Most nutrition arguments are really arguments about substitution, badly labelled. Somebody says seed oils are dangerous and means “instead of butter”, somebody else says seed oils are protective and also means “instead of butter”, and both of them reach for the same literature and quote opposite halves of it. The two papers that get cited most are a 2013 randomized trial that found harm and a 2025 cohort analysis that found benefit. Both are correct. Neither was testing the oil in your cupboard.
The trial: 458 men, safflower oil, 1966 to 1973
The BMJ 2013 paper by Ramsden and colleagues did something unusual: it recovered data from the Sydney Diet Heart Study, a single-blinded, parallel-group randomized controlled trial run between 1966 and 1973 and never fully reported. 458 men aged 30 to 59, each with a recent coronary event.
The intervention group (n = 221) was told to replace saturated fats — from animal fats, common margarines and shortenings — with omega-6 linoleic acid from safflower oil and safflower oil polyunsaturated margarine. The controls (n = 237) got no specific dietary instruction and no study foods. All non-dietary aspects were designed to be equivalent.
All-cause death: 17.6 per cent versus 11.8 per cent. Hazard ratio 1.62 (95 % CI 1.00 to 2.64), p = 0.05. Cardiovascular death: 17.2 versus 11.0, hazard ratio 1.70 (95 % CI 1.03 to 2.80), p = 0.04. Coronary heart disease death: 16.3 versus 10.1, hazard ratio 1.74 (95 % CI 1.04 to 2.92), p = 0.04.
And then the part that decided the paper’s title: when those recovered numbers went back into the updated meta-analysis of linoleic acid intervention trials, the pooled result was no longer significant. Death from coronary heart disease, hazard ratio 1.33 (95 % CI 0.99 to 1.79), p = 0.06. Cardiovascular disease, 1.27 (95 % CI 0.98 to 1.65), p = 0.07.
So the honest reading of this paper is not “omega-6 seed oils kill people”. It is: in men who had already had a coronary event, replacing saturated fat with safflower oil margarine did not reduce death, and in the pooled analysis it did not reduce it either. The authors put it plainly — clinical benefits of the most abundant polyunsaturated fatty acid have not been established.
The cohorts: 65,179 adults and 20,571 deaths
The 2025 American Journal of Clinical Nutrition analysis followed 65,179 adults from the Nurses’ Health Study and the Health Professionals Follow-up Study who were free of cardiovascular disease, cancer and diabetes at baseline in 1994. They documented 20,571 deaths through 2020 — 1,334,603 person-years — and reassessed diet every four years.
A 5 per cent energy increment in polyunsaturated fatty acid: hazard ratio 0.83 for all-cause mortality (95 % CI 0.78 to 0.89). Monounsaturated fatty acid, 0.91 (95 % CI 0.87 to 0.94). A 1 per cent energy increment in trans fatty acid, 1.10 (95 % CI 1.04 to 1.17).
Here is the sentence that matters more than any of those: changes in saturated fatty acid intake were not associated with all-cause mortality. The authors instead ran substitution models. Replacing 5 per cent of energy from saturated fat with polyunsaturated fat was associated with 19 per cent lower all-cause mortality (hazard ratio 0.81, 95 % CI 0.75 to 0.87). Replacing 0.3 per cent of energy from saturated fat with marine n-3 polyunsaturated fat, 11 per cent lower (hazard ratio 0.89, 95 % CI 0.84 to 0.93).
Read that structure again. Saturated fat on its own predicted nothing. The result appears only when something replaces it. That is what a substitution finding is, and it is the entire difference between this paper and the 2013 one.
The same analysis in people with diabetes
The BMJ 2019 paper ran the model in 11,264 participants with type 2 diabetes from the same two cohorts, with 2,502 deaths documented including 646 from cardiovascular disease.
Polyunsaturated fatty acid intake against total carbohydrates: hazard ratio 0.76 for cardiovascular mortality (95 % CI 0.58 to 0.99, p for trend 0.03). Marine n-3, 0.69 (95 % CI 0.52 to 0.90, p = 0.007). Alpha-linolenic acid, 1.13 (95 % CI 0.85 to 1.51) — pointing the other way and not significant. Linoleic acid, 0.75 (95 % CI 0.56 to 1.01).
Isocalorically replacing 2 per cent of energy from saturated fatty acids with total polyunsaturated fatty acids, or with linoleic acid, was associated with 13 per cent (hazard ratio 0.87, 95 % CI 0.77 to 0.99) or 15 per cent (0.85, 95 % CI 0.73 to 0.99) lower cardiovascular mortality.
Note that linoleic acid — the omega-6 fatty acid that Sydney Diet Heart put into safflower oil margarine — is the compound on both sides of this argument. In this cohort it tracks with lower cardiovascular mortality. In the RCT it tracked with higher mortality. The difference is not the molecule.
Pattern-level evidence, which is what guidelines actually rest on
Guidelines do not recommend a single oil. They recommend a pattern, and the largest test of the current version of that idea appeared in The Lancet Planetary Health in 2024: the Planetary Health Diet, proposed by the EAT-Lancet Commission in 2019, scored over fifteen food groups, with unsaturated oils scored positively and starchy vegetables, dairy, red and processed meat, poultry, eggs, saturated and trans fats and added sugar scored negatively.
Three US cohorts: 62,919 women from the Nurses’ Health Study I, 88,535 women from NHS II, and 42,164 men from the Health Professionals Follow-up Study. 9,831 confirmed cardiovascular disease cases over 4,541,980 person-years. Highest versus lowest quintile of the index, hazard ratio 0.83 (95 % CI 0.78 to 0.89, p-trend < 0.0001). Coronary heart disease 0.81 (95 % CI 0.74 to 0.88). Total stroke 0.86 (95 % CI 0.78 to 0.95).
A pattern score is not a mechanism. It cannot tell you which component did what. It tells you that the combination behaves differently from the average diet, which is the unit of advice a guideline can actually give.
The American Heart Association’s 2020 science advisory in Circulation frames the same way: observational studies generally do not indicate a significant association between dietary cholesterol and cardiovascular risk, meta-analyses of intervention studies differ, and guidance should focus on food patterns — Mediterranean-style and DASH-style — that emphasise nuts, seeds and liquid vegetable oils rather than on a nutrient target.
What a bottle cannot tell you
Every one of the papers above treats “vegetable oil” or “polyunsaturated fat” as a category, and none of them assayed the oil anyone actually eats. What would settle it is a document from the batch: the fatty acid profile that confirms what the label claims, the peroxide value that shows how far the oil has already oxidised, the level of a tocopherol that protects it.
How to read the next claim
Four questions, in this order. What was substituted for what — the entire literature above moves when you change the answer. Is this a randomized comparison or an observational one, and are the two being quoted as if they were the same kind of evidence. What population — men with a previous coronary event in 1968 are not the average shopper in 2026. And what is the endpoint — LDL concentration, a modelled substitution hazard ratio, or a count of deaths.
What this evidence does not show
None of these papers tested a specific seed oil as sold today; safflower oil and margarine from the 1960s are not the same product as anything on a 2026 shelf, and no trial here compares one modern oil against another. The substitution analyses are isocaloric modelling on observational data, which assumes the model is right. Nothing here resolves the omega-6 to omega-3 ratio question, which none of the five papers measured. And a hazard ratio of 0.83 for a whole dietary pattern is not a claim any single food in that pattern can carry on its own.
This article summarises published randomized trials and prospective cohort analyses. It is not medical advice and it is not a recommendation for any individual.
