Peripheral-scope lifestyle lever (heat exposure), LOW-CERTAINTY — not a big rock. Sauna is admitted on the same evidence bar as food or activity, not by a category exemption; the Layer-1 ranking is what keeps it peripheral. The reported relative associations are large (2015 fatal CVD 0.50, all-cause 0.60 at 4-7/wk (Laukkanen et al., 2015); 2018 fatal CVD 0.36, Model 3 (Laukkanen et al., 2018)), but how much of them is causal is unknown: one cohort, plausible residual confounding, and a top category resting on few deaths. The low rank comes from that certainty, not from a measured small size. Under the wiki’s ranking rule, the attention an exposure gets is never a pass; what decides here is that every mortality estimate the wiki holds rests on one observational cohort, and its transfer to anyone outside that cohort is an open question, not an assumption.
The dominant limitation — one cohort, one research group, two analyses
Every sauna -> mortality estimate below comes from one cohort, the Finnish Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD), analysed by one research group:
- 2015 analysis — men only. A prospective cohort of 2315 middle-aged men (age 42-60) from Eastern Finland, baseline 1984-89, median follow-up 20.7 years (IQR 18.1-22.6); outcomes SCD, fatal CHD, fatal CVD, all-cause. (Laukkanen et al., 2015)
- 2018 analysis — men and women, older. The KIHD re-examination sample: «In the 11-year follow-up visit of the first cohort, women were invited to join this study.» 1688 participants (867 women, 821 men), aged 53-74 (the 53.4-73.8 range describes the sampled cohort of 2358), baseline March 1998 - December 2001, median follow-up 15.0 years (IQR 14.1-15.9), 181 fatal CVD deaths (118 men, 63 women). Fatal CVD is its only outcome — no all-cause, SCD or CHD-death results. (Laukkanen et al., 2018)
- The two analyses are not independent. Same cohort, same group (T. and J.A. Laukkanen, Khan, Zaccardi author both), so agreement between them is not replication and never type-E backing. The 2018 paper does not state how many of its 821 men were also in the 2015 analysis; its wording places the 1998-2001 visit inside the original male cohort’s follow-up, so the men are plausibly largely the same people re-examined 11 years on. (inferred from Laukkanen et al., 2018)
- A companion review (the Mayo 2018 review) shares authors with these papers, so it cannot supply type-E backing whatever it contains. Its content is not held, and it must not be cited as replication. (inferred from Laukkanen et al., 2018)
The wiki holds no independent replication cohort, no non-Finnish population, no RCT, and no systematic review over independent data; women enter only through the 2018 KIHD sample, where the frequency association was not detected and the duration estimates are imprecise (below). Whether an independent cohort or a review exists is an open acquisition question: the searches run so far did not settle it, so this is a statement about holdings, not about the literature. One named candidate is a systematic review of regular dry sauna bathing by authors outside the KIHD group, cited by the 2018 paper (ref 3); what it covers, and whether it adds any non-KIHD mortality data, is unverified. In 2018 the authors wrote of their 2015 paper: «To our knowledge, this is the only available study [19] on the prospective association between sauna habits and the risk of mortality outcomes» — a statement as of 2018, not a current survey. (Laukkanen et al., 2018) Transportability is therefore almost entirely unestablished — the association is documented in middle-aged and older Eastern Finns, with the frequency signal carried by men’s deaths, by one research group. The peripheral-scope framing and Layer-1 ranking here are the wiki’s own appraisal, applied to the sources’ reported estimates.
The dose-response — an observational gradient in one cohort
2015 analysis — men 42-60, four outcomes
Source: (Laukkanen et al., 2015)
Frequency groups (self-reported sessions/week): 1/wk (n=601), 2-3/wk (n=1513), 4-7/wk (n=201). Events over follow-up: 190 sudden cardiac deaths (SCD), 281 fatal coronary heart disease (CHD), 407 fatal cardiovascular disease (CVD), 929 all-cause. Crude proportions across 1 / 2-3 / 4-7 per week over the whole follow-up: fatal CVD 22.3% / 16.4% / 12.0%, all-cause 49.1% / 37.8% / 30.8% (unadjusted). The 4-7/wk group’s counts are small: 10 SCD, 17 fatal CHD, 24 fatal CVD and 62 all-cause deaths. Multivariable-adjusted hazard ratios vs the 1-session/week referent:
| Outcome | 2-3/wk HR (95% CI) | 4-7/wk HR (95% CI) | P-trend |
|---|---|---|---|
| SCD | 0.78 (0.57-1.07) | 0.37 (0.18-0.75) | .005 |
| Fatal CHD | 0.77 (0.60-0.99) | 0.52 (0.31-0.88) | .005 |
| Fatal CVD | 0.73 (0.59-0.89) | 0.50 (0.33-0.77) | <.001 |
| All-cause | 0.76 (0.66-0.88) | 0.60 (0.46-0.80) | <.001 |
Duration groups (<11 min n=1065, 11-19 min n=722, >19 min n=528), multivariable-adjusted vs <11 min: SCD >19 min 0.48 (0.31-0.75), P-trend .002; fatal CHD >19 min 0.64 (0.46-0.88), P-trend .007; fatal CVD >19 min 0.76 (0.59-0.97), P-trend .03. For all-cause mortality no duration trend was detected (P-trend .93) even though a frequency trend was — an outcome-specific split the source states in prose (frequency, but not duration, was inversely associated with all-cause mortality). The multivariable all-cause >19 min cell is printed as 0.83 (0.87-1.20), a point estimate outside its own interval, so that cell cannot be read. The 11-19 min band is null for SCD (0.93, 0.67-1.28); only the top duration band separates.
2018 analysis — men and women 53-74, fatal CVD only
Source: (Laukkanen et al., 2018)
Same three frequency categories, once/week referent. Model 1 = age + sex; Model 3 = + BMI, smoking, SBP, LDL-C, alcohol, previous MI, type 2 diabetes, physical activity and socioeconomic status (the full-sample fully adjusted model). HR (95% CI):
| Stratum | Category | Deaths/n | Model 1 | Model 3 |
|---|---|---|---|---|
| All | 2-3/wk | 110/1028 | 0.71 (0.52-0.98) | 0.77 (0.56-1.07) |
| All | 4-7/wk | 8/205 | 0.30 (0.14-0.64) | 0.36 (0.17-0.77) |
| Men | 2-3/wk | 71/512 | 0.61 (0.41-0.90) | 0.69 (0.46-1.03) |
| Men | 4-7/wk | 8/132 | 0.33 (0.15-0.71) | 0.39 (0.18-0.84) |
| Women | 2-3/wk | 39/516 | 0.95 (0.57-1.57) | 1.00 (0.57-1.74) |
| Women | 4-7/wk | 0/73 | not estimated | not estimated |
Referent deaths/n: all 63/455, men 39/177, women 24/278.
- Absolute rates (crude): fatal CVD per 1000 person-years, once / 2-3 / 4-7 per week: 10.1 (7.9-12.9) / 7.6 (6.3-9.2) / 2.7 (1.3-5.4). Unadjusted. The 4-7 group was younger (mean 60 vs 64 years), which widens the crude gap, and more often male (64.4% vs 38.9%), which narrows it; the crude ratio 4-7 vs once (about 0.27) sits close to the age-and-sex-adjusted Model 1 HR of 0.30. (inferred from Laukkanen et al., 2018)
- The abstract’s headline 4-7/wk HR, 0.23 (0.08-0.65), is Model 4 — Model 3 plus incident CHD as a time-varying covariate, run only in participants free of CHD at baseline («The model was limited to the population at risk and did not include those who already had coronary heart disease»; 28.1% had a CHD history). It is not the full-sample estimate. It also adjusts for incident CHD, which the authors include «as it is a known factor in the path- way for development of CVD mortality» — a step on the causal path, so adjusting for it is a second reason not to read 0.23 as the association. (inferred from Laukkanen et al., 2018)
- The top category rests on 8 deaths, all in men. The 73 women at 4-7/wk add person-time with no deaths, which pulls the pooled estimate down (overall Model 3 0.36 vs men 0.39), so the overall figure is not a men-only estimate; but every death in the top category is a man’s. (inferred from Laukkanen et al., 2018)
- Weekly duration (total minutes per week: <=15, 16-45, >45 — a different quantity from 2015’s minutes per session): Model 3 HR for >45 vs <=15 min/week 0.57 (0.35-0.94) overall (Model 1 0.49, 0.30-0.80); 16-45 min/week 0.77 (0.55-1.07). Neither sex alone reaches significance in Model 3 (men >45: 0.57, 0.31-1.05; women >45: 0.75, 0.31-1.85); sex interaction P = 0.314. Weekly duration largely tracks frequency (median 60 min/week in the 4-7 group vs 10 in the once group), and no model holds both, so it is not shown to be a separate lever. (inferred from Laukkanen et al., 2018)
Shape — no knee located; curvature not detected for frequency, detected for weekly duration
2015: three categories with point estimates ordered inversely for frequency (all four outcomes, P-trend reported) and for duration (except all-cause, where the trend P is .93 and the top multivariable cell is unreadable, above). With three categories this is an ordering plus a trend test, not a tested shape, and the middle frequency category does not separate from the referent for SCD (0.78, 0.57-1.07). No knee is located; asserting one would need it pinned. The highest-frequency group holds only n=201, and the studied ranges are frequency 1-7/wk, duration 2-90 min (mean 14.2), temperature 40-100°C (mean 78.9). (Laukkanen et al., 2015) The apparent optimum (4-7/wk, >19 min) is therefore the top studied category edge, not a curve feature; do not launder a category edge into a target, as there is no data above the top bin.
2018: a restricted cubic spline (Model-2 covariates) «decreased linearly with increasing sauna sessions from one to seven (P value for non-linearity = 0.932)»; the abstract states it «decreased linearly with increasing sauna sessions per week with no threshold effect». For weekly duration the fit was the other way: «statistical tests suggested a fit with a non-linear shape (P for non-linearity = 0.005)», and the authors leave open whether a curvilinear or linear shape fits better. (Laukkanen et al., 2018)
Reading: for frequency the statement is monotone over 1-7 sessions/week and no more. A non-linearity P of 0.932 is a failure to detect curvature, not evidence of a straight line; the top of the curve is carried by 8 deaths, and the authors themselves say the event count «precluded detailed assessment of (i) effect modification by relevant clinical characteristics on the associations and (ii) dose-response relationships of the associations». The spline’s 95% band touches or crosses 1.0 around 2-3 sessions/week and again near 7 (read from Fig 3), so even the frequency line is imprecise at both ends. No data exist above 7/week. The top category remains a studied edge, not a target. Weekly duration’s top category (>45 min/week) is open-ended; Table 1 puts the overall median at 30 min/week (IQR 15-40) and the 4-7/week group’s at 60 (40-90), so data above roughly 90 min/week are sparse, and the detected non-linearity has no located knee. The duration-quartile plot (Fig 4) is not transcribed here, so any flattening it shows is not read. (inferred from Laukkanen et al., 2018)
Women — studied, frequency association not detected: insufficient evidence
Estimates from (Laukkanen et al., 2018); the evidence-state verdict is this page’s.
- Frequency: women at 2-3/wk vs once/week, Model 3 HR 1.00 (0.57-1.74) (Model 1 0.95, 0.57-1.57); at 4-7/wk, 0 deaths among 73 women (vs 24 among 278 in the once-a-week group), not estimable. A zero cell points the same way as the association but is crude and unadjusted: at the once/week women’s crude proportion (24/278) about 6 deaths would be expected among 73, before allowing for the younger age of the 4-7 group (Table 1, both sexes), so 0 is informative in direction but carries no hazard ratio. Not detected here therefore means a null-centred 2-3/wk estimate plus an unestimable 4-7/wk cell that leans favourable, not a null across the range. (inferred from Laukkanen et al., 2018) The authors: «In the analyses by gender, there was no statistically significant evidence of associations in women, which could be attributed largely to the low event rates in the sauna exposure categories (Table 2). A test of interaction showed that the association between sauna bathing frequency and CVD mortality was not significantly modified by gender (P for interaction = 0.524).»
- Weekly duration: women 16-45 vs <=15 min/week, Model 1 0.53 (0.31-0.89), Model 3 0.61 (0.35-1.07); >45 min/week Model 3 0.75 (0.31-1.85). Interaction P = 0.314.
- How to read it: 63 deaths among 867 women. An interaction test that fails to reject (P = 0.524) is not evidence that the association is equal in women — it is a low-power test; equally, a null-centred estimate with a wide interval is not evidence of no association. The evidence state for women is insufficient, not benefit and not no-effect. The title («Sauna bathing is associated with reduced cardiovascular mortality and improves risk prediction in men and women») and the abstract’s wording («strongly, inversely, and independently associated with fatal CVD events in middle-aged to elderly males and females»; the main-text Conclusions say «may be related to a lower risk») rests on sex-adjusted pooled estimates. For frequency the top category contains only men’s deaths; the women’s 2-3/wk estimate is null-centred and their 4-7/wk cell has no deaths to estimate from. For weekly duration the women-specific estimates point lower but are imprecise: 16-45 min/week reaches significance only in Models 1 and 4 (Model 4 0.51, 0.28-0.91), and >45 min/week rests on 6 deaths (Model 3 0.75, 0.31-1.85). (inferred from Laukkanen et al., 2018)
- Older band: among men aged 53-74 the frequency estimates run the same direction as in men 42-60 (4-7/wk Model 3 0.39, 0.18-0.84, 8 deaths), within the same cohort and plausibly largely the same men, so this is a later-life look at one group, not an independent older-age confirmation. Split at age 65.2 (both sexes, Model-2 covariates, 4-7 vs once/week), the younger half gives 0.15 (0.04-0.66) and the older half 0.46 (0.20-1.09), not significant; the test for a difference (P 0.287) fails to reject, which is not evidence that the association holds equally above 65. (inferred from Laukkanen et al., 2018)
Is the gradient causal, or a healthy-adherer / reverse-causation artifact?
This is the prime suspect for any protective observational gradient: frequent sauna users plausibly have more leisure, higher income, and better baseline health, and sick people sauna less (sick-quitter / reverse causation). The U-Shaped Association Artifact sets the rule: a protective arm that has survived only covariate adjustment and early-follow-up exclusion (the weak checks) is unadjudicated, not established; the strong checks are a referent correction or a genetic / quasi-experimental test. The rule is written for U/J shapes; this page applies it to a monotone protective gradient, the same protective-arm question. The sauna gradient is reported to survive the weak checks in both analyses (the early-exclusion numbers are in unheld supplements) and faced neither strong one: the referent is once a week (possibly including the 43 non-users, below), not a never-user group, and there is no genetic instrument. The 2018 authors add that «We could not account for the longer-term duration and regularity of sauna use prior to the study entry because of the lack of data», so people who had cut down before baseline cannot be separated out of the referent. By that rule the gradient stays unadjudicated. (inferred from Laukkanen et al., 2018)
What the checks cover:
- 2015 — adjusted for the prime confounders. The multivariable model adjusts for age, BMI, systolic BP, LDL-C, smoking, alcohol, previous MI, type 2 diabetes, cardiorespiratory fitness, resting heart rate, physical activity, and socioeconomic status, and the gradient survives the adjustment. (Laukkanen et al., 2015) Fitness and SES are the candidates this page treats as most likely to drive a healthy-adherer pattern; the source does not rank confounders.
- 2018 — a thinner model. Model 3 adjusts for self-reported physical activity (duration per week) and an SES index, but not cardiorespiratory fitness. The authors call PA confounding «unlikely as our analysis accounted for the role of physical activity», leaning on the group’s own joint fitness-and-sauna papers (not held). (Laukkanen et al., 2018) So the 2015 fitness-adjustment argument does not carry over to the women or the older band.
- 2018 — baseline differences between frequency groups. The 4-7 group was younger (60 vs 64), more often male (64.4% vs 38.9%) and smoked less (9.3% vs 15.4%; three-group heterogeneity P 0.091). Once vs 4-7 per week: SBP 137 vs 135, LDL-C 3.56 vs 3.59 mmol/L, CHD history 28.1% vs 27.8%, type 2 diabetes 9.9% vs 6.3% (three-group P 0.233); the frequent users had higher BMI (27.4 vs 28.2), alcohol (median 8.75 vs 24.0 g/week) and energy intake, and more PA (7.31 vs 8.48 h/week), and they worked longer days (7.9 vs 8.7 h, three-group P < 0.001). They also differed on housing — family house 23.4% vs 70.7%, apartment 60.1% vs 19.0% — and on education, where frequent users had fewer academic degrees (10.3% vs 2.9%). Annual income was 16,970 / 15,337 / 18,377 EUR across once / 2-3 / 4-7 per week (heterogeneity P 0.001, not monotone). From «minor differences in annual salary levels» the authors argue «sauna ownership does not correlate with financial status in Finland»; housing conditions are one component of the adjusted SES index, and on that index «SES did not differ when comparing one vs. four to seven times per week frequency groups; SES level was the highest among those using sauna two to three times per week» (Table 1: 10.3 vs 10.3). (Laukkanen et al., 2018) The differences point both ways. Toward a healthy-user pattern: the frequent users were younger, more active, smoked less (not significantly), had the highest income and mostly lived in family houses. Against one: fewer academic degrees (which may partly track rural vs urban living), longer working days, higher BMI and more alcohol. The combined SES index does not separate once from 4-7 per week. Covariate adjustment covers some of these; which way the residual runs is not known.
- Sensitivity analyses addressed reverse causation (both), with the numbers in unheld supplements in both. 2015: associations remained consistent after excluding men who died in the first 5 years («eTable 2 in the Supplement») and when restricting follow-up to 10 years («eTable 3 in the Supplement») (Laukkanen et al., 2015). 2018: «consistent» after excluding the first 5 years (Tables S1-S2) (Laukkanen et al., 2018). Neither paper’s sensitivity estimates are checkable here; both rest on the authors’ narrative.
Subgroups — not evidence about confounding in either paper. 2015 states that results «were consistent among subgroups, although associations seem to be somewhat stronger in nonsmokers and in men with type 2 diabetes and poor cardiorespiratory fitness at baseline», with the subgroup figures in a supplement not held (Laukkanen et al., 2015). Fitness is already in the 2015 model, so a stronger association among the least fit does not test fitness confounding either way. Both papers ran within-study interaction tests on prespecified characteristics that include diabetes. 2015: subgroup analyses «were con- ducted using interaction tests to assess statistical evidence of any differences in hazard ratios across levels of prespecified individual-level characteristics, such as age at survey, history of diabetes» (Laukkanen et al., 2015). 2018: tests across «pre-specified clinically relevant characteristics» including history of diabetes, with results printed in Fig 5 (Laukkanen et al., 2018). Side by side:
| Parameter | 2015 (men 42-60) | 2018 (men + women 53-74) | Same quantity? |
|---|---|---|---|
| What the held text gives | a narrative summary: «associations seem to be somewhat stronger in … men with type 2 diabetes», after stating results were consistent among subgroups; the test results are in a supplement not held | printed estimates and test: diabetes the one significant modifier of frequency (Fig 5, P for meta-regression 0.021) | NO — a narrative summary of unheld test results vs printed results |
| Contrast | not given in the held text | 4-7 vs once/week: no diabetes 0.26 (0.11-0.62), 146 deaths; diabetes 1.59 (0.33-7.58), 35 deaths in 138 people | NO |
| Outcome | SCD / fatal CHD / fatal CVD (statement covers all) | fatal CVD | partly |
| Population | men, 1984-89 baseline | men + women, 1998-2001 baseline, overlapping men | NO |
No cell reads yes, so this cannot be filed as a tension; it is held as a gap. In narrative terms the directions differ (2015: «somewhat stronger» with diabetes; 2018: a point estimate above 1 with diabetes and below 1 without), but the 2015 test results are not held, so whether the two actually conflict cannot be checked here — a gap, not a finding. The 2018 authors warn that «findings from the subgroup analyses should be interpreted with caution given the multiple statistical tests of interaction and the low event rates in these subgroups»; the 2015 tests ran across a similar list of characteristics, so the same caution plausibly applies to them, though the 2015 authors do not say so. Neither paper’s diabetes reading is offered here as evidence about confounding. (inferred from Laukkanen et al., 2015, 2018)
What the checks do not remove:
- Residual confounding. 2015: «potential for residual confounding remains as with all observational studies» (Laukkanen et al., 2015). 2018: «Though we accounted for many potential confounders to ensure the validity of our associations, there is a potential for residual confounding.» (Laukkanen et al., 2018) A single cohort of one population cannot exclude an unmeasured disposition (a “healthy lifestyle” bundle) that drives both sauna use and survival. Both papers answer the reverse-causation worry with consistent subgroup results, and the 2018 paper answers confounding with the salary, SES and physical-activity arguments above; those are arguments, not identification.
- Reverse causation is named, not excluded. 2018: «It is possible that underlying diagnosed or undiagnosed diseases may have an effect on sauna bathing habits, suggesting reverse causality; however, our subgroup analyses according to various clinical characteristics were consistent and the associations remained robust in several sensitivity analyses, independent of many underlying clinical conditions and exclusion of the first 5 years of follow-up.» (Laukkanen et al., 2018) The sensitivity tables are supplementary and not held; consistent subgroups and a 5-year exclusion narrow reverse causation but do not rule out a slow-developing illness that changed habits before baseline. The 2018 subgroups were also not uniformly consistent: the diabetes subgroup differed (meta-regression P 0.021, below), which sits awkwardly with the consistent subgroups defence even if it is a chance finding.
- Exposure measured once, by self-report. 2015: «We could not correct for regression dilution bias, which may have underestimated the associations found» (Laukkanen et al., 2015). 2018: «exposure assessments based on self-administered questionnaires are prone to misclassification and recall bias»; habits may have drifted over follow-up, which the authors expect to be «minimal as sauna habits are fairly stable in the Finnish population» (Laukkanen et al., 2018). The same self-report / single-timepoint fragility that Measurement Error in Dietary Assessment makes central applies here: non-differential error in one exposure tends to pull estimates toward the null, but recall bias, or error correlated with covariates, can move them in either direction, so the data alone secure neither the point estimates nor, strictly, the direction.
- No non-user comparator. The authors note Finns are accustomed to a sauna bath «regularly at least once per week», and the referent is once/week, not none. 2015 excluded its non-users: «Twelve men who did not use a sauna were excluded» (Laukkanen et al., 2015). 2018 ran a subsidiary analysis with a 0-1/week group «including participants who did not use sauna at all (n = 43)» and reports that against 4-7/week «the associations were similar» (Table S3, not held) (Laukkanen et al., 2018). With 43 non-users folded into a group of roughly 450-500, that is not a never-user referent. Where those 43 sit in the main analysis is unclear: the three main frequency groups (455 + 1028 + 205) add up to the full 1688, so they may be inside the once/week referent rather than excluded.
Honest read: in men the inverse association survives adjustment for fitness/SES (2015) and a reported early-death exclusion in both papers (numbers in unheld supplements). These are the weak checks only. Against that: one cohort, one research group, residual confounding, single-timepoint self-report, a 2018 top category of 8 deaths with men’s lower categories not separating from 1 in Model 3, and no fitness adjustment in the sample that adds women. In women the estimates neither confirm nor refute the association: the 2-3/wk estimate is null-centred and wide, and the 0/73 cell points the same way but carries no estimate. That keeps it well short of a causal dose-response the design can support.
Proposed mechanism — directional, not established
The 2015 authors propose sauna heat loads the cardiovascular system like light-to-moderate exercise: heart rate may rise up to 100/min in moderate and up to 150/min in more intense sessions, and they describe «proposed functional improvements» corresponding to exercise benefits in blood pressure and left-ventricular function. Separately, they cite improved endothelial function «in patients with CHD risk factors and heart failure», ejection fraction and exercise capacity in heart failure, and a study in which regular sauna «lowers blood pressure, especially in patients with hypertension». (Laukkanen et al., 2015) The supporting studies the 2015 paper cites are largely in patients, so their transfer to healthy people is not shown by them; the 2018 paper’s own mechanism discussion is not appraised here. This is a mechanism (marked as such), not an outcome finding: the causal chain from an acute hemodynamic response to a 20-year mortality difference is not demonstrated, and whole-organism compensation could blunt a naive mechanism. It gives the association directional plausibility, discounted; it does not upgrade it. See Blood Pressure Lowering and Cardiovascular Events for the BP channel and Physical Activity Dose and Mortality for the exercise comparator the authors lean on.
Sauna frequency as a risk-prediction marker — surrogate metrics
Source: (Laukkanen et al., 2018)
Adding sauna frequency to a model of age, sex, SBP, diabetes, total cholesterol, HDL-C and smoking moved the C-index from 0.7716 to 0.7807 (+0.0091, 0.0022-0.0160, P = 0.010); categorical NRI over 10-year CVD-mortality categories (<1%, 1-<5%, >=5%) 4.14% (1.30-6.97%); IDI 0.0037 (0.0002-0.0072). These measure how well a model sorts people, not what sauna does to anyone’s risk, and they were estimated in the same data that fitted the model, with no external validation reported. They are not an effect size and do not rank the lever. Read against ESC’s five criteria for a risk modifier in Risk Modifiers - When Extra Information Changes a Risk Estimate, it meets the first in-sample only, reports nothing on public-health impact, and cannot be judged for publication bias with one cohort, so it does not clear that bar. (inferred from Laukkanen et al., 2018)
What this changes at the margin
For a reasonably healthy person the big rocks (not smoking, weight/adiposity, activity, sleep, alcohol) dominate; a sauna habit is a low-certainty add-on whose causal size is unknown, not a substitute for any of them. The 2015 authors call it «a recommendable health habit, although further studies are needed to confirm our results in different population settings» (Laukkanen et al., 2015); the same paper calls sauna «a protective factor», causal language its single-cohort design does not support. The 2018 authors generalize to «Northern European populations» and ask for studies «in populations who are not accustomed to regular sauna bathing» (Laukkanen et al., 2018).
By stratum:
- Men, middle-aged to older, Finnish-style dry sauna: the low-confidence inverse association applies. In the 2018 Model 3 only the men’s 4-7/week category separates from 1 (0.39, 0.18-0.84, 8 deaths); 2-3/week (0.69, 0.46-1.03) and >45 min/week (0.57, 0.31-1.05) do not, though both exclude 1 in the age-adjusted Model 1 (0.61, 0.41-0.90; 0.50, 0.27-0.91). The 2018 age split, which pools both sexes, gives a non-significant estimate above 65.2 with no detected difference between the halves (P 0.287). (Laukkanen et al., 2018) In the larger 2015 men-only analysis, 2-3/week did separate from once/week for fatal CHD, fatal CVD and all-cause death (table above), but not for SCD. (Laukkanen et al., 2015) No target dose — 4-7/week is the studied edge.
- Type 2 diabetes: insufficient evidence, held as a hypothesis. In 2018, a within-study test on a pre-specified characteristic, the 4-7 vs once/week estimate (Fig 5, adjusted for the Model-2 covariates) was 1.59 (0.33-7.58) with diabetes against 0.26 (0.11-0.62) without (P for meta-regression 0.021; 35 deaths in 138 people with diabetes, of whom only 13 were in the 4-7/week group, Table 1); the 2018 authors warn about multiple tests and low event counts. (Laukkanen et al., 2018) The 1.59 therefore rests on very few people at the top frequency; the insufficient, held as a hypothesis label is this page’s. The 2015 paper’s own prespecified diabetes test is summarized only in words, as «somewhat stronger» with diabetes, with its numbers in an unheld supplement, so the two cannot be checked against each other. (inferred from Laukkanen et al., 2015)
- Women: insufficient evidence — studied once, frequency association not detected, duration estimates imprecise, too few deaths to tell. Absent a detected sex difference (interaction P 0.524), the sex-pooled estimate stays the default reference (though all 8 deaths in its top frequency category are men’s), and the women’s own data neither confirm nor refute it.
- People prone to orthostatic hypotension: both papers give the same caution. 2015: «individu- als who are prone to orthostatic hypotension should be cautious when sauna bathing because of a possible decrease in blood pressure» (Laukkanen et al., 2015); 2018, for the subset of older people so prone: «How- ever, in a specific group of older individuals who are prone to orthostatic hypotension, sauna baths should be taken cautiously due to possible sudden drop in blood pressures which may occur just after a hot and dry bath [22, 30, 56].» (Laukkanen et al., 2018). An asserted caution, not a measured harm rate.
Harm: neither cohort measured it. 2015: previous studies «suggested that sauna bathing might have some harmful effects», which the authors set against their own result; they suggest «The possible differences in results obtained in Finnish studies compared with studies from other countries» (refs 2, 12) «may be at least partly due to the different conditions and temperature of the sauna»; then «Dry sauna bathing seems to be safe», with cited studies of cardiac patients, the orthostatic caution above, and the cited figure that «Only 1% to 2% of sudden deaths occurred within 24 hours of sauna bathing, with most of them being unintentional, and alcohol intake together with sauna bathing has been a major contributing factor for these adverse events» (Laukkanen et al., 2015). 2018: «Regular Finnish sauna bathing is safe and may have several additional health benefits» (no citation on that sentence), then a cited sentence that cardiac patients «can usually enjoy sauna bathing without any significant adverse cardiovascular effects [5, 22, 55]», then its own 4-7/week CVD mortality of 2.7 per 1000 person-years read as «indicating low risk», then the orthostatic caution, followed by the statement that hypotension during and after a sauna «can be easily prevented by appropriate fluid intake to avoid dehydration» (Laukkanen et al., 2018). The safety readings rest on cited studies in cardiac patients, plus in 2018 an uncited sentence and the cohort’s CVD death rate, which does not isolate sauna-related events. The 1-2% figure has no comparator (how many deaths would fall within 24 hours of a sauna by chance among weekly bathers), so it is neither a harm rate nor a safety demonstration. The specific risks the papers name are orthostatic or dehydration-related drops in blood pressure and alcohol. The harm state is no harm measured here, not safe.
Labelled practical default (standard practice, not trial-tested): for someone with access to a Finnish-style sauna, using it is a reasonable low-cost habit if they enjoy it, with no target frequency or duration. What was studied, as a description and not a target: 1-7 sessions/week of a hot dry sauna at a mean of about 76-79°C, with sessions averaging about 14 minutes in 2015 (median 13 minutes in 2018, Table 1). Skip alcohol before and during, drink fluids, and take care standing up afterwards if prone to dizziness on standing. For someone already saunaing it is not a lever to drop. It does not earn a high rank on the intervention hierarchy, and confidence stays low by construction.
Gaps (acquisition targets, not silences)
- No replication cohort outside KIHD — a second independent prospective cohort is the single most valuable missing source; until one is held, the wiki’s evidence here is un-triangulated. No specific independent cohort is named yet; the one named review candidate is the systematic review by authors outside the KIHD group, above, whose outcomes are unverified.
- Women: insufficient evidence — one KIHD sample of 867 women, 63 deaths, none at 4-7/wk; the 2015 authors asked for female cohorts and the 2018 sample is too small to answer. Needs a cohort with enough events in women, outside the KIHD group.
- No all-cause mortality beyond men 42-60 — the 2018 analysis reports fatal CVD only.
- No non-Finnish population — cannot separate the sauna exposure from Finnish lifestyle/genetics.
- No RCT — no randomized heat-exposure trial on hard endpoints (feasibility aside, none is held).
- Participant overlap between the two KIHD analyses is unreported, so how much of the 2018 men’s signal re-observes the 2015 men is unknown. For any man in both, a CVD death between the 1998-2001 visit and the end of 2011 is counted in both analyses (2015 deaths ran «by the end of 2011»; 2018 follow-up ran to 2015), so some of the same events, not only the same people, likely recur.
- Transport to other sauna types is unestablished — results «cannot be directly applied to steam rooms, hot tubs, and some other types of saunas» that run cooler and do not allow water on the rocks. The exposure is specifically the hot dry Finnish sauna (mean 79°C in 2015, 75.9°C in 2018). (Laukkanen et al., 2015) (Laukkanen et al., 2018)