General-population longevity levers — what the evidence supports

A meta-analysis of 38 RCTs (149,051 participants) associated omega-3 with modestly lower cardiovascular mortality (about 7% lower) and fewer non-fatal heart attacks and coronary events; EPA alone outperformed EPA plus DHA. Caveat: higher-dose omega-3 slightly raised atrial-fibrillation risk — more is not always better.

Creatine is well-supported for building strength, lean mass, and recovery alongside resistance training. Brain and cognitive benefits are plausible but not proven — a 2024 EFSA panel found the cognition evidence inconclusive (methodological double-counting).

Zone 2 (a conversational pace) builds an aerobic base at low injury risk and is metabolically useful. The popular claim that it is uniquely optimal for mitochondria or fat oxidation is contested — a 2025 review argues higher intensities may do more, especially for non-elite trainees. Both matter; do not let easy cardio crowd out intensity. (Library record of the debate; the Zone 2 drawer stays focused on the practical case.)

A 2025 meta-analysis (11 RCTs, 3,177 people) found cold-water immersion may modestly help stress, sleep, and quality of life (time-dependent), with no significant immune benefit, a small and low-quality evidence base, and an acute rise in inflammation right after immersion. Practical caveat: do not ice right after lifting — it can blunt strength and muscle gains.

VITAL randomised 25,871 US adults (men 50 and over, women 55 and over) to 2,000 IU/day of vitamin D3 or placebo, in a two-by-two design that also tested omega-3, and followed them for a median of 5.3 years. Vitamin D did not reduce the primary endpoint, total invasive cancer of any type (hazard ratio 0.96, 95% CI 0.88-1.06), or major cardiovascular events (0.97, 0.85-1.12). All-cause mortality showed no detectable difference (0.99, 0.87-1.12) — but as a secondary endpoint its confidence interval still spans anything from a 13% reduction to a 12% increase, so read that as no evidence of benefit rather than proof of none. One signal did emerge: death from cancer trended lower (0.83, 0.67-1.02) and reached statistical significance in analyses that excluded the first two years of follow-up to allow for latency (0.75, 0.59-0.96) — suggestive, not definitive. No excess hypercalcemia, kidney stones or other adverse events were identified. Important limit: this recruited a general population, not people with documented deficiency — only about 13% of those with a measured level were below 20 ng/mL — so it says nothing about correcting a low measured level; if you think yours is low, testing and dosing are a conversation with your doctor.

A meta-analysis pooled 107 cohort studies — 4,838,825 people and 425,564 deaths — to re-test the familiar claim that light drinking is associated with lower mortality. Once the models adjusted for sampling variation, former-drinker bias (people who had quit drinking, often because they were already ill, sitting in the non-drinker comparison group), and other prespecified study-quality criteria, occasional drinking (more than 0 up to 1.3 g of ethanol per day; RR 0.96) and low-volume drinking (1.3–24 g/day; RR 0.93) showed no significant mortality benefit versus lifetime non-drinkers. Risk was nonsignificantly higher at 25–44 g/day (RR 1.05), and significantly higher at 45–64 g/day (RR 1.19) and at 65 g or more (RR 1.35). Female drinkers had significantly higher risk than female lifetime non-drinkers (RR 1.22). This pools observational cohorts rather than trials, so it cannot establish cause — the authors' point is that the apparent J-curve largely shrank once former-drinker bias and study quality were accounted for. If you drink heavily and are thinking about cutting back, that is worth a conversation with your doctor.

An umbrella review in the BMJ pooled 45 analyses drawn from 14 meta-analyses, covering 9,888,373 people. Higher ultra-processed food exposure was directly associated with 32 of the 45 outcomes examined. On the review's credibility scale, evidence was convincing for cardiovascular-disease-related mortality (risk ratio 1.50, 95% CI 1.37-1.63 — though GRADE quality for that particular result was very low) and for type 2 diabetes (risk ratio 1.12, 95% CI 1.11-1.13 per 10% increment in ultra-processed food intake, moderate quality). It was highly suggestive for all-cause mortality (risk ratio 1.21), depressive outcomes (hazard ratio 1.22), adverse sleep-related outcomes (odds ratio 1.41) and obesity (odds ratio 1.55). All of this is observational, and the evidence base is weak: only 4 of the 45 pooled analyses rated moderate quality under GRADE, 22 rated low and 19 very low. The pattern is broad and consistent, but it cannot show that processing itself is the cause.

Bathgate, Galpin and colleagues (Eur J Appl Physiol, 2018) studied a single pair of identical 52-year-old male twins — one with more than 30 years of endurance training, the other with no consistent exercise. The trained twin had a VO2max 12.4 ml/kg/min higher (47.5 against 35.1), 8.6 percentage points less body fat, and a far more slow-twitch muscle: roughly 94% MHC I fibres against about 40% in his brother, a gap of 55 percentage points or about 2.4 times as many. He also had lower resting heart rate, blood pressure, cholesterol, triglycerides and glucose — but lower muscle size and strength. Two limits travel with it: this is one twin pair with no statistics, and the authors themselves note the body-composition difference may not be down to exercise alone, since the untrained twin also ate slightly more than he burned, which they estimate could account for around 7 kg over 30 years.

The claim that metformin extends life began in St Petersburg: in 2008 Vladimir Anisimov's lab reported that female outbred SHR mice given metformin in their water had a 37.8% longer mean lifespan — with no change in tumour incidence. That number travelled the world; what it rests on did not. A US National Institute on Aging study found a modest lifespan gain in male mice only at 0.1% of diet, and a ten-fold higher dose was toxic. The NIA's three-site Interventions Testing Program — genetically diverse mice, both sexes, the strongest test that exists — reported that metformin alone "did not significantly extend lifespan". And a 2025 meta-analysis of 911 effect sizes across eight vertebrate species, including Anisimov's own papers, is titled for its finding: rapamycin, not metformin, mirrors the lifespan extension of dietary restriction. All of this is in animals; none of it is a verdict on metformin for people with diabetes, who take it for a different reason. It is a verdict on the headline.