The decision this page serves. A person asks whether to take fish oil “for health”. That question has no answer as posed. Marine omega-3 supplementation (the isolate — capsules of EPA/DHA or prescription EPA, not fish as food) moves different outcomes in opposite directions, and the sign flips with the dose. This page owns the cross-outcome picture that each instance page holds only one slice of; every figure below is extracted on the linked page and delegated here, not re-sourced. — the cross-outcome join and the two decision rules are this page’s; each magnitude is quoted on its instance page.
The outcome menu — one exposure, opposite signs
| Outcome (stratum) | Best-evidenced effect | Dose / form | Held on |
|---|---|---|---|
| CV events, high-CV-risk statin-treated | BENEFIT HR 0.75 (0.68-0.83), NNT 21 (REDUCE-IT) | 4 g/d icosapent ethyl (EPA-only) | Vitamin and Mineral Supplements for Disease Prevention |
| CV events (MACE), general primary prevention | NULL HR 0.92 (0.80-1.06) (VITAL) | 1 g/d EPA+DHA | Vitamin and Mineral Supplements for Disease Prevention |
| Atrial fibrillation | HARM HR 1.25 (1.07-1.46); 1.49 (1.04-2.15) at >1 g/d (Gencer) | rises with dose over 1-4 g/d | Omega-3 Supplementation and Atrial Fibrillation |
| Preterm / early preterm birth | NULL on low-RoB (PTB<37 0.92; ePTB<34 0.82, both NS) | mixed DHA/EPA doses, pregnancy | Omega-3 Supplementation in Pregnancy |
| Depression — treatment of symptoms | small BENEFIT SMD -0.28 (-0.47, -0.09), EPA | varied | Depression and Modifiable Exposures |
| Depression — prevention in the non-depressed | NULL RR 1.01 (0.92-1.10) | >=6 mo | Depression and Modifiable Exposures |
| Child neurodevelopment (maternal supplement) | NULL on any functional outcome | DHA in pregnancy | Fish and Seafood Consumption |
| Invasive cancer | NULL HR 1.03 (0.93-1.13) (VITAL) | 1 g/d | Vitamin and Mineral Supplements for Disease Prevention |
The fourth column of a cross-source table normally asks “same quantity?”; here the answer is NO in every pairing by construction — these are distinct outcomes, so the table is a type-B disambiguation, not a contrast. That is the point: the single word “omega-3” names objects that a decision must keep apart.
The 4 g/d decision is a joint benefit-harm weigh, not “supplement: yes/no”
The benefit and the harm sit at the same end of one dose axis, and REDUCE-IT is the case that makes it undeniable: at 4 g/d icosapent ethyl, in a statin-treated high-CV-risk population, the same trial reports a CV-event reduction (HR 0.75, NNT 21) and an increase in atrial fibrillation (RR 1.35 [1.11-1.65]) — the benefit held on Vitamin and Mineral Supplements for Disease Prevention, the harm on Omega-3 Supplementation and Atrial Fibrillation. Gencer’s pooled dose-response puts the AF harm at its steepest in exactly this >1 g/d region (1.49). So a high-risk person choosing 4 g/d weighs a large ischemic benefit against a smaller-absolute but real arrhythmia harm on one dose — a net-effect judgement, not a supplement-or-not one. The low dose (1 g/d, VITAL) resolves the weigh trivially the wrong way: null CV benefit, and only the mild AF slope remains. — the join across the two pages is this page’s; both magnitudes are extracted on them.
The dose is a curve, not a two-point pair — but it bends only for the high-CV-risk stratum
The VITAL-null / REDUCE-IT-benefit contrast reads as two points; a pooled dose-response now sits behind them. Across 13 CVD-outcome RCTs (n=127,477, dose range 376-4000 mg/d), a meta-regression found a linear, non-plateauing clinical dose-response — each +1000 mg/d lowering risk — that held with and without the 4 g REDUCE-IT outlier for exactly two outcomes: total CVD (~17% per +1000 mg/d in the outlier-excluded analysis; ~9% with the 4 g trial) and major vascular events; the MI / total-CHD slope appeared only when the single 4 g trial was included, and stroke stayed null throughout (Hu 2019, figures held on Vitamin and Mineral Supplements for Disease Prevention). Two decision-relevant reads:
- It replaces “does fish oil work?” with “how much, for whom?” — the ischemic benefit rises with dose, so 1 g/d and 4 g/d are different bets, not the same supplement at two strengths. This is why the 4 g/d weigh above is a genuine dose decision.
- The pooled benefit even EXCLUDING the 4 g outlier is a high-CV-risk average, not a usual-risk one. Pooling the mostly ~840 mg/d trials without REDUCE-IT still lowered MI, CHD death, total CHD, CVD death and total CVD — but those trials are dominated by secondary-prevention / established-CVD populations, and VITAL, the only usual-risk trial, stayed null on its composite. So the curve bends for the high-risk stratum; it does not license generic low-dose supplementation for a healthy general-prevention adult (the row below). And the slope is Hu’s own «highly exploratory», 4-g-outlier-driven estimate, not a settled law.
— the “curve not a pair” framing and the high-risk-only reading are this page’s; every RR, slope and caveat is Hu 2019’s, quoted on the instance page. Hu is a type-F composite over the same trial base (VITAL, REDUCE-IT et al.) and shares an author with VITAL — it refines confidence in the dose axis, it is not independent corroboration.
”Should I take fish oil” is ill-posed — name the outcome and the dose
For a general-population adult with no established CV disease, the held evidence offers no outcome with a benefit at the doses sold over the counter: CV events null at 1 g/d, cancer null, preterm and neurodevelopment null, depression prevention null. What remains at that dose is the mild AF slope. So generic low-dose supplementation “for general health” targets no outcome the fabric can show it moves, while carrying a small harm — a null-or-net-negative bet, not a neutral one. The decision becomes positive only when it names an outcome and a dose that has evidence: high-dose EPA for a high-CV-risk statin-treated person (REDUCE-IT, and weigh the AF harm), or EPA for someone being treated for depressive symptoms (small effect, fragile to EPA-type and dose).
Two firewalls
- Isolate is not fish. Every finding here is the supplement. The fish-as-food case is separate — it carries a matrix (iodine, selenium), a mercury tradeoff, and a CHD-mortality signal the isolate RCTs do not reproduce at supplemental doses. REDUCE-IT’s own author warns the result “should not be generalized to… dietary-supplement preparations of n-3 fatty acid mixtures”. -> Fish and Seafood Consumption, Is the Food Category Doing Any Work
- The benefit may concentrate in the deficient. The isolate RCTs pool Western-replete populations, which cannot resolve whether a baseline-omega-3-deficient person benefits where the replete do not. The repletion-vs-enhancement cross-cut is owned on Deficiency Repletion vs Enhancement; the pregnancy and depression pages each flag it as their live route-(b) gap.