Diets · Metabolic & Cardiometabolic
low heart rate variability correlates with all-cause mortality
In plain terms: Does low heart-rate variability signal a higher risk of dying?
Strong support Diets 💰 Industry COI noted🔬 Includes disconfirming
RefutedContestedStrong support
consensus score 0.89
Yes — a strong association, but it flags risk rather than being proven to cause it.
📅 Last reviewed: 2026-07-14 ⓘ
Evidence ladder
How far up the ladder this claim has climbed. A high consensus on a low rung means "consistent so far," not "proven in people."
Top evidence so far: All trials, pooled (Meta-analysis)
MechanismIn-vitroAnimalObservationalRCTMeta-analysis
How the studies fall
5 support 0 contradict 1 tested null 0 mixed · 6 sources, 5 independent groups
What the evidence shows
Low heart rate variability predicts death, and the evidence spans two populations rather than one. The landmark finding is post-MI (Kleiger 1987, replicated post-PTCA), but the larger share of the evidence here is general-population cohorts of adults without known heart disease — ARIC, Framingham and Zutphen — where low HRV also tracks mortality.
The evidence (12)
| Source | Grade | Stance | Quality | Finding |
|---|---|---|---|---|
| Jarczok 2022 · Neurosci Biobehav Rev | meta-analysis | supports | high | Systematic review + meta-analysis, 32 studies plus two individual-participant datasets: 'A total of 32 studies and two individual participant datasets (IPD) with 37 samples and 38,008 participants were included.' ENDPOINT — all-cause mortality is an endpoint AS SUCH, which is what separates this row from the three neighbours held off-scope on this claim: 'The aim of the present comprehensive meta-analysis is to examine eight different HRV parameters to determine their association with all-cause and cardiac mortality.' and 'IPD sub-analysis for heart rate corrected HRV parameters confirmed the strong association between HRV and all-cause mortality.' Cardiac mortality is named only as an aim; no separate cardiac-mortality estimate is reported at abstract level, so the prior extract's 'all-cause and cardiac' credited an aim as a result. DIRECTION: 'Lower HRV parameter values were significant predictors of higher mortality across different ages, sex, continents, populations and recording lengths.' WHICH METRIC — eight parameters were pooled but only ONE is named in the abstract, and it is the index that goes null elsewhere on this claim: 'Sub-analyses of studies comparing the lowest quartile of 5-min root mean square of successive differences (RMSSD) vs. the other quartiles yielded a combined HR of 1.56 (95% CI: 1.32-1.85).' Pooled RMSSD is therefore significant here, while r-MSSD was null within ARIC (s27551828, 1.06, 0.92-1.22, p=0.43) and within the Giessen post-MI cohort (s19214416, 'RMSSD was unrelated to the long-term vital status') - the single-cohort nulls are plausibly power. SDNN, LF and HF are not named in the abstract, so no per-index result and no within-paper null can be attributed to this paper; the hedge must be kept as written - 'Most of the examined parameters showed comparable hazard ratios (HR).' - 'most', not all. Note the RMSSD figure is a lowest-quartile-vs-rest contrast, not comparable to the per-SD increments (1.16-1.27) the cohort rows report. ADJUSTMENT: 'Meta-regressions revealed no effect modifier for HRs extracted from covariate-adjusted studies.' POPULATION: 'healthy and patient populations' - inside this claim's post-MI AND general-adult scope but BROADER than it; the citation base spans haemodialysis, cirrhosis, heart failure and cancer HRV-mortality literature, so the pooled figure is a general estimate, not a post-MI or a general-population one. INDEPENDENCE CAUTION: the 32 included studies are not named in the abstract, so overlap is cannot-tell from available text - but this claim already holds four ARIC rows on the same visit-1 1987-89 two-minute recording plus Framingham and Zutphen, and this paper's Europe PMC reference list contains PMID 9149661 (Dekker 1997 Zutphen = s9149661) and PMID 12453910 (Liao 2002 ARIC = s12453910) by exact PMID, plus PMID 26047073 (Wulsin 2015, a Framingham Heart Study analysis - same cohort as s8941112, different paper). A reference is not proof of inclusion, so double-counting is likely but unproven; curation needs the included-study table. group_id 'ulm' tracks the AUTHORS' institution, not the pooled cohorts - the same defect already flagged on s23370966. Read at ABSTRACT grade only (NCBI efetch, 2026-08-20): closed access with pmcid null / isOpenAccess N in Europe PMC, no PMC record via elink, NCBI BioC returns no result, and Unpaywall and OpenAlex both report oa_status 'closed' with no repository full text. None of the pooling internals (heterogeneity, publication bias, risk of bias, included-study list) could be verified. |
| Perkiomaki 2011 · Front Physiol | observational | supports | low | Single-author NARRATIVE REVIEW (Front Physiol; PMC JATS article-type 'review-article') with no cohort, no search protocol and no pooling - every n it quotes belongs to someone else, so it carries zero weight under evidence_role: review. On the claim's exact endpoint it is affirmative and unambiguous: 'There are abundant scientific data about the prognostic significance of the conventional measurements of HRV in patients with various conditions, particularly with myocardial infarction', and ALL-CAUSE mortality is named as such twice - short-recording spectral measures 'predict all-cause mortality and sudden cardiac death' (Bigger 1993) and a steep power-law slope is 'a better predictor of all-cause mortality or arrhythmic death' (Bigger 1996). Metrics that carry the finding here: SDNN, HRV triangular index, ULF/VLF/total power, power-law slope, and the short-term DFA scaling exponent; RMSSD is never mentioned in the paper, so no RMSSD credit can be taken from it. The review's own caveats: 'The predictive power of HRV variables varies also depending on whether total mortality, different modes of mortality, or other adverse events are selected as endpoints'; in REFINE, HRV 'had no association with the primary endpoint when measured from 2 to 4 weeks after the AMI'; and 'none of these methods are in widespread clinical use at the moment'. Oulu/Huikuri group and heavily self-citing (Huikuri is the handling editor) - independent of ARIC and the Liao group, but not an independent synthesis of its own programme. No commercial COI declared. |
| Erdogan, Coch, Bilgin, Parahuleva, Tillmanns, Waldecker, Soydan 2008 · Herzschrittmacherther Elektrophysiol 2008 Dec;19(4):161-8 | observational | supports | moderate | Prospective single-centre Giessen cohort, n=412 consecutive AMI patients reperfused by direct PTCA, HRV measured in the whole cohort 11+/-9 days post-MI; 31 deaths (24 cardiac) over mean 4.3+/-3 y. SDNN predicted ALL-CAUSE mortality: 'SDNN was higher in long-term survivors (102+/-39 ms) as compared to nonsurvivors (81+/-33 ms, p=0.02) but RMSSD was unrelated to the long-term vital status'; 4-year survival with SDNN <50 vs >50 ms was 80% vs 92% (p<0.001), and low SDNN was independent of LVEF and age (OR=2.0, p<0.05). RMSSD is a within-paper null. Authors' own caveat: 'Only 3/31 fatalities and 1/24 cardiac deaths were predicted by a SDNN <50 ms' — real association, poor positive predictive value. Severely depressed HRV was uncommon in the reperfusion era (SDNN <50 ms in 7%). |
| Tsuji 1996 · Circulation | observational | supports | high | Framingham community cohort, n=2501 free of clinical CHD/CHF, mean age 53, mean follow-up 3.5 years, 58 events. HRV measured from 'the first 2 hours of ambulatory ECG recordings'. Endpoint is a COMPOSITE, not mortality: 'the incidence of new cardiac events (angina pectroris, myocardial infarction, coronary heart disease death, or congestive heart failure)' - CHD death is one of four components and is not broken out. 'A one-standard deviation decrement in the standard deviation of total normal RR intervals (natural log transformed) was associated with a hazard ratio of 1.47 for new cardiac events (95% confidence interval of 1.16 to 1.86)', after adjustment for age, sex, cigarette smoking, diabetes, left ventricular hypertrophy and other risk factors. All HRV measures except the LF/HF ratio were significantly associated (P = .0016 to .0496). Landmark community cohort; supports composite cardiac risk, cannot isolate mortality from the abstract. |
| Hillebrand, Gast, de Mutsert, Swenne, Jukema, Middeldorp, Rosendaal, Dekkers 2013 · Europace. 2013 May;15(5):742-9 | meta-analysis | supports | high | OFF-SCOPE for an all-cause-mortality claim: the endpoint is a fatal-OR-non-fatal cardiovascular composite and the population is people WITHOUT cardiovascular disease, the inverse of this claim's post-MI scope. 'Primary outcomes were (non)fatal cardiovascular events.' 'Eight studies with a total number of 21 988 participants were included.' On the endpoint it did measure the direction is positive but index-dependent: 'The pooled relative risk (RR) comparing the lowest level to the highest level of SDNN was 1.35 (95% CI 1.10, 1.67).' and 'The pooled RRs for LF and HF were 1.45 (95% CI 1.12, 1.87) and 1.32 (95% CI 0.96, 1.81), respectively.' - i.e. the parasympathetic index (HF) is NOT significant, its CI crossing 1. The abstract's headline range therefore spans a null bound: 'In conclusion, low HRV is associated with a 32-45% increased risk of a first cardiovascular event in populations without known CVD.' - the '32' is the HF RR of 1.32 (95% CI 0.96, 1.81) and only the '45' (LF, 1.45) and the SDNN 1.35 are significant. The prior extract inherited that 32-45% wording without the caveat. A dose-response gradient is present but at study level, not within-person: 'In a meta-regression, the predicted RR of incident CVD of the 10th and 90th HRV (SDNN) percentiles compared with the 50th percentile were 1.50 (95% CI 1.22, 1.83) and 0.67 (95% CI 0.41, 1.09).' and 'An increase in SDNN of 1% results in an ∼1% lower risk of fatal or non-fatal CVD.' Independence caution: this pools eight CVD-free prospective cohorts using SDNN/LF/HF up to Dec 2011, which is the same literature the vault already appraises directly on this claim via ARIC (s10982537, s9125996, s27551828, s12453910), Framingham (s8941112) and Zutphen (s9149661); the abstract does not name the eight included studies, so the exact overlap is cannot-tell from available text, but double-counting is likely and group_id 'leiden' tracks the AUTHORS' institution, not the pooled cohorts. Read at abstract grade only (PubMed efetch, 2026-08-20); the paper is subscription-locked with pmcid:null / isOpenAccess:N in Europe PMC, though a free publisher PDF exists via Unpaywall. |
| Liao 1997 · Am J Epidemiol | observational | supports | moderate | ARIC case-cohort (137 incident CHD cases + stratified random subcohort of 2,252 CHD-free examinees; 3-year follow-up). Endpoint is NOT mortality: 'Incident CHD was defined as hospitalized myocardial infarction, fatal CHD, or cardiac revascularization procedures during 3 years of follow-up' - fatal CHD is one of three components and no mortality-only estimate is given. Exposure was 'Baseline, supine, resting beat-to-beat heart rate data' (short-term resting spectral/time-domain HRV, not 24-hour ambulatory; recording duration cannot-tell from available text). Adjusted (age, race, gender, other CHD risk factors) relative risks for lowest quartile vs upper three quartiles: HF power 1.72 (95% CI 1.17-2.51), LF power 1.09 (0.72-1.64), HF/LF ratio 1.25 (0.84-1.86), SDNN 1.39 (0.94-2.04) - only the parasympathetic HF index was significant. Abstract-grade. |
| Maheshwari, Norby, Soliman, Adabag, Whitsel, Alonso, Chen 2016 · PLoS One | observational | supports | high | OFF-SCOPE on outcome: this claim's object is ALL-CAUSE MORTALITY, and the paper never reports it. 'The outcome of our study was SCD.' - the sole endpoint is physician-adjudicated sudden cardiac death drawn from fatal CHD events alone: 'In order to identify SCD, all fatal CHD events through 2001 in ARIC were reviewed by an independent panel of physicians.' 'In our cohort of 12,543 participants, we identified 215 SCD events over a median follow-up period of 13 years.' The introduction itself treats all-cause mortality as prior work by others and SCD as the gap it fills: 'In the general population, low HRV has been linked to increased risk of cardiovascular events and all-cause mortality [9-11]; however, the relationship between low HRV and SCD risk is less well-characterized.' ON ITS OWN ENDPOINT the finding is real and the prior extract understated its range: 'In multivariable adjusted Cox proportional hazards models, each standard deviation decrement in SDNN, LF, and HF were associated with 24%, 27% and 16% increase in SCD risk, respectively' - i.e. 16-27%, not 16-24%. WHICH METRIC: SDNN 1.24 (1.06-1.46) p=0.009, LF 1.27 (1.12-1.43) p=0.0002, HF 1.16 (1.02-1.32) p=0.02 per 1-SD decrement (Model 2); the fourth index was NULL and the prior extract omitted it - 'r-MSSD was not significantly associated with SCD when analyzed in Model 2 as a continuous variable' (1.06, 0.92-1.22, p=0.43), and r-MSSD was the only index not differing at baseline (29.2 vs 27.3 ms, p=0.25). In the low-risk cohort excluding prevalent CHD/heart failure (n=11,531) both r-MSSD (1.05, 0.88-1.26) and HF (1.12, 0.96-1.31) went null; only SDNN (25%) and LF (24%) survived. DENOMINATOR: HRV attempted cohort-wide, not in a subsample - 15,792 enrolled minus 3,131 excluded for HRV-recording reasons (812 missing, 210 short, 636 poor quality, 1,473 too few acceptable RR intervals) plus 118 other, 'resulting in a final cohort of 12,543 participants'; ~157,000 person-years. Recording = a single supine 2-minute resting ECG at ARIC visit 1 (1987-89) after 20 min supine, 8:30-11:30 AM. POPULATION: general biracial community cohort aged 45-64, not post-MI - inside the claim's population scope; the exclusion is purely on endpoint. INDEPENDENCE: fourth ARIC appraisal on this claim and the same exposure measurement as Dekker 2000 (s10982537, visit-1 case-cohort through 1993), Liao 2002 (s12453910, visit-1 HRV, ~8y) and Liao 1997 (s9125996); this analysis's 1987-2001 person-time subsumes all three - they are one cohort, not four. QUALITY high on design: prospective, n=12,543, median 13 y, 15-covariate adjustment, PH assumption verified, NHLBI-funded, 'Competing Interests: The authors have declared that no competing interests exist.' Authors' own limit: 'The population attributable risk, however, is modest and the clinical utility, if any, of our findings is not established and will need to be clarified.' |
| Liao 2002 · Diabetes | observational | tested-null | moderate | ARIC (n=11,654, ~8y): on this claim's endpoint - death - low HF-HRV was NOT significantly associated even in diabetics: HR 1.50 (0.65-3.42) for fatal CHD and 1.27 (0.84-1.91) for non-CHD deaths (lowest quartile vs upper three). All-cause mortality was never reported as a single endpoint. The paper's significant results were morbidity, not mortality: incident MI 2.03 (1.28-3.23) and incident CHD 1.60 (1.12-2.27) in diabetics, with 'no consistent pattern of associations among individuals without diabetes'. |
| Dekker 2000 2000 · Circulation | observational | supports | high | [abstract] ARIC prospective case-cohort, 14,672 adults aged 45-65 free of CHD at baseline; HRV measured from a 2-minute rhythm strip in a case-cohort subsample (random n=900 + all 395 incident CHD + all 443 deaths through 1993). 'Subjects with low HRV had an adverse cardiovascular risk profile and an elevated risk of incident CHD and death. The increased risk of death could not be attributed to a specific cause and could not be explained by other risk factors.' No effect size available at abstract level. |
| Dekker 1997 · Am J Epidemiol | observational | supports | moderate | Zutphen Study, Dutch men (878 middle-aged followed 1960-1985; 885 elderly 1985-1990). HRV = SDNN from the resting 12-lead ECG ('brief electrocardiographic recordings'), not ambulatory Holter. All-cause mortality IS the paper's outcome: '5-year age-adjusted relative rate of total mortality of men with heart rate variability of < 20 milliseconds (msec) compared with men with heart rate variability of 20-39 msec was 2.1 (95 percent confidence interval 1.4-3.0) in middle-aged men and 1.4 (95% confidence interval 0.9-2.2) in elderly men' - significant in the middle-aged, null in the elderly. The cause-specific story runs the other way: 'The association of low heart rate variability with sudden death or coronary heart disease mortality was less consistent', while 'Death from noncoronary causes, especially cancer, contributed significantly to this elevated risk.' Authors read low HRV as 'an indicator of compromised health' rather than a cardiac-specific risk. Adjustment is age only; men-only cohort; abstract-grade read (not in PMC OA). |
| Kleiger RE, Miller JP, Bigger JT Jr, Moss AJ 1987 · Am J Cardiol | observational | supports | moderate | Post-MI cohort n=808, 24h SDNN at 11+/-3 d, mean follow-up 31 months: 'The relative risk of mortality was 5.3 times higher in the group with HR variability of less than 50 ms than the group with HR variability of more than 100 ms' — an unadjusted tail-vs-tail contrast excluding the 50-100 ms band. Separately, HRV 'remained a significant predictor of mortality after adjusting for clinical, demographic, other Holter features and ejection fraction', with no adjusted magnitude reported. Origin of the finding, not its effect size. |
| Yadav I, Waqas R, Mohammad A, Lashari UG, Sabra M, Dwayat A, Rajput J 2025 · Cureus | meta-analysis | supports | low | OFF-SCOPE on endpoint and on composition, read from full text (local PMC JATS XML, PMC12794729). ENDPOINT: no all-cause-mortality estimate is produced anywhere in the paper. The single pooled figure aggregates all-cause mortality with cardiac death, a CV-death/HF-hospitalisation composite, vascular-access failure and studies reporting no mortality at all - Table 4's own 'Mortality Outcomes' column reads 'All-cause mortality', 'Cardiac mortality', 'Cardiac death', 'CV death/HF hospitalization', 'Vascular access failure' and 'Not reported' across the ten rows. COMPOSITION: six of the ten pooled 'studies' cannot bear on low-HRV-to-death. Ref 21 (Mentz, VICTORIA) is 'Recurrent hospitalizations and response to vericiguat in heart failure with reduced ejection fraction.' and its HRV parameters are listed as 'Not specified'; ref 22 (Huang) is 'Heart rate variability during hemodialysis is an indicator for long-term vascular access survival in uremic patients.'; ref 24 (Melin) is 'Variability in physical activity assessed with accelerometer as an independent predictor of mortality in CHF patients.' - accelerometer activity variability, not HRV, tabulated as 'Physical activity HRV'; ref 23 (Shehab) is an n=24 spironolactone/losartan crossover whose mortality outcome is 'Not reported' yet which carries ES 0.60; ref 25 (Bissinger) is 'Cardiac autonomic neuropathy: why should cardiologists care about that?', a narrative review with population 'N/A' carrying ES 3.50, the second-largest effect in the forest plot; ref 26 (El-Malahi) is itself a meta-analysis of 18 studies of exercise training, pooled as a primary study at ES 0.45. Only La Rovere 2012 (GISSI-HF) and Cygankiewicz 2008 pair an HRV exposure with an all-cause-mortality endpoint. WHICH METRIC: no index-specific result exists. SDNN and RMSSD are merged into one 'time-domain' bucket and LF/HF is merged with heart-rate turbulence, so this paper contributes nothing to the SDNN-vs-RMSSD split this claim turns on, and reports no index-level null. ABSTRACT CONTRADICTS RESULTS: the abstract credits SDNN with 'the strongest prediction (ES = 1.75, I² = 71.49%)', but Table 7 makes time-domain the SMALLEST of the three domains - 'Time-domain measures showed moderate effects (ES = 1.75) with lower heterogeneity (I² = 89.12%) compared to frequency-domain (ES = 2.66, I² = 92.16%) and composite measures (ES = 1.79, I² = 98.37%).' - with no significant difference between them: 'While between-subgroup differences were non-significant (p = 0.320), the prediction intervals revealed considerable variability, particularly for frequency-domain measures (PI: -9.12 to 14.45).' The 71.49% is imported from a different (population) subgroup, and the Discussion repeats the wrong figure. The abstract's sudden-death line ('particularly for sudden death (hazard ratio (HR): 2.1-3.2)') is not a pooled result - no sudden-death meta-analysis was run; 2.1 and 3.2 are two individual Table-4 cells. INTERNAL INCOHERENCE: three mutually inconsistent pooled estimates for the same 10 studies - 1.99 (1.36-2.61) in the abstract, 1.93 (1.24-2.62) in Table 6 and 'The overall combined effect across all studies was 1.77 (95% CI: 1.59-1.96)' in Figure 7; Table 6 simultaneously reports 'Correlation 1.93' and 'Effect size (correlation) 0.31', and a correlation of 1.93 is impossible; hazard ratios, relative risks, odds ratios and a standardized mean difference are pooled on one scale. The main model's prediction interval crosses the null: 'The wide prediction interval (-0.37 to 4.23) and high heterogeneity statistics (I² = 97.67%, T² = 0.94) indicated substantial variability between studies, suggesting caution in generalizing the findings.' PUBLICATION BIAS: Egger's regression is significant - 'showing a significant intercept (estimate = 9.81, p = 0.045) but a non-significant slope' - yet the paper reads the slope as the test and the limitations state the opposite: 'Publication bias might favor studies reporting significant HRV-mortality associations, though Egger’s test did not detect strong asymmetry.' POPULATION is heart failure, a third population outside this claim's declared post-MI and general-adult scope. PRISMA counts do not reconcile: 2,059 screened minus 1,904 excluded leaves 155, while 'Full-text retrieval was attempted for 140 articles, but none were inaccessible.' and only 15 full-text reports were assessed. Venue Cureus (author-pays); 'All authors have declared that no financial support was received from any organization for the submitted work.'; the listed corresponding author's e-mail does not match his name. Direction as reported is supportive, but nothing here is a usable all-cause-mortality estimate. |
Disagree, or know a study we missed?
We grade by evidence, not opinions. The way to weigh in is to point us to a study we haven't cited (check the evidence table above first), or to flag a problem with one we have. Every submission is reviewed; if it holds up, the grade updates and shows in Science Changes Its Mind.
Opens a short form. You'll sign in with Google so submissions are tied to a real account — we don't display your identity, and we only accept a link we can verify (PubMed, DOI, ClinicalTrials.gov).
Educational only, not medical advice. Grades and scores reflect published evidence weighted by study design and quality; see the methodology.