Fish oil: formulation, outcomes and actual risks

Updated 10 Oct 2026

The decision in my food plan

I want a small oily fish that I can enjoy and buy in a manageable amount. The very salty anchovies were a different product from the Crown Prince sardines in my latest food report. The sardine label I transcribed gives 1,740 mg total omega-3 per can. It does not verify how that total divides into EPA (eicosapentaenoic acid) and DHA (docosahexaenoic acid). The calculator’s generic EPA and DHA estimates remain separate, provisional information; adding them to the label total would count overlapping fats twice.

The more ambitious question is whether concentrated omega-3 products would improve my training or prevent disease. A result for a prescription formulation in a selected patient population does not establish the same benefit from an ordinary can of fish or a different capsule. Formulation, dose, comparator, population and measured outcome all matter.

The two cardiovascular trials that prompted the debate

These trials tested products added to statin treatment in people with substantial cardiovascular risk. Neither was a trial of young adults choosing sardines instead of another meal.

Both primary composites included cardiovascular death, heart attack, stroke, coronary revascularization and hospitalization for unstable angina. REDUCE-IT reported an absolute difference of 4.8 percentage points over its follow-up. STRENGTH’s corresponding proportions were essentially similar; its confidence interval does not prove an exactly zero effect. A 4 g product dose also does not automatically mean 4 g of EPA plus DHA. Bhatt et al., REDUCE-IT, 2019; Nicholls et al., STRENGTH, 2020.

Comparing the two randomized trials is informative, but their difference is not itself a randomized comparison of EPA against EPA/DHA. Several features changed together. It cannot isolate DHA as the reason STRENGTH was null.

The comparator question remains consequential

A later REDUCE-IT biomarker analysis found that, at 12 months, the mineral-oil arm had median increases from its own baseline of 21.9% in high-sensitivity CRP and 16.2% in IL-6. The EPA arm changed much less. These are within-arm changes, not those percentages of extra heart attacks. The authors explicitly regarded their implications for interpreting the clinical benefit as uncertain. Ridker et al., 2022.

The reasonable inference is that the comparator deserves scrutiny. This substudy cannot calculate how much of the event difference came from EPA benefit, comparator effects or other mechanisms. There was no third, untreated arm that could settle that decomposition. It would be too strong either to dismiss all of REDUCE-IT as an artifact or to claim the comparator was conclusively irrelevant.

Exposure alone does not settle the discrepancy either. Among 10,382 STRENGTH participants with baseline and 12-month measurements, the highest achieved-EPA third had a median concentration of 151 micrograms/mL. Its adjusted cardiovascular hazard ratio versus corn oil was 0.98, 95% CI 0.83–1.16. This was a secondary analysis of achieved levels after randomization, subject to differences in adherence, metabolism and selection. It was not a new trial assigning people to blood-level targets. Still, it provides counterevidence to the idea that attaining a high EPA concentration guarantees benefit. Nissen et al., 2021.

Later trials prevent a simple EPA-good, DHA-bad rule

RESPECT-EPA tested 1.8 g/day icosapent ethyl against usual care without an added placebo in Japanese patients with chronic coronary disease, statin treatment and a low EPA-to-arachidonic-acid ratio. It randomized 2,506, with 2,460 in the primary analysis. Over median five-year follow-up, primary events occurred in 9.1% versus 12.6%; HR 0.79, 95% CI 0.62–1.00, P=.055. The primary result did not meet conventional statistical significance. A secondary coronary composite favored EPA. The open-label design and different comparator also matter. New atrial fibrillation was higher with EPA: 3.1% versus 1.6%, P=.017. This is suggestive evidence with a real safety signal, rather than a clean replication of every REDUCE-IT claim. Miyauchi et al., 2024.

Conversely, the newer PISCES trial found a benefit from an EPA/DHA mixture in 1,228 adults receiving maintenance hemodialysis. The daily fish-oil product supplied 1.6 g EPA and 0.8 g DHA, against corn oil. Across 3.5 years of follow-up, rates of serious cardiovascular events were 0.31 versus 0.61 per 1,000 patient-days, HR 0.57, 95% CI 0.47–0.70. Its primary endpoint counted all qualifying events, including recurrent events; these are rates, not percentages of participants. The result refutes a universal claim that mixtures cannot work, while its dialysis population makes direct application to my food plan inappropriate. Lok et al., online 2025, print 2026.

Atrial fibrillation and bleeding were measured outcomes

These AF outcomes have different definitions and cannot be treated as identical incidence measures. The observed increases are nevertheless relevant to decisions about high-dose products. Bleeding results were also different, so a statement that every fish-oil formulation caused excess bleeding would misrepresent the trials. REDUCE-IT safety results; STRENGTH safety results.

Lower-dose evidence has a different scope. VITAL Rhythm followed 25,119 older adults without prior AF, testing a daily product with 460 mg EPA and 380 mg DHA. Over median 5.3 years, confirmed AF occurred in 3.7% versus 3.4%, HR 1.09, 95% CI 0.96–1.24, P=.19. That trial did not show a significant increase, but its confidence interval is not a guarantee of no risk. It also does not directly test eating small fish. Albert et al., 2021.

Does extra fish oil improve lifting?

Twenty-one of 28 recruited young adults completed a 2023 trial: ten weeks of training plus 3.85 g EPA+DHA daily, versus safflower oil. The adjusted extra bench-press gain was 5.0 kg, 95% CI 0.07–9.93; lean mass and absolute squat gains did not differ. Multiple primary outcomes were tested, and one assessor knew allocations. Most training sessions were unsupervised. The nominal bench result is a limited positive signal. Heileson et al., 2023.

A 2026 double-blind trial randomized 46 trained men; 39 completed 14 weeks of supervised lower-body training. The paper reports 6.264 g/day total omega-3, including 4.473 g EPA+DHA, versus sunflower oil. Habitual protein averaged 1.8 g/kg/day. Muscle incorporation increased, but strength and muscle-growth gains did not differ. The primary leg-press difference was 1.11 kg, 95% CI −38.40 to 40.62: no detected benefit, not equivalence. Author-uploaded full text was inspected; supplements and registry history remain unread. Santo Andre et al., 2026.

Neither trial establishes treatment for my shoulder. The muscle investigation separates laboratory responses from function.

What I would change now

For daily shopping, an enjoyable small oily fish with a tolerable sodium label remains a workable food choice. The present record provides no established personal reason to escalate to high-dose omega-3 capsules. Prescription cardiovascular treatment belongs to a decision based on the relevant disease history, lipid results, medicines and competing risks; those inputs are not in my food log.

The immediate tracking improvements are simpler: record the exact fish variant, the drained amount actually eaten, and whether a capsule was actually taken. Keep a label’s total omega-3 separate from a verified EPA/DHA split. If I later evaluate a supplement, its study population, formulation and intended endpoint need to match the question. A claim about heart attacks in statin-treated patients should not quietly become a claim about my training recovery.

Back to investigations