← Back

Longevity and aging

Everything in the library studied for it, medical and holistic together. Within each kind, what was shown to help comes first, strongest evidence first. What was tested and not shown to help stays on the list.

Strong several reviews agree · Moderate one review or consistent trials · Limited one or a few small studies · Untested no study in people, listed with what it rests on

Medication11 Hormone therapy2 Peptides and SARMs3 Supplements29 Practices5 Daily life16 Food and drink14 Diets8 Lab markers76

Medication

Inhaled corticosteroidsStrong

Triple inhalers in COPD

Shown to cut moderate or severe COPD exacerbations against dual inhalers, and to lower deaths against two bronchodilators without a steroid, but not against a steroid-bronchodilator pair.

Compared with a long-acting antimuscarinic alone, a long-acting antimuscarinic with a long-acting beta-agonist, or an inhaled steroid with a long-acting beta-agonist [11][12][13]

The numbers and sources

Effect: Triple therapy lowered moderate or severe exacerbations against an antimuscarinic alone (rate ratio 0.71), two bronchodilators (0.78) and a steroid-beta-agonist pair (0.77). Pneumonia was higher than with two bronchodilators (RR 1.53) [11]. Over 52 weeks, deaths were lower than with two bronchodilators (RR 0.69, 95% CI 0.53 to 0.90) but not than with a steroid-beta-agonist pair (RR 0.94, 95% CI 0.72 to 1.24) [12]. A 2025 analysis also found fewer deaths than with two bronchodilators (RR 0.76, 95% CI 0.60 to 0.97) [13].
Population: 21 trials [11]; 6 trials with 22,669 patients [12]; randomized trials in moderate to severe COPD [13].
Sources: [11], [12], [13]
Caveats: Most trials enrolled people with advanced COPD and recent exacerbations [11]. The three analyses share many trials.

StatinsStrong

Heart attack, stroke and death

Shown to cut major vascular events about 22% and deaths about 10% for each 1 mmol/L drop in LDL cholesterol, including in lower-risk people.

The numbers and sources

Effect: Per 1.0 mmol/L reduction in LDL cholesterol, major vascular events fell with a rate ratio of 0.78 (95% CI 0.76 to 0.80) and all-cause mortality with 0.90 (95% CI 0.87 to 0.93). More intensive regimens cut major vascular events a further 15% (95% CI 11 to 18) against less intensive ones [1]. In people with a 5-year risk under 10%, each 1 mmol/L reduction cut major vascular events by about 11 per 1,000 over 5 years. In people without vascular disease, all-cause mortality RR was 0.91 (95% CI 0.85 to 0.97) per 1 mmol/L [2].
Population: Individual data from 26 trials with about 170,000 participants, median follow-up about 5 years [1]; 27 trials with 174,149 participants, grouped by baseline risk [2].
Sources: [1], [2]
Caveats: Both analyses come from the same collaboration, and the same 5 intensity trials appear in both. Effects are expressed per unit of LDL lowered, not per drug or dose.

ADHD stimulantsLimited

Death rates in people with ADHD

Linked to lower two-year mortality in people starting ADHD medication after diagnosis, 39.1 against 48.1 deaths per 10,000, in one Swedish cohort.

The numbers and sources

Effect: Two-year mortality was 39.1 per 10,000 in people who started ADHD medication and 48.1 per 10,000 in those who did not (HR 0.79, 95% CI 0.70 to 0.88). Deaths from unnatural causes were lower (HR 0.75, 95% CI 0.66 to 0.86), and deaths from natural causes did not differ (HR 0.86, 95% CI 0.71 to 1.05) [20].
Population: 148,578 people aged 6 to 64 with newly diagnosed ADHD in Sweden, 56.7% of whom started medication [20].
Sources: [20]
Caveats: Observational, and it covers all ADHD medications, not stimulants alone. People who start medication may differ from those who do not in ways the analysis could not measure.

BisphosphonatesLimited

New fractures and survival after a hip fracture

One trial found yearly zoledronic acid after hip fracture surgery lowered new fractures, 8.6% against 13.9%, and deaths, 9.6% against 13.3%, over about 2 years.

Compared with placebo, with calcium and vitamin D in both groups [7]

The numbers and sources

Effect: New clinical fractures occurred in 8.6% on zoledronic acid and 13.9% on placebo, a 35% reduction. New clinical spine fractures were 1.7% against 3.8%, and non-vertebral fractures 7.6% against 10.7%. Deaths from any cause were 9.6% against 13.3%, 28% fewer [7].
Population: 2,127 adults, mean age 74.5, given the first infusion within 90 days of hip fracture repair, median follow-up 1.9 years [7].
Sources: [7]
Caveats: One trial. The pooled analysis of mortality below did not confirm a drop in deaths across trials [8].
Why limited: This rests on one trial, of 2,127 people. Strength counts independent studies that agree, so one trial is limited at any size.

MetforminLimited

Longevity and diseases of aging

Limited evidence: observational studies disagree on mortality, and in a long trial follow-up, cancer and multiple chronic conditions did not differ from placebo in prediabetes.

The numbers and sources

Effect: A meta-analysis of 53 studies found people with diabetes on metformin had lower all-cause mortality than people without diabetes (HR 0.93, 95% CI 0.88 to 0.99) and than people with diabetes on other drugs (HR 0.72, 95% CI 0.65 to 0.80) [8]. A Danish registry study found the reverse. People starting metformin alone for type 2 diabetes had higher mortality than matched people without diabetes (adjusted IRR 1.32, 95% CI 1.16 to 1.50), and 1.64 (95% CI 1.10 to 2.46) against their own twin without diabetes [9]. In adults with prediabetes randomized in the Diabetes Prevention Program, cancer incidence did not differ between metformin and placebo over a median 21 years (HR 0.90, 95% CI 0.73 to 1.10) [10], nor did the risk of having two or more chronic conditions (HR 0.91, 95% CI 0.78 to 1.07) [11]. In a trial in metabolic syndrome, fewer major noncommunicable diseases on metformin were fully explained by less diabetes [3].
Population: Observational studies of people with type 2 diabetes [8][9], including discordant twin pairs [9]; 3,234 adults with prediabetes followed a median 21 years [10]; 1,173 of them with Medicare records [11]; 1,442 adults with metabolic syndrome [3].
Sources: [8], [9], [10], [11], [3]
Caveats: [8] and [9] are not randomized trials. The meta-analysis authors note baseline differences between groups that could bias results [8]. In the Diabetes Prevention Program follow-ups, metformin became open-label and use outside the study rose over time [10], and [11] is an observational follow-up of the trial cohort. No trial in people without diabetes measuring lifespan was sourced.

BisphosphonatesModerate

Death from any cause

Not shown to lower deaths across pooled placebo-controlled trials of bisphosphonates or of zoledronate alone.

Compared with placebo [8]

The numbers and sources

Effect: Bisphosphonate trials showed no significant association with overall mortality (RR 0.95, 95% CI 0.86 to 1.04), nor did zoledronate trials (RR 0.88, 95% CI 0.68 to 1.13), with some heterogeneity among the zoledronate trials (I² 48.2%) [8].
Population: 42,867 people in 21 bisphosphonate trials, 13,870 of them in 6 zoledronate trials [8].
Sources: [8]
Caveats: Trials were designed around fractures, not deaths. One zoledronic acid trial after hip fracture found fewer deaths [7].

Ezetimibe and fenofibrateModerate

Ezetimibe: death from any cause

Not shown to lower deaths from any cause or from heart disease when ezetimibe was added to a statin or fenofibrate.

Compared with a statin or fenofibrate alone [3]

The numbers and sources

Effect: All-cause mortality did not differ (RR 0.98, 95% CI 0.91 to 1.05), nor did cardiovascular mortality (RR 1.00, 95% CI 0.89 to 1.12) [3].
Population: 21,222 people in 8 trials for all-cause and 19,457 in 6 trials for cardiovascular mortality [3].
Sources: [3]
Caveats: Trials ran a median of up to 6 years.

Antiplatelets: clopidogrel and low-dose aspirinLimited

Healthy life span in adults over 70

One trial found daily aspirin did not lengthen survival free of dementia or disability in healthy adults aged 70 or over.

The numbers and sources

Effect: Over a median 4.7 years, death, dementia or persistent physical disability occurred at 21.5 events per 1,000 person-years on aspirin 100 mg and 21.2 on placebo (HR 1.01, 95% CI 0.92 to 1.11). Major hemorrhage occurred in 3.8% against 2.8% (HR 1.38, 95% CI 1.18 to 1.62) [16].
Population: 19,114 community-dwelling adults in Australia and the US, median age 74, without cardiovascular disease, dementia or disability [16].
Sources: [16]
Caveats: The trial was stopped at a median 4.7 years. Adherence was about 62% in the final year [16].
Why limited: This rests on one trial, of 19,114 people. Strength counts independent studies that agree, so one trial is limited at any size.

RapamycinLimited

Healthspan measures in healthy older adults

Small trials found no change in visceral fat, cognition or physical performance in healthy older adults, with only scattered gains in subgroups.

The numbers and sources

Effect: Over 48 weeks, weekly rapamycin 5 mg or 10 mg did not change visceral fat on DXA, the primary outcome, against placebo (p = 0.942). Blood biomarkers stayed in normal ranges. Lean tissue mass and self-reported pain improved in women on 10 mg, and emotional well-being and general health scores improved on 5 mg. No other effects were significant [1]. In a pilot in adults aged 70 to 95, 8 weeks or more of rapamycin 1 mg daily did not change cognitive, physical performance or self-rated health measures, or glucose tolerance [2].
Population: Healthy adults in a decentralized 48-week trial [1]; 25 adults with mean age 80.5 [2].
Sources: [1], [2]
Caveats: The positive findings in [1] were in subgroups among many secondary outcomes. [1] used compounded rapamycin. Both trials were small.

Methylene blueUntested

Longevity

Not studied in people. In a multi-site mouse lifespan study it did not change median lifespan, with a 6% rise in maximum lifespan in females only.

No study in people. This rests on animal studies, lab studies.

The numbers and sources

Effect: In genetically mixed mice tested at three sites, methylene blue did not alter median lifespan in males or females, and raised female maximum lifespan by 6% (P = 0.004) [9]. In culture, it extended the lifespan of human fibroblasts by more than 20 population doublings [8].
Population: Genetically heterogeneous mice at three sites [9]. Human IMR90 fibroblasts in culture [8].
Sources: [9], [8]
Caveats: Mouse and cell results do not show what happens to human lifespan. No study of methylene blue and aging outcomes in people was located.

RapamycinUntested

Longer human lifespan

Not studied in people. Started late in life, rapamycin lengthened median and maximum lifespan in both male and female mice at three test sites.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In genetically mixed mice fed rapamycin from 600 days of age, median and maximum lifespan rose in both sexes. Age at 90% mortality rose 14% in females and 9% in males, at three independent sites, and disease patterns did not differ from controls. Mice started at 270 days of age also survived longer at an interim analysis [9]. No study measuring lifespan in people taking rapamycin was located. In kidney transplant recipients, mortality was higher on sirolimus (see Risks) [6].
Population: Genetically heterogeneous mice at three sites [9].
Sources: [9], [6]
Caveats: Mouse lifespan does not show what happens in people. The transplant finding comes from people on combined immunosuppression, not healthy adults, and does not measure aging [6].

Hormone therapy

Menopausal hormone therapyModerateStudied in women

Long-term mortality

Not shown to change death from any cause over 18 years in postmenopausal women, 27.1% against 27.6% on placebo.

The numbers and sources

Effect: Over 18 years of cumulative follow-up, all-cause mortality was 27.1% with hormone therapy and 27.6% with placebo (HR 0.99, 95% CI 0.94 to 1.03). Cardiovascular mortality HR was 1.00 (95% CI 0.92 to 1.08) and cancer mortality 1.03 (95% CI 0.95 to 1.12). During the treatment phase, the HR for all-cause mortality in women aged 50 to 59 relative to women aged 70 to 79 was 0.61 (95% CI 0.43 to 0.87) [7].
Population: 27,347 postmenopausal women aged 50 to 79 in both WHI trials, treated a median of 5.6 or 7.2 years [7].
Sources: [7]
Caveats: Post-treatment follow-up was observational. The age comparison is a ratio of nominal hazard ratios.

Growth hormoneUntested

Longer lifespan

Not studied in people as a lifespan outcome. In mice, less growth hormone goes with longer life, the opposite of the claim.

No study in people. This rests on animal studies.

The numbers and sources

Effect: Mice with the growth hormone gene knocked out lived longer than normal mice in both sexes, while weighing less, carrying more body fat as a share of weight and handling glucose less well [13]. Mice lacking growth hormone releasing hormone were long-lived, and calorie restriction lengthened their lives further [14]. In SAMP8 mice, a drug that blocks the growth hormone releasing hormone receptor raised mean life expectancy by 8 weeks, with no rise in maximum lifespan [15].
Population: Growth hormone knockout mice [13]. GHRH knockout mice [14]. SAMP8 mice, a strain with shortened life expectancy [15].
Sources: [13], [14], [15]
Caveats: These studies lowered growth hormone rather than adding it, and mice are not people. No trial has measured lifespan in adults given growth hormone. The mortality findings after childhood treatment are covered under Risks [10].

Peptides and SARMs

MOTS-cGrey marketUntested

Aging and longevity

Not studied in people. MOTS-c treatment started late in life raised physical capacity and healthspan in old mice.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In mice, MOTS-c treatment begun at 23.5 months of age increased physical capacity and healthspan [12]. A 2022 review reports that MOTS-c levels in plasma decline with age, and reviews reported effects on diabetes, cardiovascular disease, osteoporosis, postmenopausal obesity and Alzheimer's disease [14].
Population: Old mice [12]. Studies reviewed in [14].
Sources: [12], [14]
Caveats: The study abstract reports physical capacity and healthspan, not lifespan [12]. Lower MOTS-c levels with age in people do not show that taking it changes how people age.

CJC-1295 and ipamorelinGrey marketNo research yet

Anti-aging

We haven't sourced research on this yet.

EpitalonGrey marketNo research yet

Aging, longevity and telomeres

We haven't sourced research on this yet.

Supplements

PotassiumModerate

Stroke, heart events and death

Shown to lower stroke about 14% and death about 12% with a potassium salt substitute, mostly from one trial in older rural Chinese adults at high stroke risk.

The numbers and sources

Effect: Stroke was lower on a salt substitute of 75% sodium chloride and 25% potassium chloride than on regular salt (29.14 against 33.65 per 1,000 person-years, rate ratio 0.86, 95% CI 0.77 to 0.96). Major cardiovascular events (rate ratio 0.87) and death (rate ratio 0.88, 95% CI 0.82 to 0.95) were also lower [6]. Pooled trials found lower total mortality (RR 0.89, 95% CI 0.85 to 0.94) and cardiovascular events (RR 0.89) [2]. In cohort studies, higher potassium intake went with a 24% lower stroke risk (RR 0.76, 95% CI 0.66 to 0.89) [1]. In a post hoc analysis of the same trial, total cancer did not differ (RR 0.94, 95% CI 0.81 to 1.08), and stomach cancer showed a non-significant trend lower (RR 0.77, 95% CI 0.54 to 1.08) [7].
Population: 20,995 people in 600 rural Chinese villages with prior stroke, or aged 60 or older with high blood pressure, mean follow-up 4.74 years [6][7]; 5 trials reporting clinical outcomes [2]; 11 cohorts with 127,038 people [1].
Sources: [1], [2], [6], [7]
Caveats: One trial carries most of the clinical events, and it enrolled people at high stroke risk. Cochrane rated the effects as slight and the certainty moderate [3]. The trial was funded by Australia's National Health and Medical Research Council [6].

Vitamin AModerateStudied in children

Child deaths and illness

Shown to lower child deaths about 12% in children aged 6 months to 5 years, mostly in Asia and Africa.

The numbers and sources

Effect: Across 19 trials, death from any cause was 12% lower with vitamin A than with control (RR 0.88, 95% CI 0.83 to 0.93). Deaths from diarrhea were also 12% lower (RR 0.88, 95% CI 0.79 to 0.98). Deaths from measles did not significantly differ (RR 0.88, 95% CI 0.69 to 1.11) [1].
Population: 47 trials with about 1,223,856 children aged 6 months to 5 years in 19 countries, mostly in Asia and Africa. 19 trials with 1,202,382 children reported mortality [1].
Sources: [1]
Caveats: The authors rated the mortality evidence high certainty. Children in hospital or with disease were left out. Most trials ran about a year [1].

Fisetin and quercetin (senolytics)Limited

Senescent cell burden

Small trials found fewer senescent cell markers after dasatinib, a prescription cancer drug, plus quercetin. A review found clinical benefit not established.

The numbers and sources

Effect: Eleven days after 3 days of dasatinib 100 mg and quercetin 1,000 mg, fat tissue had fewer p16- and p21-expressing cells and fewer cells with senescence-associated β-galactosidase. Skin p16 and p21 cells and blood SASP factors, including IL-1α and IL-6, also fell [1]. In fibrotic MASH, liver single-nucleus sequencing showed decreased senescence and fibrosis gene signatures [2]. A 2026 systematic review of 6 completed human studies found reductions in senescence biomarkers with dasatinib plus quercetin [3]. In 97 coronary bypass patients, quercetin alone at 500 mg twice a day from 2 days before surgery to discharge reversed senescence and inflammaging gene pathways in artery cells from men, improved endothelial relaxation in men but not women (p 0.049 overall), and showed a trend toward lower C-reactive protein (p 0.073). Post-operative atrial fibrillation occurred in 4% on quercetin and 18% on placebo (p 0.033) [4].
Population: 9 adults with diabetic kidney disease, mean age 68.7 [1]; 31 adults with fibrotic MASH [2]; 6 completed human studies [3]; 97 patients (78 men) with coronary artery disease [4].
Sources: [1], [2], [3], [4]
Caveats: Trial [1] was open-label with no control group. These studies used dasatinib, a prescription cancer drug, not quercetin or fisetin alone. The review calls the evidence insufficient to establish clinical efficacy [3]. In the bypass trial, the gene findings were exploratory and the endothelial test was done on artery samples outside the body [4].

NiacinLimitedStudied in men

Long-term mortality after a heart attack, before statins

One trial from before statins found 11% lower mortality at 15 years in men after a heart attack, appearing after treatment had stopped.

The numbers and sources

Effect: At a mean 15 years, about 9 years after the trial ended, mortality was 52.0% in the niacin group and 58.2% on placebo, 11% lower (P = 0.0004). During the trial, niacin lowered nonfatal repeat heart attacks and did not lower total mortality [4].
Population: 8,341 men aged 30 to 64 with a prior heart attack, treated 1966 to 1975 [4].
Sources: [4]
Caveats: One trial from before statins. The mortality difference appeared after treatment stopped [4].
Why limited: This rests on one trial, of 8,341 people. Strength counts independent studies that agree, so one trial is limited at any size.

Nicotinamide ribosideLimited

Biological age markers

One small trial found lower epigenetic age on two clocks in an exploratory analysis in older adults with mild cognitive impairment.

The numbers and sources

Effect: In an exploratory analysis, NR was linked to a modest rise in global DNA methylation and a lower epigenetic age on the PhenoAge and GrimAge clocks [3].
Population: 20 older adults with mild cognitive impairment over 10 weeks [3].
Sources: [3]
Caveats: An exploratory outcome in a 20-person pilot. Cognition did not change in the same trial [3].

NMNLimited

Biological age markers

One manufacturer-funded trial found blood-test biological age held steady on NMN over 60 days while it rose on placebo.

The numbers and sources

Effect: Blood biological age, estimated with the Aging.Ai 3.0 calculator, rose in the placebo group over 60 days and stayed unchanged in all NMN groups, a significant difference between groups (all p < 0.05) [1].
Population: 80 healthy middle-aged adults [1].
Sources: [1]
Caveats: One manufacturer-funded trial. The calculator is a blood-test model, not an epigenetic clock, and 60 days cannot show effects on aging itself.

Vitamin B6Moderate

Heart attack, stroke and death

Mixed results: stroke was about 10% lower in pooled trials of B6, mostly with folic acid and B12, while heart attack and death did not change.

The numbers and sources

Effect: Across 15 trials of vitamins B6, B9 or B12, heart attack (RR 1.02, 95% CI 0.95 to 1.10) and death from any cause (RR 1.01, 95% CI 0.96 to 1.06) did not change. Stroke was lower (RR 0.90, 95% CI 0.82 to 0.99) [8]. In HOPE-2, 50 mg of B6 with folic acid and B12 lowered homocysteine by 2.4 µmol/L. Major cardiovascular events did not change (RR 0.95, 95% CI 0.84 to 1.07) and stroke fell (RR 0.75, 95% CI 0.59 to 0.97) [9]. In NORVIT, 40 mg of B6 alone did not change recurrent events (RR 1.14, 95% CI 0.98 to 1.32) [10].
Population: 71,422 participants in 15 trials [8]; 5,522 people aged 55 or older with vascular disease or diabetes, 5 years [9]; 3,749 people within 7 days of a heart attack, median 40 months [10].
Sources: [8], [9], [10]
Caveats: Most trials combined B6 with folic acid and B12, so B6 alone cannot be separated. The Cochrane authors rated the heart attack and mortality evidence high quality [8].

GlycineLimited

Glutathione and markers of aging (glycine with N-acetylcysteine)

Small trials found mixed results: one found GlyNAC improved glutathione and strength in older adults, and a larger trial found no rise in glutathione.

The numbers and sources

Effect: In 24 older adults, 16 weeks of glycine plus N-acetylcysteine (GlyNAC) against an alanine placebo improved glutathione deficiency, oxidative stress, mitochondrial function, insulin resistance, gait speed, strength and other measures [6]. In 114 healthy older adults given 2.4, 4.8 or 7.2 g a day for 2 weeks, GlyNAC did not raise the glutathione ratio (12.65 vs 12.49 with placebo, p = 0.739) or total glutathione (p = 0.278), the primary endpoint. A post hoc subgroup with high oxidative stress and low baseline glutathione showed increased glutathione [7].
Population: 24 older and 12 young adults [6]. 114 healthy older adults [7].
Sources: [6], [7]
Caveats: These test a combination, not glycine alone. The two trials disagree. Six authors of the larger trial were Nestlé employees, and Nestlé Health Science holds patents and markets GlyNAC products [7].

MultivitaminLimited

Biological aging markers

One trial found slower change in two of five DNA methylation clocks over 2 years. The authors state the clinical relevance is not known.

The numbers and sources

Effect: Over 2 years, a daily multivitamin slowed yearly change in two DNA methylation clocks against placebo, by 0.113 years for PCGrimAge (95% CI -0.205 to -0.020) and 0.214 years for PCPhenoAge (95% CI -0.410 to -0.019). The effect on PCGrimAge was larger in people with faster biological aging at baseline [12].
Population: 958 COSMOS participants (482 women, 476 men) [12].
Sources: [12]
Caveats: Five clocks were tested and two changed. The authors state that the clinical relevance of these changes is not known [12].

Urolithin ALimited

Mitochondrial and cellular markers

Small trials found changes in mitochondrial and inflammation blood markers in middle-aged and older adults. All were run or funded by the maker.

The numbers and sources

Effect: Four weeks at 500 or 1,000 mg a day changed plasma acylcarnitines and muscle mitochondrial gene expression in older adults [4]. Acylcarnitines, ceramides and CRP fell more than on placebo in older adults over 4 months [3], and acylcarnitines and CRP fell in middle-aged adults [2].
Population: Healthy, sedentary older adults [4]; 66 older adults [3]; middle-aged adults [2].
Sources: [4], [3], [2]
Caveats: These are markers, not health outcomes. All three trials were run or funded by the maker.

Fish oilStrong

Death and cardiovascular events

Not shown to change deaths, cardiovascular events or stroke in adults. Coronary events and heart attacks fell slightly, by about 8% to 9%.

The numbers and sources

Effect: Long-chain omega-3 made little or no difference to all-cause mortality (RR 0.97, 95% CI 0.93 to 1.01, 45 RCTs), cardiovascular events (RR 0.96, 95% CI 0.92 to 1.01, 43 RCTs) or stroke (RR 1.02, 95% CI 0.94 to 1.12). Coronary heart disease events fell slightly (RR 0.91, 95% CI 0.85 to 0.97, number needed to treat 167, low certainty) [1]. A meta-analysis of 13 large trials (127,477 people, mean 5.0 years) found lower myocardial infarction (RR 0.92, 95% CI 0.86 to 0.99), coronary death (RR 0.92, 95% CI 0.86 to 0.98) and total cardiovascular disease (RR 0.97, 95% CI 0.94 to 0.99) with REDUCE-IT excluded, and a linear dose-response for total cardiovascular disease [6].
Population: 143,693 adults for mortality [1]; 127,477 adults in 13 cardiovascular outcome trials [6].
Sources: [1], [6]
Caveats: The two reviews agree on small effect sizes and differ in emphasis. Adding REDUCE-IT, a trial of high-dose purified EPA, strengthened the associations but introduced significant heterogeneity [6]. Effects in [1] did not differ by dose or trial duration.

B-complex vitaminsModerate

Heart attack, stroke and death

Not shown to change heart attack or death in 15 trials of B6, folic acid or B12. Stroke was about 10% lower.

The numbers and sources

Effect: Across 15 trials of vitamins B6, B9 or B12, heart attack (RR 1.02, 95% CI 0.95 to 1.10) and death from any cause (RR 1.01, 95% CI 0.96 to 1.06) did not change. Stroke was lower (RR 0.90, 95% CI 0.82 to 0.99) [1]. In HOPE-2, folic acid, B6 and B12 lowered homocysteine and did not change major cardiovascular events (RR 0.95, 95% CI 0.84 to 1.07) [2].
Population: 71,422 participants in 15 trials [1]; 5,522 people aged 55 or older with vascular disease or diabetes, 5 years [2].
Sources: [1], [2]
Caveats: The Cochrane authors rated heart attack and mortality evidence high quality [1]. Trials tested folic acid, B6 and B12, not full B-complex products.

MultivitaminModerate

Death from any cause

Not shown to lower death from any cause in two large trials of older adults. In three US cohorts, daily users had about 4% higher mortality early in follow-up.

The numbers and sources

Effect: All-cause mortality did not change in the Physicians' Health Study II (HR 0.94, 95% CI 0.88 to 1.02) [1] or COSMOS (HR 0.93, 95% CI 0.81 to 1.08) [2]. In 390,124 adults from 3 US cohorts followed up to 27 years, daily multivitamin users had slightly higher mortality than nonusers in the first half of follow-up (HR 1.04, 95% CI 1.02 to 1.07), and the second half gave the same estimate with an interval that includes no difference (HR 1.04, 95% CI 0.99 to 1.08) [6].
Population: Trial participants [1][2]; adults without chronic disease at baseline in NIH-AARP, PLCO and the Agricultural Health Study [6].
Sources: [1], [2], [6]
Caveats: The cohort data are observational, and a 4% difference is within the reach of confounding. Daily users were more often female and college educated and less often current smokers [6]. The cohort's authors state their conclusion as no mortality benefit [6].

NiacinModerate

Heart attack, stroke and death

Not shown to lower heart attacks, strokes or deaths in people with vascular disease, including those on statins.

The numbers and sources

Effect: In AIM-HIGH, the primary composite occurred in 16.4% on niacin and 16.2% on placebo (HR 1.02, 95% CI 0.87 to 1.21). The trial stopped at a mean 3 years for lack of efficacy [1]. In HPS2-THRIVE, major vascular events occurred in 13.2% and 13.7% (rate ratio 0.96, 95% CI 0.90 to 1.03) [2]. Across 23 trials, niacin did not change overall mortality (RR 1.05, 95% CI 0.97 to 1.12), heart attacks (RR 0.93, 95% CI 0.87 to 1.00) or strokes (RR 0.95, 95% CI 0.74 to 1.22) [3].
Population: 3,414 people [1]; 25,673 adults [2]; 39,195 participants in 23 trials, median dose 2 g a day [3].
Sources: [1], [2], [3]
Caveats: AIM-HIGH was funded by the US National Heart, Lung, and Blood Institute and Abbott Laboratories [1]. HPS2-THRIVE was funded by Merck and others, and added laropiprant to reduce flushing [2]. The Cochrane authors rated the mortality evidence high quality [3].

SeleniumModerate

Heart attack, stroke and mortality

Not shown to lower death, heart attack, stroke or other cardiovascular events in people, with events mostly from two US trials.

The numbers and sources

Effect: Selenium did not change death from any cause (RR 0.97, 95% CI 0.88 to 1.08), cardiovascular death (RR 0.97, 95% CI 0.79 to 1.2) or all cardiovascular events (RR 1.03, 95% CI 0.95 to 1.11) [3]. A Cochrane review of antioxidants found no effect of selenium on mortality (RR 0.97, 95% CI 0.91 to 1.03, 17 trials) [4].
Population: 12 trials with 19,715 people, mostly the US SELECT and NPC trials [3]; 17 low-bias selenium trials [4].
Sources: [3], [4]
Caveats: Clinical events came mostly from two US trials [3].

Vitamin DModerate

Cancer, cardiovascular disease and death

Not shown to lower cancer, cardiovascular disease or death in adults. Reviews disagree on cancer death, which one found about 16% lower.

The numbers and sources

Effect: In VITAL, invasive cancer did not differ (HR 0.96, 95% CI 0.88 to 1.06), nor did major cardiovascular events (HR 0.97, 95% CI 0.85 to 1.12) or death from any cause (HR 0.99, 95% CI 0.87 to 1.12). Cancer death had HR 0.83 (95% CI 0.67 to 1.02) [18]. Across 52 trials (75,454 participants), all-cause mortality did not change (RR 0.98, 95% CI 0.95 to 1.02) and cancer death fell (RR 0.84, 95% CI 0.74 to 0.95) [19]. Across 9 RCTs (114,379 participants), cardiovascular events (RR 0.95, 95% CI 0.88 to 1.04), cardiovascular death (RR 1.04, 95% CI 0.87 to 1.24) and heart attack (RR 0.96, 95% CI 0.83 to 1.11) did not differ from placebo [20]. Across 5 RCTs (56,901 participants), melanoma incidence (RR 0.94, 95% CI 0.71 to 1.24) and recurrence (HR 1.23, 95% CI 0.29 to 5.18) did not differ [21]. In 455 people starting chemotherapy for metastatic colorectal cancer, 4,000 IU a day after a loading dose did not change progression-free survival against 400 IU a day (median 11.8 against 10.3 months, P 0.25) or overall survival (25.6 against 27.0 months) [22].
Population: 25,871 US adults over a median 5.3 years [18]; adults in 52 RCTs [19]; adults in 9 placebo-controlled RCTs [20]; adults in 5 RCTs, with or without calcium [21]; US adults with untreated metastatic colorectal cancer, median follow-up 20 months [22].
Sources: [18], [19], [20], [21], [22]
Caveats: The sources disagree on cancer death. VITAL's interval crossed 1.0 [18] and the pooled analysis did not [19]. Melanoma was a secondary endpoint in some trials and events were few [21].

ResveratrolLimited

Longer life

Limited evidence: in older Italian adults, dietary resveratrol levels were not linked to death over 9 years.

The numbers and sources

Effect: Over 9 years, 268 of 783 older adults died. Deaths by quartile of urinary resveratrol metabolites, lowest to highest, were 34.4%, 31.6%, 33.5% and 37.4%. The lowest quartile had a hazard ratio of 0.80 (95% CI 0.54 to 1.17) against the highest. Resveratrol levels were not associated with inflammatory markers, heart disease or cancer [11].
Population: 783 community-dwelling adults aged 65 and older in the Chianti region of Italy [11].
Sources: [11]
Caveats: An observational cohort of resveratrol from diet, not supplements. Doses from diet are far smaller than supplement doses.

Alpha-ketoglutarateUntested

Longer life

Not studied in people. AKG lengthened life in worms and in one mouse study, but a multi-site mouse program found no longer lifespan.

No study in people. This rests on animal studies.

The numbers and sources

Effect: AKG extended the lifespan of adult C. elegans worms, an effect that required ATP synthase and TOR signaling [6]. In C57BL/6 mice, calcium AKG in the diet gave a longer life with less frailty and lower levels of inflammatory cytokines [7]. In the Interventions Testing Program, AKG given to genetically mixed UM-HET3 mice at three sites did not significantly increase lifespan in males or females [8].
Population: C. elegans worms [6]. C57BL/6 mice [7]. UM-HET3 mice at three US sites [8].
Sources: [6], [7], [8]
Caveats: Lifespan has not been measured in people. The mouse results disagree, and the multi-site program tested a different dose or starting age from earlier work [8]. The epigenetic age studies above do not measure lifespan.

AstaxanthinUntested

Longer life

Not studied in people. One mouse study found 12% longer median life in males, and a later study at other doses found no longer lifespan.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In genetically mixed UM-HET3 mice fed astaxanthin from 12 months of age, median male lifespan rose 12% (p = 0.003). Lifespan in females was not significantly affected [11]. In a follow-up program at three sites, astaxanthin given at different doses or later starting ages did not increase lifespan in either sex, and an apparent shorter female lifespan disappeared once one site's unusual controls were set aside [12]. Astaxanthin also lengthened the life of C. elegans worms, an effect that depended on the SKN-1 and PHA-4 genes [13].
Population: UM-HET3 mice at three US sites [11][12]. C. elegans worms [13].
Sources: [11], [12], [13]
Caveats: Lifespan has not been measured in people. The mouse results differ by sex, dose and starting age [11][12].

Fisetin and quercetin (senolytics)Untested

Longer life

Not studied in people. Fisetin given late in life lengthened mouse lifespan in one study but not in a multi-site program, and dasatinib plus quercetin extended survival in old mice.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In wild-type mice given fisetin late in life, median and maximum lifespan rose, with less age-related tissue damage [14]. In the Interventions Testing Program, fisetin did not significantly change lifespan in either sex of genetically mixed mice at the dose and schedule used [15]. In old mice, intermittent dasatinib plus quercetin raised survival after treatment by 36% and cut the mortality hazard to 65% [16].
Population: Progeroid and aged wild-type mice [14]. UM-HET3 mice at three US sites [15]. Old mice and mice given transplanted senescent cells [16].
Sources: [14], [15], [16]
Caveats: Lifespan has not been measured in people. The mouse results for fisetin disagree [14][15]. Quercetin was tested with dasatinib, a cancer drug, not alone [16].

GlyNACUntested

Longer life

Not studied in people. In one study from the group behind the human trials, mice given GlyNAC lived 24% longer than control mice.

No study in people. This rests on animal studies.

The numbers and sources

Effect: C57BL/6J mice given GlyNAC lived 24% longer than mice given placebo, with improved glutathione, oxidative stress, mitochondrial and DNA damage measures in heart, liver and kidney [7].
Population: C57BL/6J mice [7].
Sources: [7]
Caveats: One mouse study by the same research group that ran most of the human trials [7][3]. Lifespan has not been measured in people, and the study has not been repeated by another lab.

NAD+ infusionsUntested

Longer life

Not studied in people. No lifespan study of infused NAD was located, and mouse studies of oral NR, an NAD precursor, disagree.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In one lab, NR raised NAD, rejuvenated muscle stem cells in aged mice and increased mouse lifespan [9]. In genetically mixed mice tested at three sites by the US National Institute on Aging program, NR did not increase lifespan in either sex [10].
Population: Aged mice [9]. Genetically heterogeneous mice at three sites [10].
Sources: [9], [10]
Caveats: Both studies gave NR by mouth, not NAD by infusion. Mouse lifespan results do not show what happens in people.

Nicotinamide ribosideUntested

Longer life

Not studied in people. Mouse studies disagree: NR lengthened lifespan in one lab, while a multi-site US aging institute program found no lifespan gain in either sex.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In one lab, NR raised NAD, rejuvenated muscle stem cells in aged mice, delayed senescence of neural and melanocyte stem cells and increased mouse lifespan [11]. In genetically mixed mice tested at three sites by the US National Institute on Aging program, NR did not increase lifespan in males or females at the doses tested [12].
Population: Aged mice [11]. Genetically heterogeneous mice at three sites [12].
Sources: [11], [12]
Caveats: Mouse lifespan results do not show what happens in people. The larger multi-site test found no effect [12].

NMNUntested

Longer life

Not studied in people. NMN lengthened life in a strain of prematurely aging mice and eased age-related decline over 12 months in normal mice.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In Zmpste24-deficient mice, a prematurely aging strain, NMN prolonged lifespan and delayed signs of aging [12]. In normal mice given NMN by mouth for 12 months, NMN suppressed age-related weight gain, raised energy metabolism and physical activity, and improved insulin sensitivity, blood lipids and eye function, with no obvious toxicity [13].
Population: Zmpste24-deficient mice [12]. Wild-type C57BL/6N mice during normal aging [13].
Sources: [12], [13]
Caveats: The lifespan gain was in a premature aging strain, and the 12-month study in normal mice did not report lifespan in its abstract [13]. Mouse results do not show what happens to human lifespan.

PQQUntested

Longer life

Not studied in people. In a short-lived, fast-aging mouse strain, lifelong PQQ in the diet extended 75th percentile survival by 73%.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In senescence-accelerated SAMP8 mice, a strain with a short lifespan, lifelong PQQ at 0.02% of the diet prolonged survival at the 75th percentile by 73% against controls and delayed signs of aging and age-related muscle decline. Starting PQQ in midlife improved muscle function that had declined with age [7].
Population: Senescence-accelerated SAMP8 mice [7].
Sources: [7]
Caveats: One study in a fast-aging strain, not normally aging mice. One author also wrote the maker-run memory trial [1]. Mouse survival does not show what happens to human lifespan.

SulforaphaneUntested

Longer life

Not studied in people. In roundworms, sulforaphane lengthened lifespan and improved mobility through the FOXO aging pathway.

No study in people. This rests on animal studies, how it acts in the body.

The numbers and sources

Effect: In wild-type C. elegans worms, sulforaphane prolonged lifespan and increased mobility and food intake. The effect depended on the TRA-1 and DAF-16 genes, the worm versions of human GLI and FOXO, and disappeared when either was blocked [12].
Population: C. elegans roundworms, wild-type and 4 mutant strains [12].
Sources: [12]
Caveats: A worm study. No mammal lifespan study or human study of lifespan was located.

TaurineUntested

Longevity and aging

Not studied in people. Taurine supplements lengthened lifespan in mice and worms, but in 137 men blood taurine did not fall with age.

No study in people. This rests on animal studies, how it acts in the body.

The numbers and sources

Effect: Blood taurine declined with age in mice, monkeys and humans. Taurine supplements increased health span and lifespan in mice and worms and health span in monkeys, and reduced cellular senescence, DNA damage and inflammation markers [10]. A later study of 137 men aged 20 to 93 found no link between circulating taurine and age, muscle mass, strength, physical performance or mitochondrial function [11].
Population: Worms, mice and monkeys, with human blood level data [10]. 137 inactive and active men aged 20 to 93 [11].
Sources: [10], [11]
Caveats: The lifespan results are from animals. The human data are blood levels, not supplement trials, and the two studies disagree on whether levels fall with age [10][11]. No trial of taurine and aging outcomes in people was located.

TMGUntested

Longer life

Not studied in people. Betaine supplements extended lifespan in roundworms, and a review describes effects on methylation and inflammation linked to aging.

No study in people. This rests on animal studies, how it acts in the body.

The numbers and sources

Effect: In C. elegans worms, betaine supplementation extended lifespan through FoxO, MAPK, autophagy and mTOR pathways, and three anti-aging drugs raised betaine levels [8]. A 2025 review reports that betaine may counter aging-related changes in methylation by raising S-adenosylmethionine, lowers homocysteine, and reduced oxidative stress and inflammation in studies, and found no clear link between dietary betaine and heart disease or type 2 diabetes [9].
Population: C. elegans roundworms [8]. Human and animal studies, reviewed [9].
Sources: [8], [9]
Caveats: A worm lifespan study and a narrative review. The review states that efficacy and safety of betaine in older people need further research [9]. No study of betaine and human lifespan was located.

Urolithin AUntested

Longer life

Not studied in people. Urolithin A extended lifespan in roundworms and improved exercise capacity in aging mice and young rats.

No study in people. This rests on animal studies, how it acts in the body.

The numbers and sources

Effect: In C. elegans worms, urolithin A induced mitophagy, prevented the age-related buildup of dysfunctional mitochondria and extended lifespan, with longer normal mobility. In two mouse models of age-related muscle decline and in young rats, it improved exercise capacity [9].
Population: C. elegans roundworms, mice and rats [9].
Sources: [9]
Caveats: The lifespan result is from worms. Rodent results covered exercise capacity, not lifespan. No study of urolithin A and human lifespan was located.

Practices

Infrared saunaLimited

Cardiac events over 5 years in heart failure

One non-randomized study found cardiac events in 31.3% of heart failure patients who continued Waon therapy for 5 years, against 68.7% of matched controls.

The numbers and sources

Effect: At 60 months, cardiac events, defined as events due to heart failure or cardiac death, had occurred in 31.3% of the Waon group and 68.7% of the control group (P < 0.01). Eight patients died in the Waon group and 12 in the control group [7].
Population: 129 patients in NYHA class III or IV, 64 treated daily for 5 days in hospital and at least twice a week after discharge, and 65 controls matched for age, sex and class [7].
Sources: [7]
Caveats: Not randomized, single center, and the group that kept attending sessions may differ from those who did not.

MeditationLimited

Telomerase and cell aging

Small trials found telomerase activity in blood immune cells rose with mindfulness meditation, pooled across 4 trials and 190 people.

Compared with control groups in 4 randomized trials the source abstract does not describe [20]

The numbers and sources

Effect: A meta-analysis of 4 randomized trials found increased telomerase activity in peripheral blood mononuclear cells (d = 0.46) [20].
Population: 190 participants in 4 trials [20].
Sources: [20]
Caveats: Telomerase activity is an enzyme measure, not telomere length, lifespan or a health outcome. Four trials is a small base.

SaunaLimited

Death from heart disease

Linked to fewer cardiovascular deaths in Finnish cohorts, with a hazard ratio of 0.37 for sudden cardiac death at 4 to 7 sessions a week against one.

The numbers and sources

Effect: In 2,315 Finnish men followed for a median 20.7 years, fatal cardiovascular disease occurred in 22.3% of men using a sauna once a week, 16.4% at 2 to 3 times and 12.0% at 4 to 7 times. The adjusted hazard ratio for sudden cardiac death was 0.78 (95% CI 0.57 to 1.07) at 2 to 3 sessions a week and 0.37 (95% CI 0.18 to 0.75) at 4 to 7 sessions. Sessions longer than 19 minutes carried a hazard ratio of 0.48 (95% CI 0.31 to 0.75) against sessions under 11 minutes [9]. In 1,688 men and women followed for 15.0 years, adjusted hazard ratios for cardiovascular death were 0.75 (95% CI 0.52 to 1.08) and 0.23 (95% CI 0.08 to 0.65) [10]. In 2,277 men over 26.1 years, 3 to 7 sessions a week against 2 or fewer carried a hazard ratio of 0.74 (95% CI 0.59 to 0.94) for cardiovascular death [11].
Population: Men aged 42 to 60 from eastern Finland in the Kuopio Ischemic Heart Disease study [9][11]. 1,688 participants with a mean age of 63, 51.4% women [10].
Sources: [9], [10], [11]
Caveats: Observational. All three analyses come from one Finnish cohort program and one research group, and sauna habits were self-reported at baseline. Only 201 men were in the 4 to 7 sessions group, with 10 sudden cardiac deaths [9]. The link may reflect other differences between frequent and infrequent sauna users.

SaunaLimitedStudied in men

Death from any cause

Linked to lower all-cause mortality in Finnish men, with a hazard ratio of about 0.85 for 3 to 7 sessions a week against 2 or fewer.

The numbers and sources

Effect: Over 20.7 years, 49.1% of men using a sauna once a week died, against 37.8% at 2 to 3 times and 30.8% at 4 to 7 times, with a significant trend after adjustment (P for trend 0.005 or less). Session length was not associated with all-cause mortality [9]. Comparing 3 to 7 sessions a week with 2 or fewer, adjusted hazard ratios for death from any cause were 0.84 (95% CI 0.72 to 0.97) in 2,277 men [11] and 0.86 (95% CI 0.76 to 0.97) in 2,575 men followed for 27.8 years [12].
Population: Finnish men aged 42 to 61 at baseline in the Kuopio Ischemic Heart Disease study [9][11][12].
Sources: [9], [11], [12]
Caveats: Observational, and all three analyses use the same cohort of Finnish men. No randomized trial has measured mortality.

Hyperbaric oxygen therapyLimited

Telomere length and senescent cells

Limited evidence: blood cell telomeres lengthened by over 20% after 60 sessions in 35 older adults, in a study with no control group and no health outcome.

The numbers and sources

Effect: Telomere length in T helper, T cytotoxic, natural killer and B cells rose by over 20% after 60 daily sessions, with B cells up 37.63% ± 52.73 after treatment. Senescent T helper cells fell by 37.30% ± 33.04 and senescent T cytotoxic cells by 10.96% ± 12.59 [16].
Population: 35 healthy, independently living adults aged 64 and older [16].
Sources: [16]
Caveats: No control group. The standard deviations are as large as the changes. Telomere length in blood cells is a lab marker, and the study did not measure lifespan, disease or function. Blood was last sampled 1 to 2 weeks after the final session.

Daily life

Daily stepsStrong

Death from any cause

Linked to progressively fewer deaths with more daily steps in pooled cohorts, leveling off at about 6,000 to 8,000 steps in adults 60 and older.

Compared with people in the lowest quarter of steps, a median of 3,553 a day [1]; people in the reference quarter, a median of 3,867 steps a day [2]; 2,000 steps a day [3][4]; taking 4,000 steps a day [5]

The numbers and sources

Effect: Against the lowest quarter, the hazard of death was 0.60 at a median of 5,801 steps, 0.55 at 7,842 and 0.47 at 10,901 (95% CI 0.39 to 0.57). Risk fell until 6,000 to 8,000 steps a day in adults aged 60 and older, and until 8,000 to 10,000 in younger adults [1]. Each extra 1,000 steps was linked to 15% lower mortality (HR 0.85, 95% CI 0.81 to 0.91) [2]. Risk was significantly lower from 2,517 steps a day (HR 0.92), with the largest drop at about 8,763 steps (HR 0.40, 95% CI 0.38 to 0.43) [3]. Against 2,000 steps, 7,000 was linked to 47% lower mortality (HR 0.53, 95% CI 0.46 to 0.60), with an inflection at about 5,000 to 7,000 [4]. In US adults, 8,000 steps (HR 0.49) and 12,000 steps (HR 0.35) were linked to lower mortality than 4,000 [5].
Population: 47,471 adults in 15 cohorts with 3,013 deaths, median follow-up 7.1 years [1]. 226,889 participants in 17 cohorts [2]. 111,309 people in 12 studies [3]. 14 studies in the mortality analysis, from 57 studies of 35 cohorts [4]. 4,840 US adults aged 40 and older, followed a mean of 10.1 years [5].
Sources: [1], [2], [3], [4], [5]
Caveats: Observational, with steps measured by a device for about a week. Confounding by health and income is possible, and reverse causation is likely in part: illness and frailty lower step counts before death. Results depended on where the device was worn [3]. After adjusting for total steps, faster stepping was not linked to lower mortality in one cohort (HR 0.90, 95% CI 0.65 to 1.27) [5], while peak 30-minute cadence kept a link in another analysis (HR 0.67) [1].

Physical activityStrong

Death from any cause

Linked to about 30% fewer deaths at the guideline amount of 150 minutes a week in pooled cohorts, with the curve flattening at 3 to 5 times that amount.

Compared with people reporting no leisure-time physical activity [1]; people doing no non-occupational activity, the reference of its dose-response curves [2]

The numbers and sources

Effect: Against no leisure activity, mortality was 20% lower below the guideline minimum of 7.5 MET-hours a week (HR 0.80), 31% lower at 1 to 2 times the minimum (HR 0.69, 95% CI 0.67 to 0.70) and 37% lower at 2 to 3 times. The benefit leveled off at 3 to 5 times the minimum (HR 0.61). There was no sign of harm at 10 times or more (HR 0.69, 95% CI 0.59 to 0.78) [1]. At 8.75 mMET-hours a week, equal to 150 minutes of moderate activity, the RR was 0.69 (95% CI 0.65 to 0.73). Differences were largest between none and that amount, smaller up to double it, and small and uncertain beyond. If all insufficiently active people reached that amount, 15.7% of premature deaths would have been averted [2].
Population: 661,137 adults, median age 62, with 116,686 deaths in 6 US and European cohorts, median follow-up 14.2 years [1]. Over 30 million participants in 94 cohorts, with 811,616 deaths in the mortality analysis [2].
Sources: [1], [2]
Caveats: Observational, with self-reported activity. Illness lowers activity before death, which is reverse causation, and active people differ in diet, smoking and income. The averted-deaths figure assumes the link is causal [2].

Sitting timeStrong

Death from any cause

Linked to more deaths with more sitting in pooled cohorts, independent of exercise, with risk rising mainly above about 7 to 8 hours a day.

Compared with people with less sedentary time, with physical activity held equal [1]; each hour less of sitting, within intervals of 0 to 3, over 3 to 7 and over 7 hours a day [2]; each hour less of sitting a day [3]; the least sedentary quarter of people [4]

The numbers and sources

Effect: Sedentary time was linked to higher all-cause mortality after adjusting for physical activity (HR 1.240, 95% CI 1.090 to 1.410), more so at low activity [1]. Each extra hour of sitting carried an HR of 1.00 between 0 and 3 hours, 1.02 between 3 and 7 hours and 1.05 above 7 hours, an estimated 34% higher risk at 10 hours a day [2]. Each hour added 1% to risk up to 8 hours a day and 4% above 8 hours [3]. With sitting measured by device, the most sedentary quarter had an HR of 2.63 (95% CI 1.94 to 3.56) against the least sedentary [4].
Population: 13 studies in the mortality analysis, from 47 articles [1]. 595,086 adults with 29,162 deaths in 6 cohorts [2]. 1,331,468 participants in 34 studies [3]. 36,383 adults, mean age 62.6, in 8 cohorts with device-measured sitting [4].
Sources: [1], [2], [3], [4]
Caveats: Observational, and sitting was self-reported in all but one study in one review [1]. Confounding is likely, and reverse causation is possible because people who are ill sit more. The device-measured cohorts were mostly older women, and the size of their link is larger than in the self-report cohorts [4].

Sitting timeStrong

Whether exercise offsets sitting

Linked to no higher death rate from sitting in people doing about 60 to 75 minutes of moderate activity a day, in a harmonized analysis of over a million adults.

Compared with people sitting under 4 hours a day in the most active quarter [5]; people with less sedentary time, with physical activity held equal [1]; each hour less of sitting, with and without adjustment for activity [2]

The numbers and sources

Effect: People sitting over 8 hours a day but in the most active quarter (over 35.5 MET-hours a week) had no higher mortality (HR 1.04, 95% CI 0.99 to 1.10) than those sitting under 4 hours in the same quarter. People sitting under 4 hours in the least active quarter had higher mortality (HR 1.27, 95% CI 1.22 to 1.31). The highest risk was in the least active people sitting over 8 hours (HR 1.59, 95% CI 1.52 to 1.66) [5]. Sitting-linked risk was more pronounced at lower activity levels [1], and moderate to vigorous activity weakened the sitting link [2].
Population: 1,005,791 adults in 13 cohorts with 84,609 deaths, followed 2 to 18.1 years [5]. 47 articles [1]. 595,086 adults [2].
Sources: [5], [1], [2]
Caveats: Observational, with self-reported sitting and activity. The active quarter's level, about 60 to 75 minutes of moderate activity a day, is well above what most adults do [5].

Sitting timeStrong

TV time and death

Linked to more deaths with more TV time in pooled cohorts, rising above about 3 to 4 hours a day, a link exercise weakened but did not remove.

Compared with each 2 hours less of TV a day [6]; each hour less of TV a day [3]; people in the lowest of four TV-time groups and the most active quarter [5]

The numbers and sources

Effect: Each 2 hours of TV a day was linked to 13% higher mortality (RR 1.13, 95% CI 1.07 to 1.18), with risk rising above 3 hours a day [6]. After adjusting for activity, each hour of TV added 3% to risk up to 3.5 hours a day and 6% above that [3]. Watching 3 hours or more was linked to higher mortality at every activity level except the most active, where only 5 hours or more carried a higher risk (HR 1.16, 95% CI 1.05 to 1.28) [5].
Population: 26,509 people with 1,879 deaths in 3 cohorts [6]. 34 studies [3]. 465,450 people with 43,740 deaths in 6 cohorts [5].
Sources: [6], [3], [5]
Caveats: Observational. TV time often goes with snacking, lower income and poorer health, which may explain part of the link.

Sleep durationStrong

Death from any cause

Linked to the lowest death rates at about 7 hours a night in pooled cohorts, a J or U shape, with a steeper rise for long sleep than for short.

Compared with people sleeping 7 to 8 hours a night, the reference in most studies [1]; people sleeping 7 hours a day [2]; normal sleepers [3]; people with normal sleep duration [4]

The numbers and sources

Effect: A dose-response meta-analysis found a U-shaped curve with the lowest risk at about 7 hours a day. Below 7 hours, risk rose 6% per hour less (RR 1.06, 95% CI 1.04 to 1.07). Above 7 hours, it rose 13% per hour more (RR 1.13, 95% CI 1.11 to 1.15). The shortest and longest groups had RRs of 1.13 and 1.35 [1]. A spline analysis of 40 cohorts found a J shape: against 7 hours, RR was 1.06 at 5 hours, 1.04 at 6 hours, 1.03 at 8 hours, 1.13 at 9 hours, 1.25 at 10 hours and 1.38 at 11 hours [2]. Short sleepers had a 12% higher risk (RR 1.12, 95% CI 1.08 to 1.16), with a linear rise below 6 hours [3]. Long sleepers had a 39% higher risk (RR 1.39, 95% CI 1.31 to 1.47) [4].
Population: 3,582,016 participants in 67 articles, follow-up 2.3 to 34 years [1]. 2,200,425 participants with 271,507 deaths in 40 cohorts [2]. 5,172,710 participants in 153 studies [3]. 5,134,036 participants in 137 cohorts [4].
Sources: [1], [2], [3], [4]
Caveats: Observational. Sleep was self-reported by questionnaire in most cohorts. Confounding is likely, and reverse causation is a strong candidate for long sleep: illness, depression and low fitness lengthen sleep, and some illnesses shorten it. The authors of the long sleep review call for work on whether the link is causal [4]. Heterogeneity was high for long sleep (I² 76.2%) [1]. In one review, short sleep was linked to higher risk in women and not in men [2].

SmokingStrong

Death from any cause after quitting

Linked to fewer deaths after quitting at every age in large cohorts, with about 90% of the excess avoided by quitting before 40. A quit-program trial lowered deaths.

Compared with people who continued to smoke [1][2]; never-smokers, with ex-smokers' excess set against continuing smokers' excess [3]; people who continued to smoke, and never smokers [4]; never smokers, by years since quitting [5]; usual care, in a randomized trial of a quit program [6]

The numbers and sources

Effect: Adults who quit at ages 25 to 34, 35 to 44 or 45 to 54 gained about 10, 9 and 6 years of life against those who continued to smoke. Quitting before 40 lowered the death risk of continued smoking by about 90% [1]. In British doctors, stopping at 60, 50, 40 or 30 gained about 3, 6, 9 or 10 years [2]. Women who stopped at 25 to 34 or 35 to 44 had death rates 1.05 and 1.20 times those of never-smokers, 3% and 10% of the excess in continuing smokers [3]. Across four national cohorts, former smokers had a hazard ratio of 1.3 against never smokers. Quitting for under 3 years was linked to 90% to 95% less excess risk in people under 40, and quitting for 10 years or more to survival similar to never smokers [4]. In people 60 and older, former smokers' excess mortality (RM 1.34, 95% CI 1.28 to 1.40) fell with years since quitting, and gains were seen in people 80 and older [5]. In a randomized trial, a quit program raised 5-year quitting to 21.7% against 5.4% with usual care. Deaths were 8.83 against 10.38 per 1,000 person-years (HR for usual care 1.18, 95% CI 1.02 to 1.37) [6].
Population: 202,248 US adults with 15,715 deaths [1]. 34,439 male British doctors followed 50 years [2]. 1.2 million UK women followed about 12 years [3]. 1.48 million adults in the US, UK, Norway and Canada with 122,697 deaths over 15 years [4]. 17 cohorts of people 60 and older in 7 countries [5]. 5,887 middle-aged smokers with mild airway obstruction, followed 14.5 years [6].
Sources: [1], [2], [3], [4], [5], [6]
Caveats: The cohorts are observational, and people who quit may differ in other ways from those who do not. Some people quit because they are already ill, which hides part of the gain in the first years. The trial's effect comes from a program that only a minority of its group used to quit, and it enrolled only people with airway obstruction [6].

Strength trainingStrong

Death from any cause

Linked to 15% to 21% fewer deaths with muscle-strengthening activity in pooled cohorts, a J shape with the lowest risk at about 30 to 60 minutes a week.

Compared with people doing no muscle-strengthening activity [1]; no exercise [2]; no resistance training [3]

The numbers and sources

Effect: Muscle-strengthening activity was linked to 15% lower mortality (RR 0.85, 95% CI 0.79 to 0.93). The dose-response curve was J-shaped, lowest at 40 minutes a week (RR 0.83), with estimates below 1 up to about 140 minutes a week. Combined with aerobic activity, risk was lower still [1]. A second review found 21% lower mortality (HR 0.79, 95% CI 0.69 to 0.91), and 40% lower when combined with aerobic exercise (HR 0.60) [2]. A third found 15% lower mortality (RR 0.85, 95% CI 0.77 to 0.93), with the largest drop, 27%, at about 60 minutes a week, and smaller drops at higher volumes [3].
Population: 263,058 adults with 42,133 deaths in 7 cohorts, 236,331 in the dose-response analysis [1]. 370,256 participants in 10 cohorts and 1 trial, mean follow-up 8.85 years [2]. 10 studies of nonclinical adults, 4 in the dose-response analysis [3].
Sources: [1], [2], [3]
Caveats: Observational, with self-reported activity. Certainty was rated very low and heterogeneity was high (I² 83%) [1]. Few studies informed the dose-response curves, so the shape above 60 minutes a week is uncertain [1][3].

Time in natureStrong

Death and living near green space

Linked to fewer deaths in people living amid more greenery, in two meta-analyses, about 4% fewer per step of greenness within 500 meters of home.

Compared with people with lower greenspace exposure [2]; people with 0.1 lower NDVI greenness within 500 meters of home [12]

The numbers and sources

Effect: Greenspace exposure was linked to lower all-cause mortality (0.69, 95% CI 0.55 to 0.87) and cardiovascular mortality (0.84, 95% CI 0.76 to 0.93) [2]. Each 0.1 rise in satellite-measured greenness within 500 meters of home was linked to 4% lower mortality (HR 0.96, 95% CI 0.94 to 0.97, I² 95%) [12].
Population: 143 studies [2]. 8,324,652 people in 9 cohorts from 7 countries [12].
Sources: [2], [12]
Caveats: Observational, and about where people live, not time they spend in nature. Income, air pollution and activity differ between greener and less green areas. Heterogeneity was very high [12], and the broader review notes poor study quality [2].

Volunteering and purposeStrong

Death from any cause in older volunteers

Linked to about 22% to 24% fewer deaths in older adults who volunteered, in three meta-analyses of cohorts after adjusting for health.

Compared with older adults who did not volunteer [1]; non-volunteers [2]; non-volunteers in studies with a well-defined control group [3]; people who volunteered 0 hours a year [4]

The numbers and sources

Effect: After adjustment for health and other factors, volunteering was linked to 24% lower mortality (95% CI 16% to 31%), and to 47% lower without adjustment [1]. Volunteers had lower mortality (RR 0.78, 95% CI 0.66 to 0.90) [2]. In people 65 and older, volunteers had lower mortality (HR 0.76, 95% CI 0.72 to 0.80) with no heterogeneity between studies [3]. Volunteering 100 hours or more a year was linked to lower mortality over 4 years [4].
Population: Adults aged 55 and older in organizational volunteering studies [1]. 5 cohort studies [2]. Adults 65 and older in 10 mortality studies, average follow-up 5 years [3]. 12,998 US adults over 50 [4].
Sources: [1], [2], [3], [4]
Caveats: Observational. People in better health, with more money and more social ties, are more able to volunteer, which is confounding and reverse causation. Adjustment cut the link by about half [1].

Volunteering and purposeStrong

Purpose in life and death from any cause

Linked to about 17% to 24% fewer deaths in people with a stronger sense of purpose in life, in two meta-analyses of cohorts.

Compared with people with a lower sense of purpose in life [10][11]

The numbers and sources

Effect: Higher purpose in life was linked to lower mortality (adjusted RR 0.83, 95% CI 0.75 to 0.91) [10]. Across 25 samples, purpose was linked to lower mortality (HR 0.76, 95% CI 0.70 to 0.83), 0.85 after adjusting for behavior and clinical risk factors, and 0.91 after adjusting for depression [11].
Population: 136,265 participants in 10 prospective studies [10]. 488,765 participants with 48,928 deaths, followed up to 32 years [11].
Sources: [10], [11]
Caveats: Observational, with purpose measured by questionnaire. Adjustment for health habits and depression shrank the link, so part of it runs through them [11]. Illness can lower a sense of purpose, which is reverse causation.

Pet ownershipModerate

Cardiovascular death

Linked to about 19% to 31% fewer cardiovascular deaths in pet or dog owners in two meta-analyses of cohorts. One review found no link for dogs alone.

Compared with people without a dog [1]; people without pets [2]; people without a dog, in single- and multiple-person households [3]

The numbers and sources

Effect: Dog ownership was linked to 31% lower cardiovascular death (RR 0.69, 95% CI 0.67 to 0.71) [1]. Pet owners had lower cardiovascular death after adjustment (HR 0.81, 95% CI 0.68 to 0.97), and cat owners too (HR 0.79, 95% CI 0.63 to 0.99), while dog ownership showed no significant link [2]. In a Swedish registry cohort, dog ownership was linked to lower cardiovascular death in single-person households (HR 0.64, 95% CI 0.59 to 0.70) and multiple-person households (HR 0.85, 95% CI 0.81 to 0.90) [3].
Population: 3,837,005 participants in 10 studies, mean follow-up 10.1 years [1]. 26 studies [2]. 3,432,153 Swedish adults followed up to 12 years [3].
Sources: [1], [2], [3]
Caveats: Observational. People in better health are more able to keep and walk a dog, which is reverse causation. The first review did not adjust for confounders [1]. In the Swedish twin subsample, the lower risk was not replicated [3].

SunlightLimitedStudied in women

Death from any cause

One cohort of Swedish women linked active sun habits to fewer deaths from heart disease and other non-cancer causes, and 0.6 to 2.1 years longer life.

Compared with women who avoided sun exposure [13]

The numbers and sources

Effect: Women with active sun exposure habits had lower risk of cardiovascular and non-cancer, non-cardiovascular death than women who avoided the sun. Life expectancy of avoiders was 0.6 to 2.1 years shorter than in the highest sun exposure group. Because they lived longer, cancer made up a larger share of deaths among sun-seeking women [13].
Population: 29,518 Swedish women aged 25 to 64 at the start, followed 20 years [13].
Sources: [13]
Caveats: One observational cohort. Women who seek the sun may be healthier, more active and better off, and ill women may avoid going out, which is reverse causation.

Pet ownershipModerate

Death from any cause

Mixed results: one meta-analysis linked dog ownership to 24% fewer deaths, another found no adjusted link for pets or dogs, and most older-adult cohorts found none.

Compared with people without a dog [1]; people without pets [2]; people without a dog [3]; older adults without pets [4]

The numbers and sources

Effect: Dog ownership was linked to 24% lower death from any cause (RR 0.76, 95% CI 0.67 to 0.86), and to 65% lower death in people with prior coronary events (RR 0.35, 95% CI 0.17 to 0.69) [1]. Pet owners had lower death rates unadjusted (HR 0.73, 95% CI 0.62 to 0.87) but not after adjustment (HR 0.40, 95% CI 0.04 to 3.78), and dog ownership was not significantly linked [2]. In Sweden, dog owners had lower death rates living alone (HR 0.67) and with others (HR 0.89) [3]. In older adults, 2 of 6 studies found lower death from any cause and 4 found no link [4].
Population: 3,837,005 participants in 10 studies [1]. 26 studies [2]. 3,432,153 Swedish adults [3]. Community-dwelling older adults in 26 longitudinal studies from 21 countries and regions [4].
Sources: [1], [2], [3], [4]
Caveats: Observational. The pooled estimates were not adjusted for confounders in the first review [1]. Health, income and living situation shape who owns a pet.

NappingStrong

Short naps and heart disease and death

Not linked to heart disease or death for naps under an hour in two pooled cohort analyses. One Swiss cohort linked napping once or twice weekly to fewer heart events.

Compared with people who did not nap [6][7][8]

The numbers and sources

Effect: Naps under 60 minutes a day were not linked to cardiovascular disease (p = 0.98) or death from any cause (p = 0.08). The dose-response curve for heart disease was J-shaped: risk dipped from 0 to 30 minutes, rose slightly to about 45 minutes and rose sharply after that [6]. Naps under an hour were not linked to death (HR 1.00, 95% CI 0.90 to 1.11) or cardiovascular disease (HR 1.03) [7]. In one cohort, napping 1 to 2 times a week was linked to fewer cardiovascular events (HR 0.52, 95% CI 0.28 to 0.95), and nap length was not linked to events [8].
Population: 151,588 participants in 11 cohorts, mean follow-up 11 years [6]. 371,306 participants in 21 studies [7]. 3,462 Swiss adults without heart disease followed 5.3 years, 155 events [8].
Sources: [6], [7], [8]
Caveats: Observational, with naps self-reported. The dip near 30 minutes was not tested against a trial. The Swiss finding rests on few events with a wide interval [8].

Food and drink

CoffeeStrong

Death from any cause

Linked to about 15% lower death from any cause at 3 to 4 cups a day than no coffee, in pooled cohorts, including decaffeinated coffee.

Compared with people drinking no coffee [2][3]

The numbers and sources

Effect: An umbrella review of 201 meta-analyses found the largest relative risk reduction at 3 to 4 cups a day against none, for all-cause mortality (RR 0.83, 95% CI 0.79 to 0.88) [2]. A dose-response meta-analysis of 21 cohorts found the largest reduction at 4 cups a day, 16% (95% CI 13 to 18%) [4]. An update of 40 studies found the lowest risk at 3.5 cups a day (RR 0.85, 95% CI 0.82 to 0.89), with no further fall at higher intakes. The link held by age, weight, smoking, alcohol and caffeine content of the coffee, and was weaker in the US than in Europe and Asia [3].
Population: 201 meta-analyses of observational studies and 17 of trials [2]. 997,464 participants and 121,915 deaths in 21 prospective studies [4]. 3,852,651 participants and 450,256 deaths in 40 studies [3].
Sources: [2], [4], [3], [5]
Caveats: Observational. Coffee drinkers differ from non-drinkers in smoking and other habits, and people who fall ill may stop drinking coffee, which can make coffee look protective (reverse causation). The umbrella review authors call for randomized trials to test cause [2]. A review of at least 15 Mendelian randomization studies found most gave no consistent support for coffee causing the health outcomes studied, while noting low statistical power [5].

Fish and seafoodStrong

Death from any cause

Linked to 6% lower death from any cause at the highest fish intake than the lowest, and 12% lower at 60 g a day than none, in pooled cohorts.

Compared with the lowest category of fish intake, or people who never ate fish [4]; 20 g a day less fish, in a dose-response model [5]; the lowest fatty fish intake [6]

The numbers and sources

Effect: Across 12 cohorts, the highest category of fish intake was linked to lower death from any cause than the lowest (RR 0.94, 95% CI 0.90 to 0.98), and 60 g of fish a day to 12% lower death than never eating fish (RR 0.88, 95% CI 0.83 to 0.93) [4]. In a second review, each extra 20 g a day was marginally linked to lower death (RR 0.98, 95% CI 0.97 to 1.00, I² 81.9%). The link was significant only in Asian studies, and Western studies suggested a U-shaped curve with the lowest risk near 20 g a day [5]. Fatty fish was linked to lower total death (RR 0.97, 95% CI 0.94 to 0.99), and lean fish was not [6].
Population: 672,389 participants and 57,641 deaths in 12 cohorts [4]. 911,348 participants and 75,451 deaths in 14 cohorts [5]. 104,763 deaths in the cohorts of fatty fish [6].
Sources: [4], [5], [6]
Caveats: Observational. People who eat more fish may differ in income, diet and other habits (confounding), and people who fall ill may change what they eat, which can run the link backward (reverse causation). The regional differences may reflect the type of fish and how it is cooked [5].

Olive oilStrong

Death from any cause

Linked to 11% lower death per extra 25 g of olive oil a day, and 19% lower at over 7 g a day than never, in cohorts.

Compared with 25 g a day less, in a dose-response model [3]; the bottom third of intake [4]; people who never or rarely used olive oil [5]

The numbers and sources

Effect: Pooled cohorts found a relative risk of 0.89 (95% CI 0.85 to 0.93) per extra 25 g a day [3]. The top against the bottom third of intake of monounsaturated fat sources gave 0.89 (95% CI 0.83 to 0.96), and in subgroups only olive oil, not monounsaturated fat in general, was linked to lower death [4]. In US nurses and health professionals followed 28 years, more than 7 g a day was linked to lower death than never or rarely using it (HR 0.81, 95% CI 0.78 to 0.84), and to lower death from heart disease, cancer, neurodegenerative and respiratory disease. Olive oil did not differ from other vegetable oils combined [5].
Population: 733,420 participants and 174,081 deaths [3]. 841,211 participants in 32 cohorts [4]. 60,582 women and 31,801 men free of heart disease and cancer at the start [5].
Sources: [3], [4], [5]
Caveats: Observational. Olive oil users may differ in diet, income and other habits. Heterogeneity was present for death [3]. Replacing butter, margarine, mayonnaise or dairy fat with olive oil was linked to lower death, but replacing other vegetable oils was not [5].

Whole grainsStrong

Death from any cause

Linked to 16% to 17% lower death from any cause at the highest whole grain intake or per three daily servings, in pooled cohorts.

Compared with 90 g a day less whole grain, in a dose-response model [3]; the lowest whole grain intake [4]

The numbers and sources

Effect: Per extra 90 g a day, the relative risk was 0.83 (95% CI 0.77 to 0.90, I² 83%) for death from any cause, 0.85 (95% CI 0.80 to 0.91) for cancer death and 0.78 (95% CI 0.70 to 0.87) for respiratory death. Lower risks were seen up to 210 to 225 g a day [3]. The highest against the lowest intake was linked to lower total death (RR 0.84, 95% CI 0.80 to 0.88) and cancer death (RR 0.88, 95% CI 0.83 to 0.94), and each extra 16 g a day to 7% lower total death (RR 0.93, 95% CI 0.92 to 0.94) [4].
Population: 45 prospective studies [3]. 786,076 participants and 97,867 deaths in 14 cohorts [4].
Sources: [3], [4]
Caveats: Observational, with high heterogeneity for total death (I² 83% and 74%) [3][4]. Confounding and reverse causation as above.

Green teaModerate

Death from any cause

Linked to 10% lower death in men and 18% lower in women at 5 or more cups a day than under 1 cup, in 8 pooled Japanese cohorts.

Compared with people drinking under 1 cup a day [2]

The numbers and sources

Effect: At 5 or more cups a day, death from any cause was lower in men (HR 0.90, 95% CI 0.87 to 0.94) and women (HR 0.82, 95% CI 0.74 to 0.90), as was death from heart disease (HR 0.82 in men, 0.75 in women) and cerebrovascular disease (HR 0.76 in men, 0.78 in women). In women, 1 to 4 cups a day was linked to slightly lower cancer death [2].
Population: 313,381 Japanese adults with 52,943 deaths over a mean 17.3 years in 8 cohorts [2].
Sources: [2]
Caveats: Observational, with intake reported once at the start. All participants were Japanese, where green tea is a daily habit, so the results may not carry to other populations.

SaltModerate

Stroke, heart events and death, with salt substitutes

Shown to lower stroke by 14%, major heart events by 13% and death by 12% in one large trial of older Chinese adults, an effect that drives pooled results.

Compared with regular salt [10][7][8]

The numbers and sources

Effect: Over a mean 4.74 years, stroke was less frequent with a 75% sodium chloride, 25% potassium chloride substitute than with regular salt (29.14 against 33.65 events per 1,000 person-years, rate ratio 0.86, 95% CI 0.77 to 0.96), as were major cardiovascular events (rate ratio 0.87, 95% CI 0.80 to 0.94) and death (rate ratio 0.88, 95% CI 0.82 to 0.95) [10]. A Cochrane review rated substitutes as probably reducing non-fatal stroke, non-fatal acute coronary syndrome and cardiovascular death slightly, by 20, 150 and 180 fewer per 100,000 person-years [7]. A network meta-analysis estimated 7 to 17 fewer deaths per 1,000 people, and found that removing the large Chinese trial left no significant effect on death from any cause [8].
Population: 20,995 people in 600 Chinese villages with past stroke, or aged 60 or older with high blood pressure, mean age 65.4, 49.5% women [10]. Adults in 3 trials with event data [7]. 34 trials [8].
Sources: [10], [7], [8]
Caveats: One open-label trial in rural China, where most salt is added at home, carries the pooled results, and no trial outside China reported deaths [8]. The reviewers describe very limited generalizability beyond Chinese populations who add much of their salt at home [8].

Fermented foodsLimited

Death from any cause, in cohorts

Linked to lower death from any cause and heart disease for cheese and fermented milks in cohorts. Miso and bread showed no link.

Compared with the lowest intake categories [12]

The numbers and sources

Effect: Comparing highest with lowest intake, cheese and fermented milks including yogurt were linked to lower death from any cause and from cardiovascular disease, and fermented milk to lower cancer death. Miso and bread were not linked to death [12].
Population: More than 3 million participants in 50 cohort studies [12].
Sources: [12]
Caveats: Observational. People who eat yogurt and cheese may differ in other habits, so confounding is possible. The abstract gives no effect sizes.

NutsLimited

Death from any cause, in cohorts

Linked to 17% to 22% lower death from any cause per daily serving of nuts in two cohort meta-analyses, with similar results for tree nuts and peanuts.

Compared with people eating 28 g a day less [6]; people eating one serving a day less, and the lowest intake group [7]

The numbers and sources

Effect: Per 28 g a day more, death from any cause was lower (RR 0.78, 95% CI 0.72 to 0.84), as were total cancer (RR 0.85) and deaths from respiratory disease, diabetes and infections [6]. Per serving a day, death from any cause was lower (RR 0.83, 95% CI 0.76 to 0.91), and 0.85 (95% CI 0.79 to 0.91) for the highest against the lowest intake [7].
Population: 15 cohort estimates for death from any cause [6]. 18 prospective studies with 48,818 deaths [7].
Sources: [6], [7]
Caveats: Observational, open to confounding and reverse causation. Heterogeneity was high (I² 66%) [6].

AlcoholModerate

Death from any cause and the light drinking debate

Mixed results: older pooled cohorts linked drinking to 13% lower death than not drinking, while a 2023 analysis correcting for former drinker bias found no lower risk.

Compared with non-drinkers [2]; lifetime nondrinkers [3]; current drinkers of more than 0 and up to 100 g a week [4]

The numbers and sources

Effect: Pooled cohorts found lower death from all causes in drinkers than non-drinkers (RR 0.87, 95% CI 0.83 to 0.92) [2]. Across 107 cohorts, after adjusting for former drinker bias and study quality, death did not differ between lifetime nondrinkers and occasional drinkers of under 1.3 g a day (RR 0.96, 95% CI 0.86 to 1.06) or low-volume drinkers of 1.3 to 24 g a day (RR 0.93, P = 0.07). Risk rose at 45 to 64 g a day (RR 1.19) and 65 g or more (RR 1.35), and female drinkers overall had higher risk than female lifetime nondrinkers (RR 1.22) [3]. Among current drinkers only, death was lowest at about 100 g a week or less. Against drinkers of up to 100 g a week, those drinking over 350 g a week had 4 to 5 years shorter life expectancy at age 40 [4]. A global analysis put the lowest-risk intake at 0 to 0.6 standard drinks a day for people aged 15 to 39, and 0.1 to 1.9 for people 40 and older, varying by region [5].
Population: 84 prospective cohorts [2]. 4,838,825 participants and 425,564 deaths in 107 cohorts [3]. 599,912 current drinkers without cardiovascular disease in 83 studies from 19 high-income countries [4]. Disease rates for 204 countries [5].
Sources: [2], [3], [4], [5]
Caveats: All observational. Many cohorts count "sick quitters", former drinkers who cut down or stopped for health reasons, among non-drinkers, which makes drinkers look healthier by comparison. Adjusting for this removed the lower risk [3]. Light drinkers also have other healthier habits, and adjusting for them weakened the apparent benefit in a UK cohort [6].

EggsModerate

Death from any cause

Mixed results: the highest egg intake was not linked to death in pooled cohorts, while dose-response models linked each extra egg to slightly higher death.

Compared with the lowest intake [10]; one fewer 50 g egg a day [6]; half an egg a day less [8]

The numbers and sources

Effect: The highest against the lowest intake was not linked to death from any cause (RR 1.02, 95% CI 0.94 to 1.11), heart disease or stroke, while each extra egg a week was linked to 2% higher death [10]. In 27,078 men, each extra 50 g egg a day was linked to higher death (HR 1.06) [6]. In US cohorts, each extra half egg a day was linked to higher death (HR 1.08, 95% CI 1.04 to 1.11), and not after adjusting for dietary cholesterol [8].
Population: 2,216,720 participants and 232,408 deaths in 33 cohorts [10]. 27,078 men in the ATBC study [6]. 29,615 US adults [8].
Sources: [10], [6], [8]
Caveats: Observational. The review rated certainty low to moderate [10].

SaltModerate

Death from any cause

Mixed results: salt reduction advice trials found no fewer deaths, one salt substitute trial found 12% fewer, and one cohort linked both low and high intakes to more deaths.

Compared with control groups without salt reduction advice or substitution [11]; control groups without sodium reduction counseling [12]; regular salt [10]; people excreting 4.00 to 5.99 g of sodium a day [13]

The numbers and sources

Effect: In trials of salt reduction advice, death did not fall in people with hypertension (RR 1.00, 95% CI 0.86 to 1.15) and the estimate in people without it was imprecise (RR 0.67, 95% CI 0.40 to 1.12) [11]. In the trial follow-up, deaths did not differ (RR 0.80, 95% CI 0.51 to 1.26) [12]. In one large trial of a salt substitute, deaths were fewer (rate ratio 0.88, 95% CI 0.82 to 0.95) [10]. In a cohort of 101,945 adults in 17 countries, estimated sodium of 7 g a day or more (OR 1.15, 95% CI 1.02 to 1.30) and below 3 g a day (OR 1.27, 95% CI 1.12 to 1.44) were both linked to more deaths or cardiovascular events than 4 to 6 g a day [13].
Population: 3,518 adults without and 3,085 with hypertension in trials [11]. 3,126 adults with prehypertension [12]. 20,995 older Chinese adults [10]. 101,945 adults in 17 countries followed a mean 3.7 years [13].
Sources: [11], [12], [10], [13]
Caveats: The cohort estimated sodium from one morning urine sample, and people eating little may be sicker to begin with, so reverse causation and confounding are possible [13]. Advice trials cut salt less than the substitute trial.

DairyModerate

Heart disease, stroke and death from any cause

Not linked to heart disease, cardiovascular disease or death from any cause for total dairy or milk, in 29 pooled cohorts.

Compared with lower intakes, in dose-response models [11]

The numbers and sources

Effect: No associations were found for total dairy, high-fat or low-fat dairy, or milk with death from any cause, coronary heart disease or cardiovascular disease. Fermented dairy was linked to slightly lower death (RR 0.98, 95% CI 0.97 to 0.99, per 20 g a day) and cardiovascular disease (RR 0.98), and cheese to 2% lower cardiovascular disease per 10 g a day, but these links weakened when one large Swedish study was removed [11].
Population: 938,465 participants with 93,158 deaths, 28,419 coronary cases and 25,416 cardiovascular cases in 29 cohorts [11].
Sources: [11]
Caveats: Observational. Heterogeneity was very high for the fermented dairy estimates (I² 94.4% and 87.5%) [11].

Dark chocolateLimited

Cardiovascular events, cocoa flavanol extract

One trial of a cocoa flavanol extract did not significantly lower total cardiovascular events in 21,442 older adults. Cardiovascular death, a secondary outcome, fell 27%.

Compared with placebo [6]

The numbers and sources

Effect: Over a median 3.6 years, total cardiovascular events occurred in 410 people taking cocoa extract and 456 taking placebo (HR 0.90, 95% CI 0.78 to 1.02). Cardiovascular death was lower (HR 0.73, 95% CI 0.54 to 0.98), and death from any cause did not differ significantly (HR 0.89, 95% CI 0.77 to 1.03). The authors reported no safety concerns [6].
Population: 21,442 US women aged 65 and older and men aged 60 and older without major cardiovascular disease [6].
Sources: [6]
Caveats: This tested a capsule of cocoa extract with 500 mg of flavanols, not chocolate. Secondary outcomes do not carry the weight of the primary one.
Why limited: This rests on one trial, of 21,442 people. Strength counts independent studies that agree, so one trial is limited at any size.

Red meat and processed meatLimitedStudied in women

Heart disease, cancer and death on diets lower in red meat

One trial in 48,835 women found a diet lower in red meat made little or no difference to death, heart disease or cancer, at low certainty.

Compared with diets higher in red meat [3]

The numbers and sources

Effect: The trial found little or no effect of lower red meat intake on all-cause death (HR 0.99, 95% CI 0.95 to 1.03), cardiovascular death (HR 0.98), cardiovascular disease (HR 0.99), cancer death (HR 0.95, 95% CI 0.89 to 1.01), colorectal cancer (HR 1.04, 95% CI 0.90 to 1.20) or breast cancer (HR 0.97). The review found 12 eligible trials, most measuring only blood markers [3].
Population: 48,835 women in one randomized trial, within a review of 12 trials lasting at least 6 months [3].
Sources: [3]
Caveats: The difference in red meat between the groups was small. The reviewers rated certainty low to very low [3].
Why limited: This rests on one trial, of 48,835 people. Strength counts independent studies that agree, so one trial is limited at any size.

Diets

Mediterranean dietModerate

Heart attack, stroke and cardiovascular death

Shown to lower heart attacks, strokes and cardiovascular deaths against minimal advice in adults at raised heart risk, with reviews disagreeing on how certain this is.

Compared with minimal intervention, such as a healthy diet brochure, and low-fat programmes, in a network meta-analysis [1]; a low-fat diet in the PREDIMED trial, and usual care in the Lyon Diet Heart Study [2]; a control diet of advice to reduce dietary fat, in Spanish adults [3]; a low-fat diet, in patients with coronary heart disease [4]; a prudent Western-type diet, in patients after a first heart attack [5]

The numbers and sources

Effect: In a network meta-analysis of 40 trials (35,548 people), Mediterranean programs lowered all-cause death (OR 0.72, 95% CI 0.56 to 0.92), cardiovascular death (OR 0.55, 95% CI 0.39 to 0.78), stroke (OR 0.65, 95% CI 0.46 to 0.93) and non-fatal heart attack (OR 0.48, 95% CI 0.36 to 0.65) against minimal intervention, at moderate certainty. Mediterranean and low-fat programs did not clearly differ for death or non-fatal heart attack [1]. A Cochrane review found low-quality evidence of little or no effect of the PREDIMED diets against a low-fat diet on cardiovascular death (HR 0.81, 95% CI 0.50 to 1.32) or total death (HR 1.0, 95% CI 0.81 to 1.24), and fewer strokes (HR 0.60, 95% CI 0.45 to 0.80, moderate quality) [2]. In PREDIMED, major cardiovascular events occurred in 3.8% on the olive oil diet, 3.4% on the nut diet and 4.4% on the low-fat control diet over a median 4.8 years (HR 0.69, 95% CI 0.53 to 0.91, and HR 0.72, 95% CI 0.54 to 0.95) [3]. In CORDIOPREV, events ran at 28.1 per 1,000 person-years on the Mediterranean diet against 37.7 on a low-fat diet over 7 years (adjusted HR 0.719 to 0.753) [4]. In the Lyon Diet Heart Study, cardiac death plus non-fatal heart attack totaled 14 events on a Mediterranean-type diet against 44 on a prudent Western-type diet over a mean 46 months [5].
Population: Adults at raised cardiovascular risk in 40 trials followed at least 9 months [1]; 30 RCTs with 12,461 people [2]; 7,447 Spanish adults aged 55 to 80 at high risk with no heart disease at entry [3]; 1,002 patients with coronary heart disease, 82.5% men [4]; 605 patients after a first heart attack [2][5].
Sources: [1], [2], [3], [4], [5]
Caveats: PREDIMED's first report was withdrawn after randomization problems at 2 of 11 sites. The reanalysis gave similar results after leaving out 1,588 affected participants [3]. The Cochrane review rated the death evidence low quality and the network meta-analysis rated it moderate [1][2]. In CORDIOPREV the effect appeared in men, and no difference was found in the 175 women [4]. CORDIOPREV was funded in part by olive oil foundations [4]. The Lyon trial added canola margarine and compared against usual care, and the Cochrane review rated it low quality [2].

Nordic dietModerate

Heart disease, diabetes and death in cohorts

Linked to lower death, heart attack, stroke and type 2 diabetes rates with higher adherence in cohort studies, at low certainty.

Compared with people with the lowest adherence to Nordic dietary patterns, in European cohorts [2]; people with lower adherence to the Nordic dietary pattern [9]; people in the lowest category of Nordic diet adherence [10]

The numbers and sources

Effect: In 13 cohorts (930,153 people), high against low adherence was linked to lower all-cause death (RR 0.78, 95% CI 0.69 to 0.87), cardiovascular death (RR 0.78), stroke (RR 0.88), heart attack (RR 0.80) and type 2 diabetes (RR 0.90, 95% CI 0.82 to 0.99) [10]. In 15 cohorts, total cardiovascular disease was 7% lower (RR 0.93, 95% CI 0.88 to 0.99, I² 88%) [2]. A 2025 meta-analysis of 32 cohorts found 12% lower type 2 diabetes and 8% lower cardiovascular disease [9].
Population: Adults in prospective cohorts, observational [2][9][10].
Sources: [2], [9], [10]
Caveats: Observational. Certainty ranged from very low to low [10], and was low for total cardiovascular disease [2].

Alternate-day fastingLimited

Markers of aging

One trial in healthy adults without obesity found 4 weeks of strict alternate-day fasting improved cardiovascular markers, with lower sICAM-1 and LDL after longer use.

Compared with eating every day, with no fasting days or daily fasting window

The numbers and sources

Effect: Strict alternate-day fasting for 4 weeks improved cardiovascular markers and the fat-to-lean ratio and raised β-hydroxybutyrate. After long-term alternate-day fasting, sICAM-1, LDL and triiodothyronine were lower [4].
Population: Healthy middle-aged adults without obesity [4].
Sources: [4]
Caveats: The abstract gives no participant numbers or effect sizes. Lifespan was not measured [4].

Paleo dietLimited

Long-term mortality

Linked to 10% lower all-cause mortality among people whose diets scored highest for a Paleolithic pattern in cohort studies.

Compared with people whose diets scored lowest for a Paleolithic pattern, in cohorts [3]

The numbers and sources

Effect: In 12 prospective cohorts, the highest adherers had lower all-cause mortality (RR 0.90, 95% CI 0.87 to 0.94), cancer mortality (RR 0.90, 95% CI 0.85 to 0.97) and coronary heart disease incidence (RR 0.84, 95% CI 0.70 to 1.00) [3].
Population: Adults in 12 prospective cohort studies, observational [3].
Sources: [3]
Caveats: Observational studies scoring how closely ordinary diets matched a Paleolithic pattern, not trials of people following the named diet. They cannot separate the diet from other differences between people.

Intermittent fastingLimited

Longevity and heart disease

Limited evidence: no pooled trial measured deaths, heart attacks or strokes, and cohort studies linked eating windows of 8 hours or less to more cardiovascular deaths.

Compared with no trial reported these outcomes [5]; US adults whose eating window was 12 to 14 hours [14]; US adults with other eating windows [15]

The numbers and sources

Effect: A Cochrane review of 18 pooled trials with 1,125 participants found no data on all-cause death, cardiovascular death, stroke, heart attack or heart failure [5]. In US survey cohorts, eating windows under 8 hours were linked to higher cardiovascular death against 12 to 14 hours [14], and mortality was lowest at about 11 to 12 hours a day [15]. These cohort results appear under Risks.
Population: Trials of 4 weeks to 6 months [5]; observational cohorts of 19,831 and 33,052 US adults [14][15].
Sources: [5], [14], [15]
Caveats: The cohorts measured habitual eating duration from two 24-hour diet recalls, not a chosen fasting plan [14][15].

One meal a dayLimited

Longevity

Limited evidence: in two large cohorts, adults who ate one meal a day had higher death rates than those who ate three meals a day.

Compared with people who ate three meals a day [5][6]

The numbers and sources

Effect: In 24,011 US adults aged 40 or older followed through 2015, eating one meal a day was linked to higher all-cause mortality (HR 1.30, 95% CI 1.03 to 1.64) and cardiovascular mortality (HR 1.83, 95% CI 1.26 to 2.65) than eating three meals a day [5]. In 148,438 South Korean adults aged 40 or older, the once-a-day group had higher all-cause and cancer mortality than the three-meals group, partly explained by waist size [6].
Population: 24,011 US adults in NHANES 1999 to 2014, 4,175 deaths [5]. 148,438 South Korean adults in the Korean Genome and Epidemiology Study [6].
Sources: [5], [6]
Caveats: Observational. Meal pattern was reported once, by 24-hour recall [5] or food questionnaire [6], and people who eat one meal a day may differ in health, work or income. Illness can also cause people to eat less often. No trial has measured lifespan on one meal a day.

Fasting-mimicking dietUntested

Longevity

Not studied in people. Fasting-mimicking cycles started at middle age lengthened life in mice, and in prematurely aging mice, and extended yeast lifespan.

No study in people. This rests on animal studies.

The numbers and sources

Effect: In mice, cycles of a fasting-mimicking diet twice a month started at middle age extended longevity, lowered visceral fat and cancer incidence, and slowed bone density loss. Alternating fasting and feeding also extended yeast lifespan [11]. In prematurely aging mice, the diet significantly prolonged lifespan, and in naturally aging mice it improved cognitive and gut measures [12].
Population: Yeast and mice [11]. Prematurely and naturally aging mice [12].
Sources: [11], [12]
Caveats: Lifespan has not been measured in people. The biological age findings above are markers, not lifespan [9]. The diet cycles in mice differ from the 5-day monthly cycles used in human trials [5].

Sirtfood dietUntested

Longevity

Not studied in people. Sirtuins are linked to aging in lab and animal research, but a review notes poor uptake of the food compounds that modulate them.

No study in people. This rests on lab studies, animal studies, how it acts in the body.

The numbers and sources

Effect: A 2022 review of sirtuin research describes their role in metabolism, antioxidant protection and aging, and reports that compounds modulating sirtuins worked in numerous lab and animal studies. It names low bioavailability and short retention of these modulators as a central problem, and warns of risks from careless sirtuin modulation [4]. Resveratrol did not change SIRT1 in people across 11 randomized trials [3].
Population: Lab and animal studies, reviewed [4]. Adults in 11 RCTs of resveratrol [3].
Sources: [4], [3]
Caveats: No study of the Sirtfood diet and lifespan or mortality in people was located. Sirtuin activation by food has not been shown in people [3].

Lab markers

25-hydroxyvitamin DStrong

Death from any cause

Linked to more deaths at low levels in cohorts, 67% more below 12 ng/mL, while 52 pooled trials of vitamin D supplements found no fewer deaths.

The numbers and sources

Effect: In pooled individual data with standardized measurements, adjusted hazard ratios for death against a reference of 30 to 40 ng/mL (75 to 99.99 nmol/L) were 1.15 (95% CI 1.00 to 1.29) at 16 to 20 ng/mL, 1.33 (1.16 to 1.51) at 12 to 16 ng/mL and 1.67 (1.44 to 1.89) below 12 ng/mL (30 nmol/L). Risk was not significantly higher at levels up to 50 ng/mL (125 nmol/L) [8]. Across 73 cohort studies, the bottom third of baseline levels had a relative risk of death of 1.35 against the top third (95% CI 1.22 to 1.49) [9]. In 52 randomized trials, vitamin D supplementation was not associated with death from any cause (RR 0.98, 95% CI 0.95 to 1.02) [10]. In a trial of 2,000 IU a day in 25,871 US adults, the hazard ratio for death was 0.99 (95% CI 0.87 to 1.12) [11].
Population: 26,916 participants in 8 European prospective studies, median age 61.6, 6,802 deaths over a median 10.5 years [8]. 849,412 participants in 73 cohort studies [9]. 75,454 participants in 52 trials [10]. 25,871 men aged 50 or older and women aged 55 or older [11].
Sources: [8], [9], [10], [11]
Caveats: The cohort findings are observational, so confounding and reverse causation, where illness lowers the level, cannot be ruled out. An earlier pooled analysis of trials found fewer deaths with vitamin D3 (RR 0.89, 95% CI 0.80 to 0.99) and not with vitamin D2 [9]. The Endocrine Society notes that average starting levels in many large trials were in a range most would consider adequate [3].

Alkaline phosphatase (ALP)Strong

Death from any cause

Linked to more deaths at higher levels, 38% more in the top third than the bottom third across pooled general population cohorts.

The numbers and sources

Effect: In a meta-analysis of general population cohorts, the relative risk of death from any cause was 1.38 (95% CI 1.17 to 1.63) comparing the top and bottom thirds of alkaline phosphatase, and 1.03 (1.01 to 1.06) per 5 U/L higher [8]. A second meta-analysis, in people with normal or preserved kidney function, found a non-linear association with a relative risk of 1.57 (95% CI 1.27 to 1.95) for the high group against the reference group [9]. In survivors of a heart attack, the top third had an adjusted hazard ratio for death of 1.43 (95% CI 1.08 to 1.89) against the bottom third, and the association was repeated in a US national survey sample [10].
Population: 19 cohorts with over 9.24 million participants and 242,953 deaths, which also covered other liver enzymes [8]. 24 studies with 147,634 participants [9]. 4,115 people with a previous heart attack and 14,716 US adults [10].
Sources: [8], [9], [10]
Caveats: Observational. A raised total can come from liver or from bone [2]. In the heart attack cohort the excess risk was present in people without evidence of kidney disease and was highest in people with both alkaline phosphatase and serum phosphate raised [10].

Apolipoprotein B (ApoB)Strong

Cardiovascular events and death in trials that lowered ApoB

Shown in trials: each 10 mg/dL lowered went with about 7% fewer major cardiovascular events, except with drugs that act outside the LDL receptor.

The numbers and sources

Effect: Across 29 randomized trials, each 10 mg/dL decrease in ApoB went with a relative risk of 0.93 for major adverse cardiovascular events (95% CI 0.90 to 0.97), 0.93 for cardiovascular death (0.88 to 0.98) and 0.95 for death from any cause (0.92 to 0.99). For major events the relative risk was 0.88 (0.83 to 0.93) with statins and 0.96 (0.94 to 0.99) with non-statin drugs that clear ApoB through the LDL receptor. Drugs that lower ApoB by other routes (CETP inhibitors, fibrates, niacin, omega-3 fatty acids) showed no reduction (1.02, 0.81 to 1.30), and the reduction in death from any cause was limited to statins [14]. An earlier analysis of 25 trials found each 10 mg/dL decrease went with 9% less coronary heart disease, 6% less major cardiovascular disease and no decrease in stroke. Across all drug classes, the fall in ApoB did not consistently predict risk better than the fall in LDL or non-HDL cholesterol [15].
Population: 332,912 patients in 29 trials of at least 1,000 people and 1 year [14]. 131,134 participants in 25 trials: 12 of statins, 4 of fibrates, 5 of niacin and 4 others [15].
Sources: [14], [15]
Caveats: Both analyses compare averages between trials, not individuals within them. None of the trials behind current treatment thresholds enrolled people by ApoB, and ApoB was often measured in only part of a trial population [2].

Cystatin CStrong

Death from any cause

Linked to more deaths at higher levels, and a cystatin C-based kidney estimate at least 30% below the creatinine-based one went with 69% more deaths.

The numbers and sources

Effect: In elderly US adults, hazard ratios for death rose across fifths of cystatin C, from 1.08 (95% CI 0.86 to 1.35) in the second fifth to 2.58 (2.03 to 3.27) in the top part of the highest fifth, while creatinine showed a J-shaped association [11]. In a meta-analysis of 16 cohorts, being reclassified to a lower kidney function category by cystatin C than by creatinine went with a higher risk of death, and to a higher category with a lower risk, with a net reclassification improvement for death of 0.23 (95% CI 0.18 to 0.28) [12]. In 440,526 UK adults, the cystatin C-based estimate of kidney function was the one most strongly associated with cardiovascular disease and death [13]. Across 23 outpatient cohorts, people whose cystatin C-based estimate was at least 30% lower than their creatinine-based one had 28.4 deaths per 1,000 person-years against 16.8 in those whose two estimates were within 30% (HR 1.69, 95% CI 1.57 to 1.82) [14].
Population: 4,637 participants of the Cardiovascular Health Study [11]. 90,750 participants in 11 general-population studies and 2,960 in 5 cohorts with chronic kidney disease, 13,202 deaths [12]. 440,526 UK Biobank participants [13]. 821,327 people in 23 outpatient cohorts, mean follow-up 11 years [14].
Sources: [11], [12], [13], [14]
Caveats: Observational. Cystatin C rises with inflammation, smoking and other conditions apart from kidney function [2][15], so a high level may mark illness already present.

Estimated GFR (eGFR)Strong

Death from any cause and from cardiovascular disease

Linked to more deaths at lower eGFR: against 95, risk was 18% higher at 60, 57% higher at 45 and about 3 times higher at 15.

The numbers and sources

Effect: In a collaborative meta-analysis of general population cohorts, risk of death was unrelated to eGFR between 75 and 105 mL/min/1.73 m2 and rose at lower values. Against an eGFR of 95, adjusted hazard ratios for death from any cause were 1.18 (95% CI 1.05 to 1.32) at 60, 1.57 (1.39 to 1.78) at 45 and 3.14 (2.39 to 4.13) at 15, with similar findings for cardiovascular death [6]. In a later individual-participant analysis of 114 cohorts, lower eGFR was significantly associated with higher all-cause and cardiovascular mortality, including in the mildest categories of chronic kidney disease [7].
Population: 105,872 participants in 14 studies with urine albumin-to-creatinine measurements and 1,128,310 in 7 studies with dipstick measurements [6]. 27,503,140 individuals in 114 cohorts, mean age 54, mean follow-up 4.8 years [7].
Sources: [6], [7]
Caveats: Observational. Urine albumin predicted death independently of eGFR, so eGFR alone gives part of the picture [6]. KDIGO notes that creatinine-based eGFR has a J-shaped association with most outcomes, with risk also rising above 105 mL/min/1.73 m2, and that very low creatinine often reflects frailty or low muscle mass [3].

Fasting glucoseStrong

Death from any cause

Linked to more deaths above 100 mg/dL and not between 70 and 100 mg/dL, in 97 pooled cohorts of 820,900 people.

The numbers and sources

Effect: In 97 prospective studies, fasting glucose above 100 mg/dL was associated with death and levels of 70 to 100 mg/dL were not. Diabetes went with a hazard ratio for death from any cause of 1.80 (95% CI 1.71 to 1.90) [10]. In an updated meta-analysis, prediabetes by any definition went with a relative risk of death from any cause of 1.13 (95% CI 1.10 to 1.17) in the general population [11]. An earlier meta-analysis put the relative risk at 1.13 for impaired fasting glucose by both the American and WHO definitions [9].
Population: 820,900 people in 97 prospective studies, with 123,205 deaths [10]. 10,069,955 people in 129 studies, median follow-up 9.8 years [11].
Sources: [10], [11], [9]
Caveats: Observational. The updated meta-analysis pooled several definitions of prediabetes, and found impaired glucose tolerance carried a higher risk than impaired fasting glucose [11].

Gamma-glutamyl transferase (GGT)Strong

Death from any cause

Linked to more deaths from any cause, about 7% higher risk for each 5 U/L in a meta-analysis of general population cohorts.

The numbers and sources

Effect: In a meta-analysis of 19 cohort studies, the relative risk of death from any cause was 1.60 (95% CI 1.42 to 1.80) comparing the top and bottom thirds of baseline GGT, and 1.07 (1.04 to 1.10) per 5 U/L higher, consistent with a linear relationship [7]. A second meta-analysis of 35 prospective studies found GGT was associated with increased all-cause and cardiovascular mortality even at levels inside the reference interval [8]. In 14,950 US adults without viral hepatitis followed for 12 years, GGT above 51 U/L in men or 33 U/L in women went with more deaths from any cause (HR 1.5, 95% CI 1.2 to 1.8), from liver disease (HR 13.0, 95% CI 2.4 to 71.5), from neoplasms (HR 1.5, 95% CI 1.01 to 2.2) and from diabetes (HR 3.3, 95% CI 1.4 to 7.6) [4].
Population: Over 9.24 million participants and 242,953 deaths in 19 cohorts, which also covered other liver enzymes [7]. 571,511 participants and 72,196 deaths in 35 studies [8]. 14,950 adults in the third US National Health and Nutrition Examination Survey [4].
Sources: [7], [8], [4]
Caveats: Observational. Many of the pooled studies did not exclude people with liver disease or heavy alcohol use, which the authors of one meta-analysis say may have obscured the results [8]. Liver disease deaths were 0.13% of the US cohort, so the interval around that estimate is wide [4].

High-sensitivity C-reactive protein (hs-CRP)Strong

Death from any cause

Linked to more deaths at higher levels, about 75% more in the highest than the lowest category across pooled general population studies.

The numbers and sources

Effect: In a meta-analysis of prospective studies in the general population, the highest category of hs-CRP had a pooled risk ratio of 1.75 for death from any cause against the lowest (95% CI 1.55 to 1.98), 2.03 for cardiovascular death (1.65 to 2.50) and 1.25 for cancer death (1.13 to 1.38) [10]. In pooled individual records, the risk ratio per threefold higher CRP after adjustment for conventional risk factors was 1.55 for vascular death (95% CI 1.37 to 1.76) and 1.54 for non-vascular death (1.40 to 1.68) [6].
Population: 83,995 participants in 14 studies [10]. 160,309 people in 54 studies [6].
Sources: [10], [6]
Caveats: Observational. CRP rises with many illnesses, so a higher level may mark disease already present. The link with cancer death was seen in men and not in women [10].

IGF-1Strong

Death from any cause

Linked to more deaths at both low and high levels in two pooled analyses, 27% to 33% more at low levels and 18% to 23% more at high levels.

The numbers and sources

Effect: In a meta-analysis of 12 studies, people with low IGF-1 had a hazard ratio for death from any cause of 1.27 (95% CI 1.08 to 1.49) and people with high IGF-1 had 1.18 (1.04 to 1.34), each against the middle of the distribution. The relation was U-shaped, and it held for both cancer deaths and cardiovascular deaths. The predicted hazard ratio was 1.56 (1.31 to 1.86) for the 10th percentile against the 50th, and 1.29 (1.06 to 1.58) for the 90th against the 50th [4]. A later meta-analysis of 19 cohorts found 1.33 (1.14 to 1.57) for low against middle levels and 1.23 (1.06 to 1.44) for high against middle, with the lowest mortality at 120 to 160 ng/mL [5].
Population: 14,906 participants in 12 studies [4]. 30,876 adults in 19 prospective cohorts [5].
Sources: [4], [5]
Caveats: Observational. The cohorts used different assays, and results from different assays differ, most of all at the upper end [2], so the 120 to 160 ng/mL figure is not tied to any one laboratory's result. The later analysis notes that low IGF-1 may reflect frailty and illness already present [5].

Thyroid stimulating hormone (TSH)Strong

Death from any cause

Linked to more deaths at low TSH, with no rise in deaths at mildly raised TSH, across three pooled analyses of cohorts.

The numbers and sources

Effect: People with TSH below 0.45 mIU/L and normal free T4 had a hazard ratio for death from any cause of 1.24 (95% CI 1.06 to 1.46) against people with TSH of 0.45 to 4.49 mIU/L [17]. Total mortality was not increased in people with TSH of 4.5 to 19.9 mIU/L and normal thyroxine [16]. In 134,346 adults, the lowest risk of cardiovascular disease and death fell at the 60th to 80th percentiles of TSH, medians of 1.90 to 2.90 mIU/L. Against that band, the lowest fifth of TSH had a hazard ratio for death from any cause of 1.09 (1.05 to 1.14) [18].
Population: 52,674 adults in 10 cohorts, 8,527 deaths [17]. 55,287 adults in 11 cohorts, 9,664 deaths [16]. 134,346 adults in 26 cohorts, median age 59 [18].
Sources: [17], [16], [18]
Caveats: Observational. A US guideline notes that acute illness can suppress TSH [3], so reverse causation cannot be ruled out. The percentile figures are adjusted for age and sex [18].

Total testosteroneStrongStudied in men

Death from any cause in men

Linked to more deaths in men at low testosterone: below 213 ng/dL for death from any cause and below 153 ng/dL for cardiovascular death in pooled cohorts.

The numbers and sources

Effect: In pooled individual data from cohorts that measured testosterone by mass spectrometry, men with baseline total testosterone below 7.4 nmol/L (213 ng/dL) had higher mortality from any cause, and men below 5.3 nmol/L (153 ng/dL) had higher cardiovascular mortality [16]. An earlier meta-analysis of 11 studies of death from any cause and 7 of cardiovascular death found low testosterone associated with both, with considerable differences between studies that tracked the men's age and testosterone levels, the length of follow-up and the time of day of sampling [17].
Population: 9 cohorts with individual data covering 255,830 participant-years, plus 11 studies with summary estimates [16]. 16,184 men for death from any cause and 11,831 for cardiovascular death, mean age 61, mean follow-up 9.7 years [17].
Sources: [16], [17]
Caveats: Observational. The earlier analysis concluded that the differences between studies suggest effects driven by the underlying health of the cohorts, so low testosterone may mark illness [17].

Urine albumin to creatinine ratioStrong

Death from any cause and cardiovascular death

Linked to more deaths at higher ratios with no threshold, 63% more at about 30 mg/g than at about 5 mg/g across 105,872 people.

The numbers and sources

Effect: In a meta-analysis of general population cohorts, the ratio was related to death on a continuous scale without a threshold. Against a ratio of 0.6 mg/mmol (about 5 mg/g), the adjusted hazard ratio for death from any cause was 1.20 (95% CI 1.15 to 1.26) at 1.1 mg/mmol (10 mg/g), 1.63 (1.50 to 1.77) at 3.4 mg/mmol (about 30 mg/g) and 2.22 (1.97 to 2.51) at 33.9 mg/mmol (about 300 mg/g). Findings were similar for cardiovascular death, and the ratio and estimated GFR each added risk without interaction [5]. In a later meta-analysis of 114 cohorts, higher ratios were associated with higher rates of each of 10 outcomes studied: kidney failure, death from any cause, cardiovascular death, acute kidney injury, hospitalization, coronary heart disease, stroke, heart failure, atrial fibrillation and peripheral artery disease [8].
Population: 105,872 participants in 14 general population cohorts with ratio measurements [5]. 9,067,753 people with albuminuria data in 114 cohorts, with a median ratio of 11 mg/g in the main population [8].
Sources: [5], [8]
Caveats: Observational, with the ratio measured once at baseline. The first meta-analysis printed its ratios in mg/mmol, and the mg/g figures other than 10 are converted here with the factor on this page [5].

AlbuminModerate

Death from any cause

Linked to more deaths at lower albumin, with the odds of death 24% to 56% higher for each 0.25 g/dL lower in a review of cohort studies.

The numbers and sources

Effect: A review of cohort studies in people with and without illness found that albumin was inversely related to the risk of death in a graded manner over its entire range. The estimated increase in the odds of death was 24% to 56% for each 2.5 g/L (0.25 g/dL) lower albumin, and the association covered cardiovascular death as well as death overall [7]. In a meta-analysis of 90 cohort studies of acutely ill patients, each 10 g/L (1 g/dL) decline in albumin raised the odds of death by 137% and of complications by 89%, and the association appeared independent of nutritional status and inflammation [8].
Population: Cohort studies of many populations, from people with no known illness to patients with acute or chronic illness [7]. 291,433 acutely ill patients in 90 cohort studies [8].
Sources: [7], [8]
Caveats: Observational. The review's authors consider it likely that albumin is a highly sensitive indicator of preclinical disease and disease severity, and state that there is no direct evidence that albumin itself protects [7].

Blood urea nitrogen (BUN)Moderate

Death in heart failure and after a heart attack

Linked to more deaths in people hospitalized with heart failure, about 2.3 times the risk in the top quarter of BUN in one cohort.

The numbers and sources

Effect: In 541 patients admitted for decompensated heart failure, the risk of death rose with each quartile of BUN, with an adjusted relative risk of 2.3 (95% CI 1.3 to 4.1) in the top against the bottom quartile [6]. In a US registry of heart failure admissions, a BUN of 43 mg/dL or above on admission was the single strongest predictor of death in hospital among 39 variables examined [4]. In older Medicare patients, adjusted 1-year mortality rose 3% per 5 mg/dL of BUN after a heart attack and 7% per 5 mg/dL in heart failure, with risk climbing steadily above 17 mg/dL and 16 mg/dL [7].
Population: 541 patients, mean age 63, followed for a mean of 343 days [6]. 33,046 hospitalizations in the derivation cohort and 32,229 in the validation cohort, mean age 72.5 [4]. 44,437 patients hospitalized for heart attack and 56,652 for heart failure, aged 65 or older [7].
Sources: [6], [4], [7]
Caveats: Observational, in people already in hospital with heart disease. The authors of the first study suggest that a raised BUN in this setting reflects reduced blood flow to the kidneys from the heart failure itself [6]. These cut points were derived to predict death in hospital patients and are not reference ranges.

Calcium levelModerate

Death and cardiovascular disease

Linked to slightly more deaths and cardiovascular disease at higher calcium, 13% more deaths per standard deviation, in pooled studies of people without a calcium disorder.

The numbers and sources

Effect: In a systematic review of studies in people selected neither for kidney disease nor for abnormal calcium, 5 of 11 studies reported a significant positive association between serum calcium and death, and 8 of 13 between serum calcium and cardiovascular disease. Pooled across 8 studies each, the hazard ratio per standard deviation of serum calcium was 1.13 (95% CI 1.09 to 1.18) for death and 1.08 (1.04 to 1.13) for cardiovascular disease. With hazard ratios adjusted for cardiovascular risk factors, the pooled figure for cardiovascular disease was 1.04 (1.01 to 1.08) [5].
Population: Longitudinal studies with at least 1 year of follow-up, located in a search to December 2014 [5].
Sources: [5]
Caveats: Observational. The authors state that these data cannot determine causality [5]. The abstract gives no total head count.

Copper levelModerate

Death from any cause at higher levels

Linked to more deaths from any cause at higher serum copper, about 22% more per 20 µg/dL across pooled cohort studies.

The numbers and sources

Effect: In a meta-analysis of cohort studies, elevated circulating copper went with more deaths from any cause (RR 1.56, 95% CI 1.35 to 1.79), and each 20 µg/dL higher copper had a hazard ratio of 1.22 (1.04 to 1.44). The authors rated the certainty of this evidence as moderate [7]. In US adults from the 2011 to 2016 national survey, those in the top third of serum copper (19.84 µmol/L or above, about 126 µg/dL) had more deaths from any cause than those in the bottom third (below 16.31 µmol/L, about 104 µg/dL), with a hazard ratio of 2.84 (1.66 to 4.87) [10].
Population: 47,813 people in 17 cohort studies [7]. 5,412 US adults followed for a mean of 5.85 years, with 356 deaths [10].
Sources: [7], [10]
Caveats: Observational. The US cohort had 356 deaths and wide intervals, and its authors state that causal relationships require further investigation [10]. The abstract prints the cut points for the thirds as mol/L, read here as µmol/L, and the µg/dL figures are converted with the factor on this page. Inflammation and illness raise serum copper [3], which may account for part of the link.

DHEA sulfate (DHEA-S)Moderate

Death from any cause

Linked to more deaths at low levels in older adults, 46% more in pooled cohorts, with a J-shaped pattern in one cohort where the highest levels also carried more.

The numbers and sources

Effect: In a meta-analysis of six prospective studies of older adults, the lowest DHEA-S category against higher levels had a pooled risk ratio of 1.46 for death from any cause (95% CI 1.25 to 1.70) and 1.49 for cardiovascular death (1.11 to 1.99). The association was found in men (1.41, 1.18 to 1.69) and did not reach significance in women (1.72, 0.99 to 2.99) [3]. In a German cohort followed for a median 17.7 years, both the lowest and the highest fifths for sex and age had more deaths from any cause than the middle fifth (hazard ratios 1.51, 1.25 to 1.82, and 1.31, 1.08 to 1.58) [4]. In the Atherosclerosis Risk in Communities cohort, DHEA-S below the 25th percentile for sex (70.0 µg/dL in men, 37.1 µg/dL in women) had hazard ratios for death of 1.12 in men (1.01 to 1.25) and 1.19 in women (1.03 to 1.37) [5]. In Swedish men aged 69 to 81, the lowest quarter against the rest had a hazard ratio of 1.54 (1.21 to 1.96) [6]. Among people who already had cardiovascular disease, lower DHEA-S had a pooled risk ratio for death of 1.47 (1.38 to 1.56) [7].
Population: 6,744 people in six prospective studies [3]. 7,370 men and women with mean ages of 55.0 and 52.4 [4]. 8,143 adults with a mean age of 63 and no cardiovascular disease at the start, followed for 18 years [5]. 2,644 men, mean follow-up 4.5 years [6]. 18 studies of patients with cardiovascular disease [7].
Sources: [3], [4], [5], [6], [7]
Caveats: Observational. DHEA-S falls with age and may mark illness already present. The trials that raised DHEA-S measured other outcomes (see the trial block below).

EstradiolModerate

Death from any cause in men

Linked to more deaths in men at very low estradiol, below 5.1 pmol/L or about 1.4 pg/mL, in pooled cohorts that used mass spectrometry.

The numbers and sources

Effect: In pooled individual data from cohorts of community-dwelling men, those with baseline estradiol below 5.1 pmol/L, about 1.4 pg/mL, had higher mortality from any cause. The abstract gives no hazard ratio for estradiol [11].
Population: 9 cohorts with individual data covering 255,830 participant-years, plus 11 studies with summary estimates [11].
Sources: [11], [4]
Caveats: Observational, with differences between studies and imputed missing data [11]. The level in question sits far below the 30 to 100 pg/mL limit of quantitation of most direct immunoassays [4].

Free T4 (free thyroxine)Moderate

Cardiovascular disease and death

Linked in a J shape: risk of cardiovascular disease and death was lowest at the 20th to 40th percentiles of free T4, and the top fifth had 34% more deaths.

The numbers and sources

Effect: In pooled data on 134,346 adults, risk was lowest at the 20th to 40th percentiles of free T4, medians of 13.5 to 14.8 pmol/L. Against that band, the top fifth had hazard ratios of 1.20 (95% CI 1.11 to 1.31) for cardiovascular events or death combined, 1.34 (1.20 to 1.49) for death from any cause, 1.57 (1.31 to 1.89) for cardiovascular death and 1.22 (1.11 to 1.33) for cardiovascular events [9]. In older US adults with thyroid tests in the reference range, the top quartile of free T4 had more coronary heart disease, heart failure and deaths than the bottom quartile, by 8.0, 7.8 and 14.3 per 1,000 person-years [8].
Population: 134,346 adults in 26 cohorts, median age 59 [9]. 2,843 older adults [8].
Sources: [9], [8]
Caveats: Observational, so illness that shifts free T4 before an event cannot be ruled out. The pooled figures are adjusted for age and sex, and the percentile medians are printed in pmol/L [9].

FructosamineModerate

Heart attack, stroke and death from any cause

Linked to more heart attacks and deaths at high levels in US and Swedish cohorts, and to slightly more deaths at the lowest levels.

The numbers and sources

Effect: In a US cohort followed for two decades, elevated fructosamine and glycated albumin were each significantly associated with coronary heart disease, ischemic stroke, heart failure and death after adjustment for cardiovascular risk factors. The associations were similar in size to those for HbA1c and tended to be J-shaped, with raised risk at the lowest levels as well [7]. In a Swedish cohort, the hazard ratio in people above 2.70 mmol/L on that cohort's scale, against its reference range, was 2.88 (95% CI 2.70 to 3.07) for heart attack and 2.31 (2.21 to 2.41) for death from any cause, and the associations weakened but remained after adjustment for glucose [8]. Among Swedish adults without diabetes, the lowest tenth of fructosamine had a hazard ratio for death of 1.20 against the middle eight tenths, falling to 1.11 (95% CI 1.07 to 1.15) after adjustment for six other blood markers [9].
Population: 11,104 adults with and without diabetes, with 1,096 new cases of coronary heart disease, 605 ischemic strokes, 1,432 cases of heart failure and 2,860 deaths [7]. 338,443 people followed 19 years on average, with 21,526 heart attacks and 73,458 deaths [8]. 215,011 people without diabetes, 41,388 deaths over 25 years [9].
Sources: [7], [8], [9]
Caveats: Observational. The Swedish cohort reported fructosamine in mmol/L on a scale about ten times this page's, so its cut points are not placed here. Its authors found that smoking and chronic inflammation at least partly explained the higher mortality at low fructosamine [9]. No source here tested whether lowering fructosamine changes these outcomes, and a laboratory guideline states that treatment goals have not been established [2].

HematocritModerate

Death from any cause

Linked to more deaths at both low and high hematocrit in an Iranian cohort of 49,983 adults and in the Framingham cohort.

The numbers and sources

Effect: In 49,983 Iranian adults followed for a mean of 5 years, the relation between hematocrit and death was U-shaped in both sexes. In women, against 35% to 39%, adjusted hazard ratios for death were 2.12 (95% CI 1.26 to 3.58) at 15% to 29%, 1.58 (1.29 to 1.93) at 45% to 49% and 2.05 (1.17 to 3.60) at 50% to 54%. In men, against 40% to 44%, they were 2.12 (1.30 to 3.45) at 15% to 29%, 1.64 (1.22 to 2.21) at 30% to 34% and 1.29 (1.04 to 1.61) at 50% to 54%. Deaths from cardiovascular disease followed the same pattern [7]. In the Framingham cohort over 34 years, people in the highest fifth of hematocrit had more deaths from any cause and more cardiovascular disease, and women with lower values also had more cardiovascular events, which the authors describe as a J-shaped or U-shaped relation [8].
Population: 49,983 adults aged 40 to 75 in the Golestan Cohort Study in northeast Iran, 2,262 deaths [7]. 5,209 men and women in the Framingham cohort [8].
Sources: [7], [8]
Caveats: Observational. In the Iranian cohort, leaving out the first 2 years of follow-up weakened the association at low hematocrit in both sexes and removed the association at high hematocrit in men, which points to illness before death as part of the explanation. That cohort measured hematocrit by spinning blood in a tube, not by analyzer [7]. The Framingham abstract gives no hazard ratios [8].

HemoglobinModerate

Death from any cause

Linked to more deaths at both low and high levels in three cohorts of middle-aged and older adults, with the lowest risk in the middle of the range.

The numbers and sources

Effect: In 5,888 US adults aged 65 and older followed 11.2 years, the lowest fifth of hemoglobin (below 13.7 g/dL in men, below 12.6 g/dL in women) and the highest fifth had adjusted hazard ratios for death of 1.33 (95% CI 1.15 to 1.54) and 1.17 (1.01 to 1.36) against the fourth fifth. Anemia by WHO criteria had a hazard ratio of 1.38 (1.19 to 1.54) [9]. In 17,030 Canadian adults aged 66 and older, hemoglobin below 11.0 g/dL in people with normal kidney function had a hazard ratio for death of 4.29 (3.55 to 5.12), and risk was lowest at 13.0 to 15.0 g/dL in women and 14.0 to 17.0 g/dL in men [10]. In 292,194 Korean adults over 40, the association with cardiovascular and all-cause death was U-shaped or J-shaped. Men whose anemia had resolved by a second exam had a hazard ratio for death of 0.67 (0.59 to 0.77) against men whose anemia persisted [11].
Population: 5,888 community-dwelling adults aged 65 and older in 4 US communities [9]. 17,030 community-dwelling adults aged 66 and older in Calgary, 1,983 deaths over about 3 years [10]. 170,078 men and 122,116 women without cardiovascular disease, median follow-up 8 years [11].
Sources: [9], [10], [11]
Caveats: Observational. Illness that lowers or raises hemoglobin before death cannot be ruled out, although adjustment for kidney function, inflammation and frailty did not remove the association in the US cohort [9].

Hemoglobin A1c (HbA1c)Moderate

Death from any cause

Linked to more deaths at both ends: above 6.0% and below 5.0% in people without diabetes, and above 9.0% and below 6.0% in people with diabetes.

The numbers and sources

Effect: A meta-analysis of observational studies found higher death rates from any cause in people without diabetes at HbA1c above 6.0% (HR 1.74, 95% CI 1.38 to 2.20) and below 5.0% (HR 1.19, 95% CI 1.04 to 1.36). In people with diabetes, rates were higher above 9.0% (HR 1.69, 95% CI 1.09 to 2.66) and below 6.0% (HR 1.57, 95% CI 1.14 to 2.17). Results for cardiovascular death were similar [10]. A US cohort of adults without diabetes also found a J-shaped association with death from any cause [8].
Population: 46 observational studies in the meta-analysis, of 74 reviewed [10]. 11,092 adults [8].
Sources: [10], [8]
Caveats: Observational. The excess at low values may reflect illness or conditions that lower HbA1c, not an effect of low blood sugar itself, and the studies cannot separate the two.

IGF-1Moderate

Death in acromegaly after treatment that lowered IGF-1

Linked to mortality in acromegaly: deaths were near expected when treatment brought IGF-1 into the normal range, and 2.5 times expected when it stayed raised.

The numbers and sources

Effect: In a meta-analysis of mortality studies in acromegaly, a condition of excess growth hormone, people whose IGF-1 was normal for age and sex at last follow-up after treatment had a standardized mortality ratio of 1.1 (95% CI 0.9 to 1.4). People whose IGF-1 stayed raised had 2.5 (1.6 to 4.0) [12].
Population: Published mortality series in acromegaly from 1965 to 2008 that reported deaths by hormone level at last follow-up [12].
Sources: [12]
Caveats: Observational comparisons within a rare disease, not randomized. People whose IGF-1 came down may have differed in other ways from those in whom it did not.

Mean corpuscular volume (MCV)Moderate

Death from any cause and from heart and vessel disease

Linked to more deaths at higher MCV in two East Asian cohorts, including a 44% to 55% higher death rate in the top quarter of non-anemic Korean adults.

The numbers and sources

Effect: In non-anemic Korean adults, the top quarter of MCV (95.8 fL or above in men, 94.2 fL or above in women) had adjusted hazard ratios for death from any cause of 1.44 (95% CI 1.15 to 1.80) in men and 1.55 (1.08 to 2.22) in women, against a reference group at 90.5 to 93.0 fL in men and 89.2 to 91.6 fL in women. In men the hazard ratio for death from liver cancer was 3.55 (1.75 to 7.21) [10]. In a Taiwanese community cohort, higher MCV went with more deaths from stroke and heart disease combined (adjusted hazard ratio 1.42, P for trend 0.012) [11].
Population: 36,260 adults aged 40 and older having routine check-ups in Seoul, 1,107 deaths over a mean 8.0 years [10]. 66,294 adults aged 20 and older in Keelung, Taiwan, followed 9 years [11].
Sources: [10], [11]
Caveats: Observational, in two East Asian populations. The Taiwanese abstract does not state which MCV levels were compared, and its findings for single causes of death did not reach significance [11]. Neither study shows that MCV itself changes risk.

Methylmalonic acid (MMA)Moderate

Death from any cause

Linked to more deaths at higher levels in US and Dutch cohorts, 62% more at 250 nmol/L or above than below 120.

The numbers and sources

Effect: In a US national survey cohort, adults with methylmalonic acid of 250 nmol/L or more, compared with below 120, had hazard ratios of 1.62 (95% CI 1.43 to 1.84) for death from any cause and 1.66 (1.22 to 2.27) for cardiovascular death after adjustment. The association was especially marked in people with normal B12 and was repeated in a later round of the survey [9]. In a Dutch cohort, each doubling of methylmalonic acid had a hazard ratio for death of 1.67 (95% CI 1.25 to 2.22), independent of vitamin B12, kidney function and sex, and more pronounced in people with impaired kidney function [10].
Population: 23,437 US adults, with 3,019 deaths over a median 13.5 years in the 1999 to 2004 rounds [9]. 1,533 Dutch adults aged 60 to 75 followed for 8.5 years, 72 deaths [10].
Sources: [9], [10]
Caveats: Observational. In the Dutch cohort, vitamin B12, kidney function, age and sex together explained 22% of the variation in methylmalonic acid, so a raised level reflects more than B12 status. No trial cited here tested whether lowering it changes survival [10].

NT-proBNPModerate

Death from any cause and from cardiovascular disease

Linked to about 2.4 times the deaths from any cause in the highest against the lowest category across 11 general population studies.

The numbers and sources

Effect: In a meta-analysis of prospective studies in the general population, the highest category of NT-proBNP against the lowest had a hazard ratio of 2.44 (95% CI 2.11 to 2.83) for death from any cause, 3.77 (2.85 to 5.00) for cardiovascular death and 2.35 (1.45 to 3.82) for death from coronary heart disease [7]. Among patients with acute heart failure in a pooled emergency department study, a level above 5,180 pg/mL at presentation went with death within 76 days (OR 5.2, 95% CI 2.2 to 8.1) [5]. In a meta-analysis by kidney function, the unadjusted risk ratio for death with a raised NT-proBNP was 3.01 (2.53 to 3.58) in patients with preserved kidney function and 3.25 (2.45 to 4.30) in patients with reduced function [8].
Population: 25,715 people in 11 studies [7]. 720 patients with acute heart failure among 1,256 with acute breathlessness [5]. 32,203 patients in 30 studies [8].
Sources: [7], [5], [8]
Caveats: Observational. The studies defined the highest and lowest categories by different cut points [7].

Omega-3 indexModerate

Death from any cause

Linked to fewer deaths at higher blood levels of long-chain omega-3: 15% to 18% lower risk in the top fifth than the bottom across 17 pooled cohorts.

The numbers and sources

Effect: In a pooled analysis of 17 prospective cohorts, risk of death from any cause was 15% to 18% lower in the highest than the lowest fifth of circulating EPA, DPA and DHA after adjustment for other risk factors. Similar relations were seen for death from cardiovascular disease, cancer and other causes, and none for the plant omega-3 alpha-linolenic acid [7]. In the Framingham Offspring cohort, the highest fifth of the omega-3 index had a 34% lower risk of death from any cause than the lowest [6].
Population: 42,466 people with 15,720 deaths over a median 16 years [7]. 2,500 participants with 350 deaths [6].
Sources: [7], [6]
Caveats: Observational. The pooled analysis measured omega-3 fatty acids in several blood fractions, not only as the red cell index [7].

Platelet countModerate

Death from any cause

Linked to more deaths at low counts in two European cohorts, and at counts above 300 thousand/µL in one of them.

The numbers and sources

Effect: In Danish adults with counts of 100 to 450 thousand/µL, the relationship with death was U-shaped, with more deaths below 175 or above 300. Against 201 to 250, a count of 301 to 450 had an adjusted hazard ratio for death of 1.42 (95% CI 1.06 to 1.90) [7]. In Italian adults, a count under the lower limit for age and sex had a hazard ratio of 1.92 (1.38 to 2.67) for death from any cause, 1.77 (1.03 to 3.05) for cancer death and 3.16 (1.84 to 5.42) for deaths that were neither cardiovascular nor cancer. A count over the upper limit was not associated with any cause of death [8].
Population: 21,252 adults aged 20 and older in a Danish general population study [7]. 21,563 adults, mean age 55.6, in the Moli-sani study, 1,130 deaths over a median 8.2 years [8].
Sources: [7], [8]
Caveats: Observational. The two cohorts disagree about counts toward the high end. In the Italian study 87.7% of the low counts were still above 100 thousand/µL [8]. An illness that alters the platelet count before death cannot be ruled out.

Potassium levelModerate

Death from any cause

Linked to more deaths at both low and high potassium, with the lowest risk at 4 to 4.5 mEq/L across 1.2 million people.

The numbers and sources

Effect: In an individual-level meta-analysis of 27 cohorts, the relation between potassium and death was U-shaped, with the lowest risk at 4 to 4.5 mmol/L. Against 4.2 mmol/L, the adjusted hazard ratio for death from any cause was 1.22 (95% CI 1.15 to 1.29) at 5.5 mmol/L and 1.49 (1.26 to 1.76) at 3.0 mmol/L. Risks were similar across levels of kidney function [8]. In two US community cohorts, potassium of 5.0 mEq/L or above went with more deaths from any cause than 4.0 to 4.4 mEq/L (HR 1.41, 95% CI 1.12 to 1.76), and the association was stronger in people using diuretics [9].
Population: 1,217,986 participants in 10 general population, 7 high cardiovascular risk and 10 chronic kidney disease cohorts, followed for a mean of 6.9 years [8]. 9,651 adults free of cardiovascular disease in the Multi-Ethnic Study of Atherosclerosis and the Cardiovascular Health Study [9].
Sources: [8], [9]
Caveats: Observational, from potassium measured in outpatients at baseline. The authors of the US cohort study write that whether keeping potassium within the normal range improves outcomes requires future study [9].

Prostate-specific antigen (PSA)ModerateStudied in men

Death from prostate cancer decades later, by PSA in midlife

Linked to later death from prostate cancer: 90% of deaths by age 85 were in men whose PSA at 60 was above 2 ng/mL.

The numbers and sources

Effect: In Swedish men who gave blood at age 60 and were followed to 85, PSA predicted death from prostate cancer with an area under the curve of 0.90 (95% CI 0.84 to 0.96). Men in the top quarter, above 2 ng/mL, accounted for 90% of the deaths (78% to 100%), though only a minority of them died of the disease. Men at the median or lower, 1 ng/mL or below, had a 0.5% risk of metastasis and a 0.2% risk of death from prostate cancer by 85 [17]. In men who gave blood at 45 to 49, the top tenth of PSA (1.6 µg/L or above) accounted for 44% of later prostate cancer deaths (34% to 53%). Below the median of 0.68 µg/L, the 15-year risk of metastasis was 0.09% [18]. The AUA/SUO guideline cites a US cohort in which 82%, 71% and 86% of lethal cases arose in men above the median at ages 40 to 49, 50 to 54 and 55 to 59 (medians 0.68, 0.88 and 0.96 ng/mL) [5].
Population: 1,167 men aged 60 who gave blood in 1981 [17]. 21,277 men aged 27 to 52 who gave blood from 1974 to 1984, with a median follow-up of 27 years [18].
Sources: [17], [18], [5]
Caveats: Observational, from one Swedish city where PSA testing was rare during follow-up, plus a US cohort described in a guideline. The cut points are medians and upper fractions of a distribution, not reference limits.

Red cell distribution width (RDW)Moderate

Death from any cause

Linked to more deaths at higher RDW in community cohorts, about 14% more per 1 point in pooled data on 11,827 older adults.

The numbers and sources

Effect: In a pooled analysis of individual data from 7 community studies of older adults, each 1 point higher RDW went with a 14% higher risk of death (adjusted hazard ratio 1.14, 95% CI 1.11 to 1.17). Hazard ratios were 1.15 for cardiovascular death, 1.13 for cancer death and 1.13 for other causes. Among 1,603 people without major age-associated diseases it was 1.32 (1.21 to 1.44) [6]. In a US national sample aged 45 and older, the top fifth of RDW had a hazard ratio of 2.1 (1.7 to 2.6) against the bottom fifth, and each 1 point went with 22% more deaths (1.22, 1.15 to 1.30) [5]. In UK Biobank volunteers free of major disease and anemia, an RDW of 15% or above had a hazard ratio for death of 3.10 (2.57 to 3.74) against below 12.5% [7].
Population: 11,827 older adults over 68,822 person-years [6]. 8,175 community-dwelling US adults aged 45 and older, followed through 2000 [5]. 240,477 volunteers aged 40 to 70, followed up to 9 years [7].
Sources: [6], [5], [7]
Caveats: Observational. Median RDW differed between studies, from 13.2% to 14.6% [6]. The sources do not test whether changing RDW changes the risk of death, and the pooled analysis describes the biology behind the link as not established [6].

Sex hormone binding globulin (SHBG)Moderate

Death from any cause in men

Linked to fewer deaths in men at lower levels: 15% lower mortality in the bottom fifth of SHBG than the top fifth in pooled cohorts.

The numbers and sources

Effect: In pooled individual data from cohorts of community-dwelling men, the bottom fifth of SHBG (median 20.6 nmol/L) had lower mortality from any cause than the top fifth (median 68.3 nmol/L), with an adjusted hazard ratio of 0.85 (95% CI 0.77 to 0.95). For cardiovascular death it was 0.81 (0.65 to 1.00) [10].
Population: 9 cohorts with individual data covering 255,830 participant-years, plus 11 studies with summary estimates [10].
Sources: [10]
Caveats: Observational, with differences between studies and imputed missing data. The direction runs opposite to the diabetes findings, where lower SHBG went with more disease [10][4].

Sodium levelModerate

Death from any cause at low sodium

Linked to more deaths at low sodium, about 2.6 times the risk across 81 pooled studies, with risk rising as sodium fell.

The numbers and sources

Effect: In a meta-analysis of 81 studies, people with hyponatremia had a higher risk of death from any cause than people without it (RR 2.60, 95% CI 2.31 to 2.93). The association held in heart attack (RR 2.83), heart failure (2.47), cirrhosis (3.34), lung infections (2.49) and hospitalized patients (2.48). People who died had a mean sodium 4.8 mmol/L lower than survivors (130.1 against 134.9 mmol/L) [8]. In US veterans with chronic kidney disease, the adjusted hazard ratio for death was 1.93 (95% CI 1.83 to 2.03) below 130 mEq/L and 1.28 (1.26 to 1.30) at 130 to 135.9 mEq/L, against 136 to 145 mEq/L [9]. On admission to intensive care, the adjusted odds of death in hospital were 1.32 (1.25 to 1.39) at 130 to 134 mmol/L, 1.89 (1.71 to 2.09) at 125 to 129 and 1.81 (1.56 to 2.10) below 125 [6].
Population: 850,222 patients in 81 studies, 17.4% of them hyponatremic [8]. 655,493 US veterans with chronic kidney disease not on dialysis, followed for a median of 5.5 years [9]. 151,486 adults admitted to 77 Austrian intensive care units from 1998 to 2007 [6].
Sources: [8], [9], [6]
Caveats: Observational, and most of the people studied were already ill. The authors of a later trial write that it remains unclear whether hyponatremia itself increases illness and death or is an indicator of the severity of the underlying disease [10].

Sodium levelModerate

Death from any cause at high sodium

Linked to more deaths at high sodium in intensive care and in chronic kidney disease, with the odds reaching 3.6 times above 155 mEq/L.

The numbers and sources

Effect: On admission to intensive care, the adjusted odds of death in hospital were 1.48 (95% CI 1.36 to 1.61) at 146 to 150 mmol/L, 2.32 (1.98 to 2.73) at 151 to 155 and 3.64 (2.88 to 4.61) above 155, against 135 to 145 mmol/L [6]. In US veterans with chronic kidney disease, the adjusted hazard ratio for death was 1.33 (1.28 to 1.38) at 145.1 to 150 mEq/L and 1.56 (1.33 to 1.83) at 150 mEq/L or above, against 136 to 145 mEq/L. Mortality was lowest at 140 mEq/L [9].
Population: 151,486 adults admitted to Austrian intensive care units, 6.9% of them with sodium above 145 mmol/L [6]. 655,493 US veterans with chronic kidney disease not on dialysis [9].
Sources: [6], [9]
Caveats: Observational, in people who were critically ill or had kidney disease. These studies cannot say whether high sodium caused the deaths or marked the illness behind it.

Transferrin saturationModerate

Death from any cause at high levels

Linked to more deaths at a saturation of 50% or above, about 40% more than below 50% in two Danish population studies.

The numbers and sources

Effect: In two Danish general population studies combined, a transferrin saturation of 50% or above had an adjusted hazard ratio for death of 1.4 (95% CI 1.2 to 1.6) against below 50%, 1.3 in men and 1.5 in women. Risk rose stepwise, and 80% or above against below 20% had a hazard ratio of 2.2 (1.4 to 3.3). Adding a third population study, the pooled odds ratio for death at 50% or above was 1.3 (1.2 to 1.5) [10]. In a US national cohort, a saturation above 55% had a hazard ratio of 1.60 (1.17 to 2.21) against lower values. No death certificate listed hemochromatosis, and people above 55% were more likely to have died of cirrhosis or diabetes [11].
Population: 45,159 Danish adults, 4,568 deaths over up to 18 years [10]. 10,714 US adults aged 25 to 74 at baseline, followed from the early 1970s to 1992 [11].
Sources: [10], [11]
Caveats: Observational. The share of all deaths in the Danish studies attributable to a saturation of 50% or above was 0.8% [10]. Neither source shows that lowering saturation changes the risk of death.

TriglyceridesModerate

Death from any cause

Linked to more deaths at higher levels: 9% more at 150 to 199 mg/dL and 20% more at 200 mg/dL or above in a meta-analysis of cohorts.

The numbers and sources

Effect: Against triglycerides of 90 to 149 mg/dL, pooled risk ratios for death from any cause were 0.94 (95% CI 0.85 to 1.03) below 90 mg/dL, 1.09 (1.02 to 1.17) at 150 to 199 mg/dL and 1.20 (1.04 to 1.38) at 200 mg/dL or above. For cardiovascular death they were 0.83, 1.15 and 1.25. Risk of death from any cause rose 12% per 1 mmol/L higher triglycerides [22]. In a Danish cohort, adjusted hazard ratios for death at non-fasting triglycerides of 5 mmol/L or more, against below 1 mmol/L, were 3.3 in women and 1.8 in men [12].
Population: 61 prospective studies, with 58,419 deaths from any cause among 330,566 participants [22]. 13,981 adults in Copenhagen followed a mean 26 years [12].
Sources: [22], [12]
Caveats: Observational. The studies cannot separate triglycerides from the conditions that raise them.

Vitamin A (retinol) levelModerateStudied in children

Death in young children where deficiency is common

Shown in trials to cut deaths 12% in children aged 6 months to 5 years given vitamin A in countries where deficiency is common.

The numbers and sources

Effect: In a Cochrane review, vitamin A supplementation reduced deaths from any cause (RR 0.88, 95% CI 0.83 to 0.93, 19 trials, 1,202,382 children, high-certainty evidence) and deaths from diarrhea (RR 0.88, 0.79 to 0.98). It reduced night blindness (RR 0.32, 0.21 to 0.50), Bitot's spots (RR 0.42, 0.33 to 0.53) and vitamin A deficiency itself (RR 0.71, 0.65 to 0.78, 4 studies, 2,262 children). Deaths from measles and from respiratory disease did not differ [8].
Population: 47 studies with about 1,223,856 children in 19 countries, 30 of the studies in Asia. The mean age was about 33 months [8].
Sources: [8]
Caveats: The trials gave vitamin A to children in the community whatever their own serum retinol, so they show what supplementation did in populations with widespread deficiency. They did not test treating a child according to a measured level, and they do not describe adults or populations where deficiency is rare [8].

White blood cell count (WBC)Moderate

Death from any cause

Linked to more deaths at higher counts, about 10% more for each 1 thousand/µL in a meta-analysis of 13 studies.

The numbers and sources

Effect: Each 1.0 thousand/µL higher white cell count had a pooled hazard ratio for death of 1.10 (95% CI 1.08 to 1.13). It was 1.10 (1.07 to 1.12) in people with cardiovascular disease and 1.12 (1.07 to 1.17) in the general population or people with COVID-19. For long-term death from any cause it was 1.09 (1.07 to 1.12), and for long-term cardiovascular death 1.05 (1.02 to 1.07) [10].
Population: 62,904 participants in 13 studies [10].
Sources: [10]
Caveats: Observational, and the pooled studies mixed patients with heart disease, people with COVID-19 and the general population. The authors state that causality is not established [10].

AlbuminLimited

Death and complications after surgery

One US veterans cohort linked lower albumin before surgery to more deaths within 30 days, from under 1% above 4.6 g/dL to 29% below 2.1 g/dL.

The numbers and sources

Effect: A fall in albumin from above 46 g/L (4.6 g/dL) to below 21 g/L (2.1 g/dL) was associated with an exponential rise in 30-day mortality from less than 1% to 29% and in complications from 10% to 65%. Among 62 preoperative variables, albumin was the strongest predictor of death and complications, and it predicted sepsis and major infections more closely than other complications [9].
Population: 54,215 major noncardiac surgery cases at 44 Veterans Affairs medical centers [9].
Sources: [9]
Caveats: Observational, one cohort, published in 1999. It does not show that raising albumin before surgery changes the outcome.

Apolipoprotein B (ApoB)Limited

Death and heart attack when ApoB and LDL cholesterol disagree

One cohort of statin-treated adults linked high ApoB with low LDL cholesterol to 21% more deaths and 49% more heart attacks than when both were low.

The numbers and sources

Effect: Compared with both values below the median, ApoB above the median with LDL cholesterol below it had hazard ratios of 1.21 for death from any cause (95% CI 1.07 to 1.36) and 1.49 for heart attack (95% CI 1.15 to 1.92). High LDL cholesterol with low ApoB was not associated with a higher risk of either outcome [16].
Population: 13,015 statin-treated adults in the Copenhagen General Population Study, median follow-up 8 years [16].
Sources: [16]
Caveats: Observational, in one Danish cohort. High and low were defined by the cohort's own medians, which the abstract does not report [16].

Bilirubin (total)Limited

Lung disease and death from any cause

One UK cohort of 504,206 adults linked higher bilirubin within the usual range to less lung cancer, less chronic obstructive lung disease and fewer deaths.

The numbers and sources

Effect: For each 0.1 mg/dL higher bilirubin, the adjusted rate of lung cancer was 8% lower in men (95% CI 5% to 11%) and 11% lower in women (7% to 14%). In men the rate of chronic obstructive pulmonary disease was 6% lower (5% to 7%) and the death rate 3% lower (2% to 3%), with very similar results in women. Median bilirubin was 0.64 mg/dL in men and 0.53 mg/dL in women [8].
Population: 504,206 adults in a UK primary care database with no evidence of liver, bile duct or hemolytic disease, with 1,341 lung cancers, 5,863 cases of chronic obstructive pulmonary disease and 23,103 deaths [8].
Sources: [8]
Caveats: One cohort, observational, in people whose bilirubin was within the normal range [8]. Bilirubin was lower in active smokers in US survey data, so how fully smoking was accounted for matters to this result [5].

Bilirubin (total)Limited

Death in people with Gilbert syndrome

One UK cohort found people with Gilbert syndrome had about half the death rate of people with normal bilirubin.

The numbers and sources

Effect: Mortality was 24 per 10,000 person-years in people with a diagnosis of Gilbert syndrome and a raised bilirubin, and 50 per 10,000 in a comparison cohort with normal bilirubin. After accounting for sociodemographic factors and general health indicators, the mortality rate ratio was 0.5 (95% CI 0.4 to 0.7) [10].
Population: 4,266 patients with Gilbert syndrome and 21,968 comparison patients in a UK primary care database, with 1,174 deaths over 350,000 person-years [10].
Sources: [10]
Caveats: One cohort, observational. It counted people whose Gilbert syndrome had been diagnosed and recorded, who may differ from people with the same trait and no diagnosis.

Blood urea nitrogen (BUN)Limited

Death from any cause in the general population

One US cohort found 48% more deaths from any cause in the highest quarter of BUN than in the lowest over about 12 years.

The numbers and sources

Effect: In 17,719 US adults followed for a mean of 11.65 years, people in the highest quarter of BUN (5.71 mmol/L or above, about 16 mg/dL) had more deaths from any cause than those in the lowest quarter (below 3.57 mmol/L, about 10 mg/dL), with an adjusted hazard ratio of 1.48 (95% CI 1.28 to 1.72). The hazard ratio for cardiovascular death was 1.48 (1.08 to 2.02). In subgroups the trend held most clearly in women [5].
Population: 17,719 adults in the US National Health and Nutrition Examination Survey, 1999 to 2006, with 3,628 deaths, 859 from cardiovascular disease [5].
Sources: [5]
Caveats: Observational, one cohort, one BUN measurement per person. The models adjusted for demographics, lifestyle, hypertension and diabetes. The abstract does not list kidney function among the adjustments, and the test is used as a marker of kidney disease [2][5].

C-peptideLimited

Death in adults without diabetes

Limited evidence: two analyses of one US survey linked the highest quarter of fasting C-peptide to about 70% to 80% more deaths in adults without diabetes.

The numbers and sources

Effect: In 5,153 adults aged 40 to 74 without diabetes, the highest quarter of fasting C-peptide, 0.984 nmol/L (about 2.97 ng/mL) or above, had an adjusted hazard ratio of 1.72 (95% CI 1.34 to 2.21) for death from any cause and 1.60 (1.07 to 2.39) for cardiovascular death against the lowest quarter, 0.418 nmol/L (about 1.26 ng/mL) or below. C-peptide predicted these deaths better than fasting glucose, insulin or indices built from them [7]. In 5,902 adults aged 40 and older from the same survey, the highest quarter had a hazard ratio of 1.80 (1.33 to 2.43) for death from any cause and 3.20 (2.07 to 4.93) for cardiovascular death against the lowest [8].
Population: 5,153 adults without diabetes in the third US National Health and Nutrition Examination Survey of 1988 to 1994, followed for a median of 14.4 years with 513 deaths [7]. 5,902 adults with no history of diabetes in the same survey, mean C-peptide 0.78 nmol/L [8].
Sources: [7], [8]
Caveats: Observational, and both analyses draw on the same survey, so they are not independent. The authors read a high C-peptide in a person without diabetes as a sign of insulin resistance. Neither source tested whether lowering it changes the risk of death [7][8].

Erythrocyte sedimentation rate (ESR)Limited

Death from any cause

Limited evidence: one Dutch cohort of 5,226 older adults linked an ESR of 20 to 50 to 23% more deaths, and above 50 to 89% more.

The numbers and sources

Effect: Against an ESR below 20 mm/hr, the adjusted hazard ratio for death from any cause was 1.23 (95% CI 1.12 to 1.35) at 20 to 50 mm/hr and 1.89 (1.38 to 2.60) above 50 mm/hr. In people older than 75 with no other illness recorded, an ESR above 20 mm/hr still had a hazard ratio of 1.29 (1.01 to 1.64) [7].
Population: 5,226 participants in the Rotterdam Study, mean age 70.3, with 3,749 deaths over a median of 14.9 years [7].
Sources: [7]
Caveats: One observational cohort of older adults. ESR is a nonspecific sign of illness, and the source did not test whether lowering it changes the risk of death [7].

Gamma-glutamyl transferase (GGT)Limited

Cancer

One Swedish cohort of 545,460 people found cancer risk rose across GGT categories, 32% higher at 72 U/L or above than below 18 U/L.

The numbers and sources

Effect: In a Swedish cohort with 37,809 primary cancers, hazard ratios for any cancer against GGT below 18 U/L were 1.07 (95% CI 1.04 to 1.09) at 18 to 36 U/L, 1.18 (1.14 to 1.22) at 36 to 72 U/L and 1.32 (1.26 to 1.38) at 72 U/L or above. Associations with prostate, breast and liver cancer were stronger in people with glucose of 6.11 mmol/L or above [12].
Population: 545,460 people older than 20 in the Swedish Apolipoprotein Mortality Risk study [12].
Sources: [12]
Caveats: Observational, one cohort. The abstract does not say whether the analysis adjusted for alcohol intake or smoking [12].

HDL cholesterol (HDL-C)Limited

Death from any cause at very low and very high levels

Two Danish cohorts linked both low and extremely high HDL cholesterol to more deaths, with the lowest risk at 73 mg/dL in men and 93 mg/dL in women.

The numbers and sources

Effect: The association with death from any cause was U-shaped in both sexes. The level with the lowest risk was 73 mg/dL (95% CI 54 to 77) in men and 93 mg/dL (69 to 97) in women. Against the lowest-risk groups, adjusted hazard ratios were 1.36 (95% CI 1.09 to 1.70) for men at 97 to 115 mg/dL and 2.06 (1.44 to 2.95) at 116 mg/dL or above. For women they were 1.10 (0.83 to 1.46) at 116 to 134 mg/dL and 1.68 (1.09 to 2.58) at 135 mg/dL or above [18]. A European guideline states that above about 90 mg/dL there appears to be an increased risk of atherosclerotic cardiovascular disease [4].
Population: 52,268 men and 64,240 women in two Copenhagen population studies, 745,452 person-years, 10,678 deaths [18].
Sources: [18], [4]
Caveats: Observational, from two cohorts in one Danish city, and the authors state the findings need confirmation in other studies [18]. The study cannot show that very high HDL cholesterol itself shortens life.

Luteinizing hormone (LH)LimitedStudied in men

Death from any cause in men

One cohort found men in the highest quarter of LH had 32% higher mortality from any cause than men in the lowest quarter over up to 30 years.

The numbers and sources

Effect: Men in the highest LH quartile had higher mortality from any cause than men in the lowest (hazard ratio 1.32, 95% CI 1.14 to 1.53), and higher LH went with more cancer deaths (1.42, 1.10 to 1.84). A higher ratio of LH to testosterone also went with more deaths (1.23, 1.06 to 1.43). Testosterone itself was not associated with death from any cause [7].
Population: 5,350 men from the general population aged 30 to 70 at baseline, followed for up to 30 years, with 1,533 deaths [7].
Sources: [7]
Caveats: Observational, from one cohort. The abstract does not print the LH level that marks the top quartile. The authors read the result as a sign of impaired testicular function, so LH may mark illness and not cause it [7].

ProlactinLimited

Death from any cause

One cohort found higher prolactin went with more deaths, 17% more per standard deviation in men and 22% more in women.

The numbers and sources

Effect: Each standard deviation higher prolactin had a hazard ratio for death from any cause of 1.17 in men (95% CI 1.07 to 1.29) and 1.22 in women (1.03 to 1.46). The top third against the bottom third had 1.75 in men (1.32 to 2.32) and 1.66 in women (1.08 to 2.56). Cardiovascular deaths followed the same pattern in both sexes [8].
Population: 3,929 adults aged 20 to 81 in a German population study, 419 deaths over a median 10.1 years [8].
Sources: [8]
Caveats: One observational cohort, described by its authors as the first report of this association [8].

Red cell folateLimited

Cardiovascular death in adults with hypertension

Limited evidence: two US survey analyses linked the highest red cell folate levels to more cardiovascular deaths in adults with hypertension.

The numbers and sources

Effect: In adults with hypertension surveyed from 1991 to 1994, the high folate quartile had a hazard ratio for cardiovascular death of 1.44 (95% CI 1.31 to 1.58) against the intermediate quartiles and 2.13 (1.77 to 2.57) for death from heart attack. The low quartile did not differ (1.09, 0.94 to 1.27) [9]. In adults with hypertension surveyed from 1999 to 2014, the highest quartile against the second had hazard ratios of 1.68 (1.30 to 2.16) for cardiovascular death and 1.30 (1.16 to 1.46) for death from any cause, and the lowest quartile was not associated with either. The associations were non-linear, with inflection points at 819.7 and 760.1 ng/mL [10].
Population: 2,986 US adults with hypertension, 579 cardiovascular deaths over an average 11.7 years [9]. 13,986 US adults with hypertension, 548 cardiovascular deaths over a median 7.0 years [10].
Sources: [9], [10]
Caveats: Observational, and limited to people with hypertension. Supplement use and illness may differ between people with high and low levels. No trial cited here tested the outcome of lowering a high red cell folate [9][10].

Selenium levelLimited

Death from any cause

Limited evidence: one US cohort linked serum selenium to fewer deaths up to about 130 µg/L and more above 150, and one trial of 300 µg a day raised deaths.

The numbers and sources

Effect: In US adults, the top third of serum selenium (130.39 µg/L or above) had fewer deaths from any cause than the bottom third (below 117.31 µg/L), with a hazard ratio of 0.83 (95% CI 0.72 to 0.96), and fewer cancer deaths (0.69, 0.53 to 0.90). The relation was not a straight line: deaths fell as selenium rose to 130 µg/L and rose modestly above 150 µg/L. Cardiovascular deaths showed no association [11]. In a Danish trial in older adults with relatively low selenium, 300 µg a day for 5 years was followed by more deaths from any cause over the full 15 years of follow-up than placebo (HR 1.59, 95% CI 1.02 to 2.46). Doses of 100 and 200 µg a day showed no significant difference [7].
Population: 13,887 adults in the US national survey of 1988 to 1994, followed for up to 12 years [11]. 491 volunteers aged 60 to 74 in Denmark, with 158 deaths [7].
Sources: [11], [7]
Caveats: The cohort is observational, and mean serum selenium in it was 125.6 µg/L [11]. The trial was not designed to study deaths, its sample was small, and the excess appeared only at the highest dose [7].

Serum ironLimited

Death from any cause

Limited evidence: one Taiwanese cohort linked both low and high serum iron to more deaths over 19 years, 27% and 37% more than the middle group.

The numbers and sources

Effect: The relation between serum iron and death from any cause, and from specific causes, was J-shaped. Against the group classed as normal, the adjusted hazard ratio for death was 1.27 (95% CI 1.23 to 1.31) in the low group and 1.37 (1.30 to 1.44) in the high group. People whose serum iron was outside the normal group at two visits had a hazard ratio of 1.49 (1.36 to 1.63) against people inside it at both [6].
Population: Adults in the Taiwan MJ health screening cohort, recruited 1997 to 2007 and followed to 2022, 33,005 deaths over a median of 19.0 years [6].
Sources: [6]
Caveats: One observational cohort. The abstract does not give the number of participants or the serum iron levels that defined the low and high groups. Illness that changes serum iron before death cannot be ruled out [6].

Sodium levelLimited

Sodium at the upper end of the range in midlife

One US cohort linked midlife sodium above 142 mEq/L to 39% more chronic disease, and above 144 mEq/L to 21% more early deaths, over 25 years.

The numbers and sources

Effect: Sodium above 142 mmol/L in middle age was associated with a higher risk of developing chronic diseases (HR 1.39, 95% CI 1.18 to 1.63) and with higher odds of a biological age above chronological age (OR 1.50, 1.14 to 1.96). Sodium above 144 mmol/L was associated with a higher risk of premature death (HR 1.21, 1.02 to 1.45) [14].
Population: 15,752 adults aged 45 to 66 at enrollment in the Atherosclerosis Risk in Communities study, followed for 25 years [14].
Sources: [14]
Caveats: One cohort, observational. The authors used serum sodium as a proxy for hydration habits and state that intervention studies are needed to confirm a link between hydration and aging. These values sit inside both laboratory reference ranges at the top of this page [14].

Uric acid (serum urate)Limited

Death from any cause

One Korean cohort of 375,163 adults found more deaths at both ends of the uric acid range than in the middle.

The numbers and sources

Effect: In Korean adults having health checkups, the association with death from any cause was U-shaped. Against the sex-specific reference category, adjusted hazard ratios in the lowest categories (below 3.5 mg/dL for men, below 2.5 mg/dL for women) were 1.58 (95% CI 1.18 to 2.10) and 1.80 (1.10 to 2.93). In the highest categories (9.5 mg/dL or above for men, 8.5 mg/dL or above for women) they were 2.39 (1.57 to 3.66) and 3.77 (1.17 to 12.17). Cardiovascular and cancer deaths followed the same pattern [21].
Population: 375,163 South Korean men and women, with 2,020 deaths over about 2.06 million person-years [21].
Sources: [21]
Caveats: Observational, one cohort, with few deaths in the extreme categories and wide intervals for women. Illness that lowers or raises uric acid before death cannot be ruled out as the explanation [21].

Vitamin A (retinol) levelLimitedStudied in men

Death from any cause in adults

Limited evidence: one cohort of 29,104 Finnish men who smoked linked higher serum retinol to 17% fewer deaths, with the lowest risk near 60 to 70 µg/dL.

The numbers and sources

Effect: Over 30 years, men in the highest fifth of serum retinol had fewer deaths from any cause than men in the lowest fifth (HR 0.83, 95% CI 0.80 to 0.87), and fewer deaths from cardiovascular and respiratory disease. The relation was not a straight line. Deaths were more frequent below 483 µg/L (48.3 µg/dL), and risk was lowest at about 600 to 700 µg/L (60 to 70 µg/dL). The median at the start was 576 µg/L (57.6 µg/dL). The authors cite a US survey analysis of 16,008 people in which the middle fifths, and not the top fifth, had 18% fewer deaths than the lowest fifth [3].
Population: 29,104 male smokers in Finland aged 50 to 69 at entry to a cancer prevention trial, with 23,797 deaths [3].
Sources: [3]
Caveats: Observational, in men who smoked, almost none of whom had retinol in the range WHO calls low. Illness and inflammation lower serum retinol [9], so a low level may mark disease already present. No trial that raised serum retinol in adults and counted deaths was located.

Vitamin B12 levelLimited

Death from any cause at high levels

Limited evidence: one Dutch cohort linked higher plasma B12 to more deaths over 8 years, 25% more per standard deviation.

The numbers and sources

Effect: Mortality was 33.8 deaths per 10,000 person-years in the quartile with the lowest plasma B12 and 65.7 per 10,000 person-years in the highest. After adjustment for age, sex, kidney function and other clinical and laboratory variables, the hazard ratio per 1 standard deviation higher B12 was 1.25 (95% CI 1.06 to 1.47). The median concentration was 394 pg/mL (interquartile range 310 to 497) [9].
Population: 5,571 adults in Groningen, the Netherlands, mean age 53.5, not using B12 supplements, 226 deaths over a median 8.2 years [9].
Sources: [9]
Caveats: Observational and a post hoc analysis. The authors state that the mechanisms are not established, so a high level may mark an illness and need not be its cause [9].

Hemoglobin A1c (HbA1c)Strong

Heart attack and death in trials that lowered HbA1c in type 2 diabetes

Mixed results: pooled trials of intensive glucose lowering found 9% fewer major cardiovascular events and no fewer deaths, and one trial aiming below 6.0% found more deaths.

The numbers and sources

Effect: A prospectively planned meta-analysis found that more intensive glucose control reduced major cardiovascular events by 9% (HR 0.91, 95% CI 0.84 to 0.99) and heart attack by 15% (HR 0.85, 95% CI 0.76 to 0.94). Death from any cause (HR 1.04, 95% CI 0.90 to 1.20) and cardiovascular death (HR 1.10, 95% CI 0.84 to 1.42) were not reduced, and major hypoglycemia rose (HR 2.48, 95% CI 1.91 to 3.21) [15]. In one of the pooled trials, a strategy targeting HbA1c below 6.0% reached a median of 6.4% against 7.5% with standard therapy. Deaths were higher in the intensive group (257 against 203, HR 1.22, 95% CI 1.01 to 1.46), and the intensive strategy was stopped after a mean 3.5 years [16].
Population: 27,049 participants with type 2 diabetes and 2,370 major vascular events over 4.4 years [15]. 10,251 patients, mean age 62, with cardiovascular disease or risk factors and a median HbA1c of 8.1% [16].
Sources: [15], [16]
Caveats: A marker that tracks risk is not always one whose lowering reduces it. These trials used the drugs and insulin regimens of their time in older, high-risk people with long-standing diabetes, and do not describe people without diabetes.

IGF-1Strong

Body composition and adverse events in trials of growth hormone

Shown in trials of growth hormone to shift fat and lean mass by about 2 kg in older adults, with more swelling, joint pain and carpal tunnel syndrome.

The numbers and sources

Effect: IGF-1 manages the effects of growth hormone in the body [1], and these trials gave growth hormone without aiming at an IGF-1 level. In community-dwelling adults with a mean age of 69 who were not being treated for a specific illness, growth hormone lowered fat mass by 2.1 kg (95% CI 1.35 to 2.8) and raised lean mass by 2.1 kg (1.3 to 2.9), with no change in weight, bone density or most blood lipids. Treated people were more likely to have soft tissue swelling, joint pain, carpal tunnel syndrome and breast growth in men, and somewhat more likely to develop diabetes or impaired fasting glucose [10]. In adults with confirmed growth hormone deficiency, it lowered weight by 2.31 kg (1.96 to 2.66) and body fat by 2.56 kg (2.16 to 2.97) and raised lean mass by 1.38 kg (1.10 to 1.65). Swelling had a relative risk of 6.07 (4.34 to 8.48) and joint stiffness 4.17 (1.4 to 12.38), with no significant change in bone density. Quality of life improved in 11 of 16 trials [11].
Population: 220 treated participants in 18 study populations, mean treatment 27 weeks [10]. More than 3,400 patients in 54 randomized trials [11].
Sources: [10], [11], [1]
Caveats: A marker that tracks an outcome is not always one whose correction changes it. Low IGF-1 went with more deaths in cohorts, and these trials measured body composition, not deaths. The trials in older adults were small and short [10].

Prostate-specific antigen (PSA)StrongStudied in men

Death from prostate cancer in trials of screening

Mixed results: an invitation to screening cut prostate cancer deaths by 20% in the European trial and 8% in a UK trial, with no significant change in the US trial.

The numbers and sources

Effect: In the European trial, the rate ratio for death from prostate cancer at 16 years was 0.80 (95% CI 0.72 to 0.89), an absolute difference of 0.18%. To prevent one death, 570 men had to be invited and 18 diagnosed [9]. In the US trial the rate ratio was 1.04 (0.87 to 1.24) at a median of 15 years and 0.93 (0.81 to 1.08) at about 17 years [10][11]. In the UK trial of a single invitation, it was 0.96 (0.85 to 1.08) at 10 years and 0.92 (0.85 to 0.99) at 15 years, when 0.69% of invited men and 0.78% of control men had died of prostate cancer [12][8]. A meta-analysis of five trials found screening probably has no effect on death from any cause (incidence rate ratio 0.99, 0.98 to 1.01) and may have no effect on prostate cancer death (0.96, 0.85 to 1.08), or a small effect in the trial at lower risk of bias (0.79, 0.69 to 0.91), about one fewer prostate cancer death per 1,000 men screened over 10 years [13]. The US Preventive Services Task Force estimates that screening men aged 55 to 69 may prevent about 1.3 prostate cancer deaths per 1,000 men over about 13 years [14].
Population: 162,389 men aged 55 to 69 in eight European countries [9]. 76,683 US men aged 55 to 74 [10][11]. 415,357 UK men aged 50 to 69 at 573 primary care practices [12][8]. Five trials with 721,718 men [13].
Sources: [9], [10], [11], [12], [8], [13], [15], [14], [5]
Caveats: The trials tested an invitation to screening followed by biopsy and treatment, not a PSA level. They differed in ways that bear on the results. In the US trial an estimated 86% of control men had some PSA testing, its biopsy cut-off was 4 ng/mL against 3 in the European trial, and 34% of men with a raised result were biopsied against over 90% [10][5]. In the UK trial 36% of invited men had the single test [12]. Deaths from any cause did not differ in the US trial, the UK trial or the pooled analysis [10][8][13].

25-hydroxyvitamin DModerate

Death from cancer

Mixed results: pooled trials found 16% fewer cancer deaths with vitamin D supplements, while standardized cohort data found no clear link between the blood level and cancer death.

The numbers and sources

Effect: Across 52 randomized trials, vitamin D supplementation reduced the risk of death from cancer (RR 0.84, 95% CI 0.74 to 0.95) [10]. In a trial of 2,000 IU a day, invasive cancer did not differ from placebo (HR 0.96, 95% CI 0.88 to 1.06) and the hazard ratio for death from cancer was 0.83 (95% CI 0.67 to 1.02) [11]. Across cohort studies, the bottom third of baseline levels had a relative risk of death from cancer of 1.14 against the top third (95% CI 1.01 to 1.29) [9]. Pooled individual data with standardized measurements found no significant linear association between the level and cancer death [8].
Population: 75,454 participants in 52 trials [10]. 25,871 US adults [11]. 73 cohort studies [9]. 26,916 participants in 8 European studies [8].
Sources: [10], [11], [9], [8]
Caveats: Cancer death was a secondary outcome in the trials. The largest single trial found no effect on being diagnosed with cancer [11].

Alanine aminotransferase (ALT)Moderate

Death from any cause

Mixed results: pooled cohorts found fewer deaths at higher ALT in North America and more in Asia, and a US survey found no association.

The numbers and sources

Effect: In a meta-analysis of 19 cohort studies, comparing the top and bottom thirds of baseline ALT gave a relative risk of death from any cause of 0.82 (95% CI 0.78 to 0.86) in North American populations and 1.43 (1.08 to 1.90) in Asian populations [14]. In US adults without viral hepatitis, ALT above 30 U/L in men or 19 U/L in women was not associated with death from any cause (HR 1.2, 95% CI 0.88 to 1.6) [11].
Population: Over 9.24 million participants and 242,953 deaths in 19 cohorts, which also covered other liver enzymes [14]. 14,950 US adults followed for 12 years [11].
Sources: [14], [11]
Caveats: Observational. The authors of the meta-analysis could not explain the difference between regions and called for further study [14]. These studies cannot say why a higher ALT went with fewer deaths in the North American cohorts.

Fasting insulin and HOMA-IRModerate

Death from any cause

Mixed results: a meta-analysis of adults without diabetes tied the highest HOMA-IR group to 34% more deaths, with a weaker, borderline link for fasting insulin.

The numbers and sources

Effect: Comparing the highest with the lowest category, the pooled adjusted risk ratio for death from any cause was 1.13 (95% CI 1.00 to 1.27, P = 0.058) for fasting insulin and 1.34 (1.11 to 1.62) for HOMA-IR. For cardiovascular death it was 2.11 (1.01 to 4.41) for HOMA-IR in two studies and 1.40 (0.49 to 3.96) for fasting insulin in one [14].
Population: 26,976 adults without diabetes in 7 articles [14].
Sources: [14]
Caveats: Observational. The authors call the findings for cardiovascular death unreliable because so few studies were included [14].

FerritinModerate

Death from any cause at high levels

Mixed results: one Danish cohort linked ferritin of 600 ng/mL or more to 50% more deaths, while a pooled analysis of population studies found no link.

The numbers and sources

Effect: In a Danish general population cohort, the adjusted hazard ratio for death was 1.1 (95% CI 1.1 to 1.2) for ferritin of 200 ng/mL or more against less than 200, and 1.5 (95% CI 1.2 to 1.8) for 600 or more. Median survival was 55 years at 600 or more, 72 years at 400 to 599, 76 years at 200 to 399 and 79 years below 200. For 600 or more, hazard ratios were 1.6 for cancer death and 1.5 for cardiovascular death. The same paper's meta-analysis of population studies comparing the top with the reference quartile or third found an odds ratio for death of 1.0 (95% CI 0.9 to 1.1) [9].
Population: 8,988 adults in the Copenhagen City Heart Study, 6,364 deaths over a median 23 years [9].
Sources: [9]
Caveats: Observational. Ferritin rises with infection and inflammation [2][3], so a high level may mark disease already present and need not reflect iron itself.

HOMA-IRModerate

Death from any cause

Mixed results: one meta-analysis tied the highest HOMA-IR group to 34% more deaths from any cause, while another found no link with fatal cardiovascular events or cancer deaths.

The numbers and sources

Effect: In adults without diabetes, the pooled adjusted risk ratio for death from any cause comparing the highest with the lowest category was 1.34 (95% CI 1.11 to 1.62) for HOMA-IR and 1.13 (1.00 to 1.27, P = 0.058) for fasting insulin. For cardiovascular death it was 2.11 (1.01 to 4.41) for HOMA-IR in two studies [15]. A later meta-analysis found no association of a higher HOMA-IR with fatal major cardiovascular events or with cancer mortality [12].
Population: 26,976 adults without diabetes in 7 articles [15]. 215,878 participants in 38 cohort studies across all outcomes [12].
Sources: [15], [12]
Caveats: Observational. The authors of the first analysis call the findings for cardiovascular death unreliable because so few studies were included [15].

LDL cholesterol (LDL-C)Moderate

Death from any cause

Mixed results: one Danish cohort found more deaths at both low and high LDL cholesterol, while statin trials found 10% fewer deaths per 1 mmol/L lowered.

The numbers and sources

Effect: In 108,243 Danish adults, the association with death from any cause was U-shaped. Compared with 132 to 154 mg/dL, the adjusted hazard ratio was 1.25 (95% CI 1.15 to 1.36) below 70 mg/dL and 1.15 (1.05 to 1.27) above 189 mg/dL. The level with the lowest risk was 140 mg/dL overall and 89 mg/dL in people on lipid-lowering treatment. The pattern held for cancer and other deaths and not for cardiovascular deaths, and any increase in LDL cholesterol went with more heart attacks [15]. Across 26 statin trials, death from any cause fell 10% per 1.0 mmol/L reduction (RR 0.90, 95% CI 0.87 to 0.93), with no significant effect on deaths from cancer or other non-vascular causes [12].
Population: 108,243 people aged 20 to 100 in the Copenhagen General Population Study, median follow-up 9.4 years, 11,376 deaths [15]. About 170,000 participants in 26 statin trials [12].
Sources: [15], [12]
Caveats: The cohort finding is observational, and illness that lowers LDL cholesterol before death (reverse causation) cannot be ruled out. The trials tested lowering with a statin in people selected for trials, over about 5 years.

Non-HDL cholesterol (non-HDL-C)Moderate

Death from any cause

Mixed results: higher non-HDL cholesterol went with more deaths in people with heart disease or on statins, while pooled general-population studies found no clear link with cardiovascular death.

The numbers and sources

Effect: In statin-treated Danish adults, non-HDL cholesterol above the median with LDL cholesterol below it had hazard ratios of 1.18 for death from any cause (95% CI 1.02 to 1.36) and 1.78 for heart attack (1.35 to 2.34), compared with both below the median. ApoB above the median with non-HDL cholesterol below it had a hazard ratio for death of 1.21 (1.03 to 1.41) [16]. In a meta-analysis of people with coronary heart disease, high baseline non-HDL cholesterol went with a risk ratio for death of 1.24 (95% CI 1.05 to 1.46) [17]. In a meta-analysis of observational studies, elevated non-HDL cholesterol was not associated with cardiovascular death in the general population (RR 1.64, 95% CI 0.96 to 2.80) or in people with type 2 diabetes (RR 1.08, 0.57 to 2.07), while it was associated with cardiovascular disease in both (RR 1.59 and 1.99) [18].
Population: 13,015 statin-treated adults in the Copenhagen General Population Study, median follow-up 8 years [16]. 11,057 people with coronary heart disease in 6 studies followed 18 to 148 months [17]. 156,381 people in 13 studies [18].
Sources: [16], [17], [18]
Caveats: Observational. High and low were defined by each cohort's own median or categories [16][17]. The confidence intervals for cardiovascular death are wide [18].

Sodium levelModerate

Death and readmission after low sodium is corrected

Mixed results: pooled observational studies linked a rise in low sodium to fewer deaths, while a 2,173-person trial of targeted correction found no fewer deaths or readmissions.

The numbers and sources

Effect: In a meta-analysis of 15 studies, any improvement of hyponatremia was associated with a lower risk of death (OR 0.57, 95% CI 0.40 to 0.81). The association was stronger in the 8 studies that set the threshold for improvement at above 130 mmol/L (OR 0.51, 0.31 to 0.86) [11]. In a randomized trial in hospitalized patients with sodium below 130 mmol/L, a targeted correction program brought sodium to 135 to 145 mmol/L in 60.4% of patients, against 46.2% with routine care. Death or readmission within 30 days occurred in 20.5% and 21.8% (difference -1.3 percentage points, 95% CI -4.9 to 2.2). Death occurred in 8.0% of each group [10].
Population: 13,816 patients in 15 studies [11]. 2,173 hospitalized patients at 9 European centers, median age 73, median sodium 127 mmol/L at entry [10].
Sources: [11], [10]
Caveats: The pooled studies were not randomized, so people whose sodium rose may have differed in other ways from those whose sodium did not [11]. The trial counted outcomes for 30 days and tested one program of correction in hospital. Overcorrection occurred in 2.3% of the intervention group and 1.4% of the control group, and no case of osmotic demyelination syndrome was observed [10].

Total cholesterolModerate

Death from any cause

Mixed results: a Korean cohort of 12.8 million linked both low and high total cholesterol to more deaths, while statin trials found 10% fewer deaths per 1 mmol/L lowered.

The numbers and sources

Effect: In 12.8 million Korean adults, the association with death from any cause was U-shaped. Below 200 mg/dL, each 39 mg/dL higher total cholesterol went with 23% lower mortality (95% CI 23 to 24). From 200 mg/dL upward, each 39 mg/dL higher went with 7% higher mortality (6 to 7). The levels with the lowest mortality were 210 to 249 mg/dL in most age and sex groups, and lower in men aged 18 to 34 (180 to 219 mg/dL) and women aged 18 to 34 (160 to 199 mg/dL) [12]. Across 26 statin trials, death from any cause fell 10% per 1.0 mmol/L reduction in LDL cholesterol (RR 0.90, 95% CI 0.87 to 0.93) [10].
Population: 12,815,006 Korean adults who had routine health examinations in 2001 to 2004, followed until 2013, with 694,423 deaths [12]. About 170,000 participants in 26 statin trials [10].
Sources: [12], [10]
Caveats: The cohort finding is observational, and illness that lowers cholesterol before death (reverse causation) cannot be ruled out. The trials tested lowering with a statin in people selected for trials, over about 5 years.

DHEA sulfate (DHEA-S)Strong

Body composition, physical performance and quality of life in trials that raised DHEA-S

Not shown to change body composition, physical performance, sexual function or quality of life in a meaningful way when DHEA-S was raised with DHEA in older adults.

The numbers and sources

Effect: In a 2-year trial in older adults with low DHEA-S, DHEA raised the level by a median of 3.4 µg/mL in men and 3.8 µg/mL in women, which is 340 and 380 µg/dL. It had no significant effect on body composition, and did not change peak oxygen consumption, muscle strength, insulin sensitivity or quality of life. Bone density rose at the femoral neck in men and at the ultradistal radius in women [8]. Across 25 trials in older men, DHEA reduced fat mass slightly (standardized mean difference -0.35, 95% CI -0.65 to -0.05), an effect that disappeared after adjusting for the rise in testosterone and estradiol, with no effect on blood lipids, blood sugar, bone, sexual function or quality of life [9]. A review of 8 trials called the effect on muscle strength and physical function inconclusive [10]. In postmenopausal women, 23 trials found no significant improvement in libido or sexual function (0.35, -0.02 to 0.73) and no significant effect on lipids, glucose, weight or bone density [11]. A Cochrane review found no improvement in quality of life (0.16, -0.03 to 0.34) and more androgenic side effects, mainly acne (odds ratio 3.77, 1.36 to 10.4) [12].
Population: 87 older men and 57 older women with low DHEA-S, treated for 2 years [8]. 1,353 older men in 25 trials, mean follow-up 36 weeks [9]. 661 adults aged 50 or older in 8 trials [10]. 1,188 postmenopausal women in 23 trials [11]. 1,273 women in 28 trials [12].
Sources: [8], [9], [10], [11], [12]
Caveats: A marker that tracks an outcome is not always one whose correction changes it. Low DHEA-S went with more deaths in cohorts. The trials measured body composition, function and quality of life, not deaths. The reviewers of the trials in women rated confidence in the evidence as low [11].

HomocysteineStrong

Heart attack and death in trials that lowered homocysteine

Not shown to fall when homocysteine is lowered: B vitamins cut homocysteine about 25%, and heart attacks and deaths did not differ from placebo across pooled trials.

The numbers and sources

Effect: In a meta-analysis of individual data from 8 placebo-controlled trials, folic acid lowered homocysteine by an average of 25% and had no significant effect on major vascular events (rate ratio 1.01, 95% CI 0.97 to 1.05) or major coronary events (1.03, 95% CI 0.97 to 1.10), overall or in any subgroup studied. Nor did it change overall cancer incidence (1.05, 95% CI 0.98 to 1.13) or death from any cause (1.02, 95% CI 0.97 to 1.08) [7]. A Cochrane review of 15 trials found no difference from placebo in heart attack (RR 1.02, 95% CI 0.95 to 1.10, 12 trials) or death from any cause (RR 1.01, 95% CI 0.96 to 1.06, 11 trials), and judged that further trials were unlikely to change the finding [8].
Population: 37,485 people at increased cardiovascular risk in 8 trials, median follow-up 5 years, 9,326 major vascular events [7]. 71,422 participants in 15 trials with follow-up of 1 to 7.3 years [8].
Sources: [7], [8]
Caveats: The pooled trials ran for a median of 5 years in people at increased risk of cardiovascular disease, so they say little about longer treatment or lower-risk people [7].

AlbuminModerate

Raising albumin by infusion

Not shown to lower deaths: two trials that raised albumin to 3.0 g/dL or more by infusion, in sepsis and in cirrhosis, found no benefit.

The numbers and sources

Effect: In severe sepsis, albumin infusion to a target of 30 g/L (3.0 g/dL) or more did not change death at 28 days (31.8% against 32.0% with crystalloid alone, RR 1.00, 95% CI 0.87 to 1.14) or at 90 days (41.1% against 43.6%, RR 0.94, 0.85 to 1.05) [11]. In hospitalized patients with decompensated cirrhosis and albumin below 30 g/L, infusions that raised albumin to 30 g/L or more did not change a composite of new infection, kidney dysfunction or death (29.7% against 30.2% with standard care, adjusted OR 0.98, 95% CI 0.71 to 1.33), and more severe or life-threatening serious adverse events occurred in the albumin group [12]. An earlier meta-analysis of 9 controlled trials of correcting low albumin was inconclusive [8].
Population: 1,818 patients with severe sepsis in 100 intensive care units [11]. 777 hospitalized patients with decompensated cirrhosis [12]. 535 patients in 9 trials [8].
Sources: [11], [12], [8]
Caveats: Both large trials were open-label and enrolled people who were seriously ill, so they do not address albumin that is low for other reasons [11][12].

Aspartate aminotransferase (AST)Moderate

Death from any cause

Not linked to death from any cause in a meta-analysis of general population cohorts, where the pooled estimate had a wide interval.

The numbers and sources

Effect: In a meta-analysis of 19 cohort studies of liver enzymes, the authors found no strong evidence of an association of AST with death from any cause. The relative risk comparing the top and bottom thirds of baseline AST was 1.23 (95% CI 0.80 to 1.88) [6].
Population: Over 9.24 million participants and 242,953 deaths across the 19 cohorts, which also covered other liver enzymes. The abstract does not give the number of studies that measured AST [6].
Sources: [6]
Caveats: Observational. The interval runs from 20% lower to 88% higher risk, so a moderate association in either direction is not ruled out [6].

Calcium levelLimited

Surgery for mildly raised calcium from an overactive parathyroid

One trial found parathyroid surgery did not reduce deaths, fractures, heart events or kidney stones over 10 years in mild primary hyperparathyroidism.

The numbers and sources

Effect: In people with mild primary hyperparathyroidism and modest hypercalcemia, 15 had died 10 years after randomization, 8 assigned to surgery and 7 to observation. Over an extended period to 2018 there were 44 deaths, 24 and 20. Cardiovascular events, cerebrovascular events, cancer, peripheral fractures and kidney stones numbered 52 in the surgery group and 49 in the observation group. Vertebral fractures occurred in 7 patients in each group [7].
Population: 191 patients at 8 Scandinavian centers, 95 assigned to parathyroidectomy and 96 to observation [7].
Sources: [7]
Caveats: One trial of 191 people, and 50 withdrew during the study. It enrolled mild disease without known complications, and the abstract does not report calcium levels after surgery [7].

Free T3 (free triiodothyronine)Limited

Death in older adults with normal thyroid tests

One cohort found no link between free T3 and death over 9 years in adults aged 65 and older whose thyroid tests were all in range.

The numbers and sources

Effect: Among 815 adults aged 65 and older with TSH, free T3 and free T4 all in the normal range, 181 died over 9 years. Neither free T3 nor free T4 quartile was associated with death from any cause. TSH in the lowest quartile was, with a hazard ratio of 2.22 (95% CI 1.19 to 4.22) against the highest quartile [7].
Population: 815 community-dwelling adults aged 65 and older in the Invecchiare in Chianti study in Italy [7].
Sources: [7]
Caveats: One observational cohort with 181 deaths, limited to people whose results were inside the reference ranges [7].

Parathyroid hormone (PTH)Limited

Death and illness in a trial of surgery for mild primary hyperparathyroidism

One trial found parathyroid surgery for mild primary hyperparathyroidism did not change deaths, fractures or cardiovascular events over 10 years compared with observation.

The numbers and sources

Effect: After 10 years, 8 patients had died in the surgery group and 7 in the observation group. Over the extended follow-up to 2018 there were 24 and 20 deaths. Cardiovascular events, strokes, cancer, peripheral fractures and kidney stones together numbered 52 in the surgery group and 49 in the observation group, and 7 patients in each group had vertebral fractures [15].
Population: 191 patients with mild primary hyperparathyroidism at eight Scandinavian centers, 95 assigned to surgery and 96 to observation [15].
Sources: [15]
Caveats: One trial of 191 people, 50 of whom withdrew. It studied mild disease with modest high calcium and no known complications, so it does not describe more severe disease [15].