Nut consumption and risk of cardiovascular disease, total cancer, all-cause and cause-specific mortalitya systematic review and dose-response meta-analysis of prospective studies

Four point four million. That's the number of premature deaths in a single year — 2013 — across the Americas, Europe, Southeast Asia, and the Western Pacific that a major meta-analysis linked to one dietary habit. Not smoking. Not physical inactivity. Not excess sugar. Eating too few nuts. Sit with that for a second. Then let's look at how that number was built, what the underlying data actually say, and where the honest limits of this research lie. The team behind this work, led by Dagfinn Aune and colleagues, set out to fill a specific gap. Earlier reviews had pointed toward a link between nut consumption and lower risk of coronary heart disease and overall mortality — but findings on stroke were inconsistent, cancer associations were unclear, and cause-specific mortality from diseases like diabetes, respiratory illness, and infectious disease had never been systematically examined. Worse, prior reviews had missed significant data: studies together representing forty-seven thousand sixty-one deaths and more than seven hundred forty-eight thousand additional participants had either been omitted or published after those reviews closed. Aune and colleagues searched PubMed and EMBASE through July 2016, screened nearly forty-eight thousand records, and ultimately included twenty prospective cohort studies across twenty-nine publications — nine from the United States, six from Europe, four from Asia, and one from Australia. The dose-response synthesis covered up to eight hundred nineteen thousand four hundred forty-eight participants and eighty-five thousand eight hundred seventy deaths. Before getting into the results, it's worth understanding what this kind of study can and cannot do. These are prospective cohort studies — researchers enroll people, ask about their diet, then follow them forward in time to see who develops disease or dies. The strength of pooling many such cohorts is that rare outcomes become detectable and modest effects become visible. To combine results, the authors used random-effects meta-analysis, a method that weights each study's contribution while accounting for genuine variation between studies — what statisticians call heterogeneity. They measured this with a statistic called I-squared, which tells you the proportion of variation across studies that reflects real differences rather than chance. They also standardized the dose. Everything is reported per twenty-eight grams per day — one ounce, roughly a small handful. And crucially, they tested whether risk changed in a straight line across intake levels or whether it curved. That nonlinear analysis, using a technique called restricted cubic splines, turned out to be one of the most important parts of the study. Now, what the data actually show. For coronary heart disease, the summary relative risk — the ratio comparing risk in people with higher nut intake to those with lower intake — was zero point seventy-one per twenty-eight grams per day. That's a twenty-nine percent lower risk. The ninety-five percent confidence interval ran from zero point sixty-three to zero point eighty, a range that sits well clear of one, meaning the result is statistically solid. And the dose-response curve was nonlinear: most of the benefit accumulated up to about fifteen to twenty grams per day, with little additional reduction above that level. Cardiovascular disease broadly showed a summary relative risk of zero point seventy-nine — a twenty-one percent lower risk — with a similar nonlinear pattern. For all-cause mortality, the figure was zero point seventy-eight per twenty-eight grams per day, or roughly a twenty-two percent lower risk. Total cancer came in at zero point eighty-five, about a fifteen percent lower risk, and unlike the cardiovascular outcomes, cancer showed a linear rather than curved relationship with intake. Stroke was the outlier. The summary relative risk per twenty-eight grams per day was zero point ninety-three, and the confidence interval ran from zero point eighty-three to one point zero five — it crosses one. The association is weaker and statistically non-significant in the pooled analysis. There was even a slight J-shaped signal at very high intakes, though that pattern didn't hold up in subgroup analyses. Stroke, for now, remains an open question. One finding worth pausing on: the results were essentially the same for tree nuts and for peanuts. That matters because peanuts are technically legumes, not tree nuts, and they are far more affordable and widely accessible. If the protective association holds for peanuts — almonds at one end, peanuts at the other — then this isn't a finding limited to expensive specialty foods. Beyond the cardiovascular and cancer results, Aune and colleagues pushed into less-charted territory: cause-specific mortality. And here the numbers get striking, with the important caveat that they come from far fewer studies. Respiratory disease mortality showed a summary relative risk of zero point forty-eight per serving per day — roughly a halving of risk — based on three studies. Infectious disease mortality showed a point estimate of zero point twenty-five, a seventy-five percent lower risk, though that came from only two studies and the confidence interval was wide: zero point zero seven to zero point eighty-five. Diabetes mortality sat at zero point sixty-one per serving per day across four studies, with a tighter confidence interval of zero point forty-three to zero point eighty-eight. Neurodegenerative disease mortality showed a point estimate of zero point sixty-five, but the confidence interval included one. Kidney disease produced a relative risk of zero point twenty-seven with a confidence interval stretching nearly to one point ninety-one — statistically, that's compatible with almost any effect, positive or negative. These are signals, not conclusions. Two to four studies each is not a sufficient base to establish an association. But they point toward hypotheses worth testing: that nut consumption may affect mortality through pathways that go well beyond the cardiovascular system. Now to that four point four million figure. Aune and colleagues used what's called a population-attributable risk calculation — a formula that combines the prevalence of low nut intake in a population with the relative risk associated with that low intake to estimate the fraction of deaths that could, in theory, be avoided if everyone reached a target exposure level. The target they chose was twenty grams per day, selected because the dose-response curves showed little additional benefit above roughly fifteen to twenty grams. They applied this to cause-specific mortality data from the Global Burden of Disease Study two thousand thirteen, across the Americas, Europe, Southeast Asia, and the Western Pacific. The breakdown: an estimated one point nineteen million coronary heart disease deaths, one point oh seven million respiratory disease deaths, four hundred sixty-nine thousand cancer deaths, and one hundred thirty-nine thousand diabetes deaths — summing to four point four million total. The paper is explicit: this estimate is conditional. It assumes the associations are causal. That assumption is the critical one. Why might nuts actually reduce mortality? The biological story involves the nutritional package nuts deliver: unsaturated fatty acids, protein, fiber, vitamin E, potassium, magnesium, and various antioxidants and phytochemicals. Intervention studies cited by Aune and colleagues show that nut consumption reduces total cholesterol, low-density lipoprotein cholesterol, apolipoprotein B, and triglycerides in a dose-dependent way, and also reduces endothelial dysfunction and insulin resistance. For cancer specifically, nuts' antioxidants may reduce oxidative DNA damage, lower circulating insulin-like growth factor one, and affect the gut microbiota. These are plausible pathways — they don't prove causation, but they make the associations biologically coherent. Which brings us to the honest accounting. These are observational studies. Aune and colleagues adjusted for a long list of potential confounders — smoking, body mass index, physical activity, alcohol, diet quality, and more — but adjustment is not the same as elimination. Healthy eaters tend to do many healthy things. Residual confounding cannot be ruled out. Nut intake was self-reported in most studies, introducing measurement error. Heterogeneity was substantial for several outcomes: I-squared reached sixty-six percent for all-cause mortality and sixty percent for cardiovascular disease. Small-study effects were suggested for all-cause mortality — Egger's test returned a p-value of zero point zero two, though the signal disappeared after excluding the very smallest studies. For the cause-specific mortality findings, the base of evidence is simply thin. None of this negates the findings. It calibrates them. The coronary heart disease result — relative risk zero point seventy-one, confidence interval zero point sixty-three to zero point eighty, across eleven studies — is not a fragile signal. The all-cause mortality result — zero point seventy-eight, interval zero point seventy-two to zero point eighty-four, across fifteen studies — is similarly consistent. Even discounted for observational uncertainty and possible confounding, the magnitude of these associations is large enough that dismissing them would require strong evidence to the contrary. What the field needs next are randomized trials and better-measured dietary exposures. What the data already have is a pattern that is hard to look away from. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.