If you have spent time on longevity nutrition, you have probably seen the headline number: eat the right mix of food groups and you could cut all-cause mortality risk by more than half. Much of that framing traces back to one landmark paper.
In 2017, Lukas Schwingshackl and colleagues published a systematic review and meta-analysis in the American Journal of Clinical Nutrition. They pooled prospective cohort studies on 12 predefined food groups and all-cause mortality, then modelled dose-response curves to estimate optimal intakes.
It remains one of the most cited syntheses in food-based longevity research — and it is the evidence base behind our Longtime Calculator dose-response curves.
What the study did
The authors searched PubMed, Embase, and Google Scholar through December 2016 for prospective studies linking any of 12 food groups to all-cause death:
- Whole grains, refined grains, vegetables, fruits, nuts, legumes, eggs, dairy, fish, red meat, processed meat, and sugar-sweetened beverages (SSBs).
They ran high-vs-low category analyses, linear dose-response models, and nonlinear spline models to find intake levels where risk flattened or reversed. They also applied the NutriGrade scoring system to rate confidence in the meta-evidence for each food group.
In total, the review included dozens of cohort studies and well over 100,000 recorded deaths across the food groups.
Key findings
Foods that lowered risk
Whole grains had the strongest evidence (NutriGrade: high). Each extra 30 g/day was associated with roughly 8% lower mortality (RR 0.92). Risk fell by about 25% up to ~100 g/day, then levelled off.
Vegetables and fruits both showed clear inverse associations, but with plateaus: benefit accumulated up to roughly 250–300 g/day for each, with little additional gain beyond that in the nonlinear models.
Nuts had the steepest per-serving benefit in linear analysis (RR 0.76 per 28 g/day), though the curve flattened around 15–20 g/day.
Legumes showed a roughly 16% risk reduction up to ~150 g/day in nonlinear analysis, despite a weaker linear signal.
Fish was consistently protective, with about 10% lower risk up to ~200 g/day.
Foods that raised risk
Red meat (RR 1.10 per 100 g/day) and processed meat (RR 1.23 per 50 g/day) showed linear increases — processed meat especially, with risk rising steeply up to ~200 g/day.
Eggs showed a modest positive association at higher intake (~10% higher risk above ~60 g/day), though the authors rated egg meta-evidence as very low quality and noted all included egg studies had unclear or high risk of bias.
SSBs showed no significant linear association overall, but nonlinear analysis suggested ~7% higher risk up to ~250 mL/day.
Neutral or mixed
Refined grains showed no meaningful association.
Dairy was U-shaped: modest benefit at low intake, no harm up to ~750 g/day, then rising risk at very high intake — but the authors cautioned this was largely driven by two cohorts and excluded dairy from their optimal-intake calculation.
The famous 56% number
Combining optimal servings of the protective groups — roughly 3 servings whole grains, 3 vegetables, 3 fruits, 1 nuts, 1 legumes, and 2 fish per day — the authors calculated a 56% relative reduction in all-cause mortality compared with eating none of those foods.
Conversely, stacking risk-increasing intakes (2 servings red meat, 4 processed meat, 1 egg, 1 SSB daily) implied roughly double the risk versus avoiding them.
These are modelled combinations, not a guarantee for any individual. They assume the associations are additive and that confounding is adequately controlled — both big assumptions.
What still holds up
Nearly a decade and many new cohorts later, the directional story of this paper has held remarkably well:
- Plant-forward patterns win. Higher intake of whole grains, vegetables, fruits, nuts, legumes, and fish remains associated with lower mortality in subsequent meta-analyses and large cohort updates (including the 2023 HEI/AMED/AHEI/hPDI comparison in over 120,000 people showing ~20% lower mortality across healthy eating patterns).
- Processed meat is consistently harmful. Later dose-response work has only strengthened the processed-meat signal; there is little credible debate that regular bacon, sausage, and deli meat intake raises chronic-disease and mortality risk.
- Red meat is still best treated as occasional. EPIC and US cohort data continue to show higher risk with increasing red-meat intake, even if the effect size varies by population and processing level.
- Nonlinear plateaus for produce and nuts make biological sense. You do not need infinite broccoli to benefit; the biggest gains come from moving from low to moderate intake — exactly what the 2017 splines showed.
- Whole grains remain the standout. Schwingshackl rated whole-grain evidence highest, and subsequent reviews (including Aune et al., BMJ 2016) have reinforced the link to lower all-cause and cardiovascular mortality.
If you are building a longevity-oriented diet, the 2017 optimal-intake table is still a reasonable starting blueprint:
| Food group | Optimal daily intake (from paper) | Approx. risk vs none |
|---|---|---|
| Whole grains | 3 servings (90 g) | −21% |
| Vegetables | 3 servings (~240 g) | −11% |
| Fruits | 3 servings (~240 g) | −10% |
| Nuts | 1 serving (28 g) | −15% |
| Legumes | 1 serving (100 g) | −10% |
| Fish | 2 servings (200 g) | −10% |
Our Triple-F Longevity Diet and calculator are essentially practical translations of this evidence.
What we would read differently today
Science moved on. Here is where we think the 2017 picture needs updating:
Eggs: the signal weakened
Schwingshackl found a small mortality increase with higher egg intake, but flagged very low evidence quality. Larger meta-analyses since then — including Drouin-Chartier et al., BMJ 2020 and Maziarz et al., 2022 — generally find no association between moderate egg consumption (≤1/day) and all-cause or cardiovascular mortality. A 2025 umbrella review concluded evidence is too weak to discourage eggs as part of an otherwise healthy diet.
Our take: Treat eggs like the Okinawans do — a modest garnish, not a daily centrepiece — but do not fear a couple of eggs per week. The calculator still shows a slight uptick at high intake because it tracks the 2017 curve; that is a modelling choice, not a settled verdict.
Sugar-sweetened beverages: stronger than the 2017 paper suggested
Schwingshackl’s SSB analysis included only five studies and found no significant linear association. Later work is clearer. A 2021 dose-response meta-analysis (Mullee et al.) found each daily serving (~355 mL) linked to 8% higher all-cause mortality, with high-quality evidence.
Our take: SSBs belong in the “minimize” column with processed meat. The 2017 null result was likely underpowered, not wrong directionally.
Ultra-processed foods: the missing category
The 2017 review predates the NOVA framework going mainstream. UPF intake was not one of the 12 groups. Subsequent meta-analyses (Taneri et al., 2022; updated reviews through 2025) link high UPF consumption to roughly 15–29% higher all-cause mortality, with processed meat and SSBs often driving much of the signal.
Our take: “Avoid ultra-processed food” is now as important as any single food-group target. Schwingshackl’s list captures many UPF components (processed meat, SSBs, refined grains) but misses the broader pattern.
Protein: context matters more than the paper allows
The 2017 analysis did not treat protein as its own axis — only food groups. Since then, longevity nutrition has shifted toward age-stratified protein targets. Valter Longo’s Cell review (2022) argues for lower protein until age 65, then higher intake to prevent sarcopenia — a nuance absent from food-group-only modelling.
Our take: A low-meat, plant-forward diet still looks right for mid-life longevity. But if you are over 65 or losing muscle, deliberately increasing protein (see our protein intake article) may matter more than further cutting legumes or fish.
Dairy: still genuinely unclear
The U-shaped dairy curve was fragile even in 2017. Recent cohort data remain mixed — some neutral, some modestly protective for fermented dairy, some adverse at very high intake. Fat content does not clearly split the signal.
Our take: One serving of yoghurt or similar fermented dairy fits a longevity diet; guzzling litres of milk does not.
The 56% figure: illustrative, not predictive
Combining relative risks across food groups assumes independence and multiplicativity — neither is true in real diets. People who eat more vegetables also tend to exercise more, smoke less, and earn more. Sensitivity analyses helped, but residual confounding is unavoidable in observational nutrition.
Our take: Use the 56% as motivation, not a forecast. The food4healthylife.org calculator (built on this same paper) is useful for comparing scenarios, not for predicting your exact expiry date.
Bottom line
Schwingshackl et al. did something rare and valuable: they put 12 food groups on the same mortality scale, with dose-response curves instead of just “eat more / eat less” slogans. That structure still powers how we think about longevity nutrition on this site.
The core message endures — more whole plants, fish, and nuts; less processed meat and junk drink; treat red meat as a condiment — but three areas have shifted since 2017: eggs look less scary at moderate intake, SSBs look worse, and ultra-processed food deserves its own spotlight.
If you want to explore how a specific diet stacks up against these curves, try the Longtime Calculator. If you want the original source, the paper is here on ScienceDirect (DOI: 10.3945/ajcn.117.153148).