finnish sauna and mortality literature. These are observational cohort data, not randomized trials, and the studies do not measure how many years of life sauna adds.

Sauna is one of those health habits that seems almost suspiciously pleasant. Sit in a hot room, relax, repeat several times a week, and perhaps live longer.

That idea is not invented by wellness marketing. Several Finnish studies really did report substantially lower mortality among people who used sauna more often.

But I wanted to ask a narrower question: does the association survive after accounting for the fact that frequent sauna users may simply live differently?

The answer is interesting. The association often survives broad statistical adjustment. But the published models do not let us cleanly separate “sauna” from “lifestyle.” That missing distinction changes how confidently the results can be interpreted.

What the famous 2015 Finnish study actually found

The best-known study followed 2,315 middle-aged men from Eastern Finland for a median of 20.7 years. Sauna use was reported at baseline and divided into three groups: once per week, 2–3 times per week, and 4–7 times per week.

For all-cause mortality, the numbers looked like this:

Sauna frequency Age-adjusted hazard ratio Broadly adjusted hazard ratio
1 time/week 1.00 1.00
2–3 times/week 0.69 0.76
4–7 times/week 0.61 0.60

The 4–7-times-per-week group therefore had a mortality hazard around 40% lower than the once-weekly group after the study’s broad adjustment. The result for 2–3 sessions was weaker, but still below 1.

The crude death proportions were also striking: 49.1% in the once-weekly group, 37.8% in the 2–3-times group, and 30.8% in the 4–7-times group. But those raw percentages should not be read as “sauna prevented this many deaths.” The groups differed in other ways, and people were followed over long periods.

The original paper is here: Laukkanen et al., JAMA Internal Medicine, 2015.

The result survives adjustment. That is the interesting part.

A simple healthy-user explanation would predict that the apparent sauna benefit should shrink sharply once researchers adjust for smoking, alcohol, exercise and other health differences. A related selection problem complicates longevity comparisons among elite athletes.

That did not happen cleanly.

For the highest-frequency group, the all-cause mortality hazard ratio moved from 0.61 to 0.60 after broad adjustment. In other words, it barely moved at all. For the 2–3-times-per-week group, it moved from 0.69 to 0.76, so some attenuation appeared there.

At first glance, this looks very strong for sauna.

But there is a statistical trap hidden inside the phrase “adjusted for lifestyle.”

The adjustment problem nobody notices

The broader models did not add only smoking, alcohol and physical activity.

They added those variables together with things such as BMI, blood pressure, LDL cholesterol, previous heart attack, diabetes, cardiorespiratory fitness, resting heart rate and socioeconomic status.

Those are not all the same kind of variable.

Smoking and alcohol are lifestyle behaviours. Previous heart disease is baseline health. Income or occupation is socioeconomic context. Blood pressure, fitness and BMI are even more complicated because they could be confounders, consequences of long-term behaviour, or potentially part of a pathway through which sauna itself affects health.

So when a hazard ratio changes from 0.69 to 0.76, we cannot say: “lifestyle explained this much of the sauna effect.” Too many different things changed in the model at the same time.

For the underlying review, I used a deliberately strict rule: two models counted as a valid lifestyle-only comparison only if they were otherwise identical and differed solely by verified lifestyle variables. This was a review of published model specifications and reported hazard ratios, not a reanalysis of individual participant data.

None of the 12 model transitions examined passed that test.

That does not mean lifestyle confounding is absent. It means these 12 extracted model transitions do not isolate its contribution. Even a lifestyle-only change in a Cox-model hazard ratio would not, by itself, prove what fraction was caused by lifestyle: residual confounding, model dependence and the non-collapsibility of hazard ratios would still matter.

What did we actually check? The original research note mapped eight related Finnish reports, but completed detailed model extraction for only four publications (2015–2018). The 12 comparisons below come from those four, not all eight. The other four (2022–2024) were identified but not subjected to the same complete model-by-model audit. The results are therefore provisional and limited in scope.

Show the technical audit: all 12 model comparisons

We compared successive published Cox regression models for the same outcome and sauna-frequency contrast. A transition would qualify as a lifestyle-only comparison only if the newly added variables were exclusively verified behaviours (such as smoking, alcohol or physical activity), with the other specification unchanged. Every row below adds non-lifestyle covariates as well or does not isolate lifestyle. 0 of 12 passed this narrowly defined screening test. This is an audit of reporting and comparability, not a new estimate of a causal sauna effect.

Report Outcome and frequency contrast Published models / HR sequence Reason not lifestyle-only
2015 JAMA All-cause; 2–3 vs 1/week Age-adjusted → broadly adjusted: 0.69 → 0.76 Behaviours plus socioeconomic, clinical and physiological variables
2015 JAMA All-cause; 4–7 vs 1/week Age-adjusted → broadly adjusted: 0.61 → 0.60 Same mixed adjustment set
2017 Annals of Medicine Cardiovascular; 3–7 vs ≤2/week Model 1 → 2: 0.67 → 0.73 Mixed cardiovascular-risk covariates
2017 Annals of Medicine All-cause; 3–7 vs ≤2/week Model 1 → 2: 0.77 → 0.82 Same mixed-risk adjustment
2017 Annals of Medicine Cardiovascular; 3–7 vs ≤2/week Model 2 → 3: 0.73 → 0.74 Socioeconomic status and C-reactive protein added together
2017 Annals of Medicine All-cause; 3–7 vs ≤2/week Model 2 → 3: 0.82 → 0.84 Socioeconomic status and C-reactive protein added together
2018 BMC Medicine Cardiovascular; 2–3 vs 1/week Model 1 → 2: 0.71 → 0.78 Behavioural, clinical and physiological variables added together
2018 BMC Medicine Cardiovascular; 4–7 vs 1/week Model 1 → 2: 0.30 → 0.36 Same mixed adjustment
2018 BMC Medicine Cardiovascular; 2–3 vs 1/week Model 2 → 3: 0.78 → 0.77 Physical activity and socioeconomic status added together
2018 BMC Medicine Cardiovascular; 4–7 vs 1/week Model 2 → 3: 0.36 → 0.36 Physical activity and socioeconomic status added together
2018 Progress in Cardiovascular Diseases Sudden cardiac death; 3–7 vs ≤2/week Model 1 → 2: 0.59 → 0.66 Mixed cardiovascular-risk covariates
2018 Progress in Cardiovascular Diseases Sudden cardiac death; 3–7 vs ≤2/week Model 2 → 3: 0.66 → 0.67 Socioeconomic status and C-reactive protein added together

HR = hazard ratio; ≤ = at most. Values are transcribed from the original provisional research extraction and the linked publications, rounded to two decimals. The audit includes two models from the 2015 paper, four transitions from 2017, four from the 2018 mixed-sex BMC paper and two from the 2018 sudden-cardiac-death paper. The 2018 sudden-cardiac-death paper’s intermediate model values are from the original working extraction; its abstract reports only the final 0.67 estimate. An HR moving closer to or farther from 1 under mixed adjustment cannot be labelled a lifestyle-specific effect. These 12 comparisons are not 12 independent populations.

Other Finnish studies point in the same direction

The pattern did not appear only once.

A 2017 analysis of 2,277 men followed for a median of 26.1 years compared 3–7 sauna sessions per week with two or fewer. After the broader adjustments, the reported hazard ratio was about 0.74 for cardiovascular mortality and 0.84 for all-cause mortality. See the study in Annals of Medicine.

A 2018 BMC Medicine study included 1,688 men and women aged 53–74. Compared with one sauna session per week, the broadly adjusted cardiovascular-mortality hazard ratio was 0.77 for 2–3 sessions and 0.36 for 4–7 sessions. The confidence interval for the 2–3 group included 1, while the 4–7 group remained clearly below it. See the full study.

Another long-term analysis of 2,291 men reported a hazard ratio around 0.67 for sudden cardiac death in those using sauna 3–7 times per week versus two or fewer. See the 2018 study.

So there is a recurring signal. Frequent sauna use and lower mortality keep appearing together.

But these are not independent replications

This is an important detail.

In our original, provisional eight-report source map, the papers use the Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD), its follow-up visits or overlapping subsets, rather than eight independently recruited populations. The 2018 mixed-sex study includes women invited at a later KIHD follow-up, so its population and measurement dates are not identical to the original male cohort. The four 2022–2024 papers are included in the source map, not in the 12-transition model audit.

Show the eight-report cohort map and original papers
Paper Outcome / analytical focus Dataset relationship Extraction status
Laukkanen et al., 2015 Cardiovascular and all-cause death KIHD, 2,315 men at baseline Detailed model audit
Kunutsor et al., 2017 Fitness × sauna; cardiovascular and all-cause death KIHD male subset, 2,277 Detailed model audit
Laukkanen et al., 2018 Fitness × sauna; sudden cardiac death KIHD male subset, 2,291 Detailed model audit
Laukkanen et al., 2018 (BMC) Cardiovascular mortality, men and women Later KIHD follow-up sample, 1,688; women joined at follow-up Detailed model audit
Kunutsor et al., 2022 (inflammation) High-sensitivity CRP × sauna; all-cause death KIHD men Mapped, not fully extracted
Kunutsor et al., 2022 (socioeconomic status) Socioeconomic status × sauna; all-cause death KIHD, 2,575 men Mapped, not fully extracted
Laukkanen et al., 2023 Systolic blood pressure × sauna; cardiovascular death KIHD, 2,575 men Mapped, not fully extracted
Kunutsor et al., 2024 Systolic blood pressure × sauna; all-cause death KIHD, 2,575 men Mapped, not fully extracted

Sharing a cohort does not invalidate a result. It limits how strongly multiple papers can be counted as independent population replications. This is a map of the original review’s eight reports, not a systematic census of every sauna study ever published. Different endpoints, baseline examinations and analytical subsets also mean these papers should not be pooled as if their effect estimates were interchangeable.

That means eight papers are not eight separate Finnish populations independently reproducing the same result. Much of the literature is different analyses of the same underlying cohort.

This does not make the finding meaningless. It simply makes the evidence base narrower than the publication count may suggest.

What about reverse causation?

There is another obvious problem: perhaps healthier people go to sauna more often because they are healthy enough to do so.

The 2015 researchers tried one useful test. They excluded deaths occurring during the first five years of follow-up. The broadly adjusted hazard ratios were still about 0.76 for 2–3 sessions and 0.66 for 4–7 sessions.

In the later BMC analysis, excluding the first five years produced cardiovascular-mortality estimates around 0.70 and 0.37.

That weakens one simple version of reverse causation, where somebody is already seriously ill, stops using sauna, and dies soon afterward.

But it cannot remove all health selection. Someone may reduce sauna use years before death because of frailty, cardiovascular symptoms, mobility problems or changing habits. Sauna frequency was also generally self-reported at baseline rather than repeatedly measured throughout decades of follow-up.

A 2019 KIHD study on cancer offers a useful reality check: sauna habits were measured again in a subset of men 11 years later, and the results showed considerable variation. This does not tell us whether the mortality estimates are too high or too low, but it shows why one measurement at the start may not describe decades of behaviour. Thanks to Nick Tiller for pointing out this paper.

Does sauna add years to your life?

This is where I think many headlines go too far.

The Finnish studies reported hazard ratios, event rates and survival curves. They did not report “sauna adds X years of life.”

A hazard ratio cannot simply be converted into years of extra lifespan. To do that credibly, we would need age-specific survival information and additional assumptions, ideally with repeated measurements of sauna exposure over time.

So “40% lower mortality hazard” is not the same statement as “40% longer life,” and it is not even the same as “40% fewer deaths.”

My current reading of the evidence

I think there is a real and interesting observational signal here.

I would not dismiss the Finnish sauna findings as a trivial healthy-user artifact, because the association often remains substantial after broad adjustment and after excluding early deaths. The high-frequency groups are especially interesting.

But the strongest causal claim is still not justified.

The cleanest statement I can make is this:

Frequent Finnish-style sauna use is associated with lower mortality in these Finnish cohort analyses, and the association often survives broad adjustment. We still do not know how much of that association is independent of lifestyle, nor whether sauna itself adds years of life.

That may sound less exciting than “sauna makes you live longer.” I think it is actually more interesting. The association often remains after substantial adjustment, but none of the 12 published model transitions we extracted isolated lifestyle adjustment. Four more mapped reports still require full extraction; a stronger causal conclusion would need more than this published-model comparison.

Sources

The first four papers below were used for the detailed 12-transition audit. The remaining four were mapped for cohort overlap but not fully extracted. Research status: provisional published-model review, not a participant-level reanalysis.

Additional mapped KIHD reports (not part of the 12 comparisons)

Medical note: This article reviews observational research and is not medical advice. It does not establish that sauna prevents disease or extends lifespan.