Associations Between Plasma Omega‐3 and Fish Oil Use With Risk of Atrial Fibrillation in the UK Biobank

Abstract

Background

Meta‐analyses of randomized trials with omega‐3 products found an increased risk of atrial fibrillation (AF), and a recent study from the UK Biobank concluded that fish oil supplement (FOS) use was likewise associated with increased risk of AF. Conversely, a meta‐analysis based on blood levels of omega‐3 fatty acids found an inverse relationship with AF risk. Thus, the relationship between omega‐3 fatty acids and risk of AF remains unclear.

Methods

We conducted a retrospective analysis of UK Biobank data to link both plasma omega‐3 levels and reported FOS use with AF risk. Among participants without prevalent AF, a random sample of 261 108 had data on plasma omega‐3 levels and 466 169 reported about FOS use. The primary outcome was incident AF during follow‐up (median, 12.7 years). Multivariable‐adjusted hazard ratios (HRs [95% CIs]) for fatty acids were computed continuously (per interquintile range [IQ₅R]) and by quintile. HRs were computed for dichotomous vs. continuous FOS use.

Results

Plasma omega‐3 levels were inversely associated with incident AF (HR per IQ₅R, 0.89 [95% CI, 0.86–0.93]). FOS use was reported by 31% of the cohort and was more common in older individuals. After multivariable adjustment, no association was observed between FOS use and AF risk (HR, 1.00 [95% CI, 0.97–1.02]).

Conclusions

Higher circulating omega‐3 levels were linked to reduced AF risk in the UK Biobank. Further, after age was adjusted for as a continuous variable (as opposed to a dichotomous variable as in an earlier report), no association was found between FOS use and risk for AF.

Nonstandard Abbreviations and Acronyms

AHA

American Heart Association

DHA

docosahexaenoic acid

DPA

docosapentaenoic acid

EPA

eicosapentaenoic acid

FOS

fish oil supplement

IQ₅R

interquintile range

NMR

nuclear magnetic resonance

Clinical Perspective.

What Is New?

• This study shows that individuals with the highest plasma omega‐3 levels were the least likely to develop atrial fibrillation (AF) during follow‐up, and no increase in risk of AF among fish oil supplement users in the UK Biobank was found, contradicting a previous report linking such use with a higher risk of AF.

What Are the Clinical Implications?

• The findings that individuals in the UK Biobank with the highest plasma omega‐3 fatty acid levels were at the lowest risk for developing AF suggests that moderate intake of omega‐3 fatty acids from the diet and/or fish oil supplements may have a beneficial effect on AF risk.

• Because fish oil supplements are among the most commonly used dietary supplements, a prior report suggesting that their use increased risk of AF raised serious concerns in both the medical and lay communities. The current findings provide important reassurance against those concerns.

The most prevalent sustained cardiac arrhythmia is atrial fibrillation (AF), with >3 million new cases reported each year and a global prevalence of >37 million people. Moreover, it is a significant public health concern, elevating risks of hospitalization, premature mortality, heart failure, and thromboembolic events, including stroke. ¹ Higher blood levels of omega‐3 fatty acids have been favorably associated with lower risk of AF, ² as well as all‐cause mortality, ³ heart failure, ⁴ and stroke. ⁵ However, randomized controlled trials assessing pharmaceutical omega‐3 products containing eicosapentaenoic acid (EPA) and/or docosahexaenoic acid (DHA) at doses ranging from 2 to 4 g/d have documented an increase in risk of AF. ⁶ , ⁷ These latter findings appear to be at odds with the aforementioned reduced risk for AF observed in a meta‐analysis of 17 biomarker‐based (ie, blood EPA + DHA) prospective cohort studies including nearly 55 000 individuals. ² Further complicating this are 2 reports using UK Biobank data which concluded that self‐reported use of fish oil supplements (FOSs; over‐the‐counter sources of EPA and DHA) was associated with an increased risk of incident AF. ⁸ , ⁹ In light of these conflicting findings, we sought to further investigate the relationship between omega‐3 levels and AF risk in the UK Biobank using serum omega‐3 levels as the exposure and to reevaluate the reported link between self‐reported FOS use and AF risk.

MATERIALS AND METHODS

Data Availability

Because of the sensitive nature of the data collected for this study, requests to access the data set from qualified researchers trained in human subject confidentiality protocols may be sent to the UK Biobank at ukbiobank@ukbiobank.ac.uk.

Sample

The UK Biobank is a prospective, population‐based cohort of 502 366 individuals, aged 40 to 69 years, recruited in the United Kingdom between April 2007 and December 2010. ¹⁰ UK Biobank has ethical approval (reference 11/NW/0382) from the Northwest Multi‐centre Research Ethics Committee as a Research Tissue Bank. All participants gave electronic signed informed consent. The UK Biobank study was conducted according to the guidelines established by the Declaration of Helsinki. The UK Biobank protocol is available online (http://www.ukbiobank.ac.uk/wp114content/uploads/2011/11/UK‐Biobank‐Protocol.pdf). The University of South Dakota institutional review board approved the use of these deidentified, publicly available data for research purposes (IRB‐21‐147). Within the cohort, a random sample of 274 123 patients had data on serum fatty acids. Of these, 4639 had prevalent AF and another 8376 were missing at least one covariate, leaving 261 108 patients as the primary analytic sample size. Secondary analyses used a larger sample of individuals (n=466 169) who answered a question on self‐reported FOS use, did not have prevalent AF, and were not missing continuous covariates, regardless of whether serum fatty acid data were available.

Outcome Assessment

The date of first diagnosis of AF is available via the primary UK Biobank data set (variable source code: 131350) and is primarily based on electronic medical record information from hospital admissions or primary care, with a limited number of events based on death registries or self‐report. Individuals with date of first AF before measurement of serum fatty acids were dropped from the analytic sample (see previous paragraph). A complete list of UK Biobank variable identification corresponding to covariates, exposures, and outcomes is available in Tables S1 and S2.

Exposure Assessment

The UK Biobank includes data on 2 serum omega‐3 metrics: DHA and total omega‐3s. We constructed a third metric, other omega‐3 fatty acids (ie, the sum of alpha linolenic acid, EPA, and docosapentaenoic acid [DPA]), as the difference between these two. Serum samples were collected at baseline and analyzed for omega‐3 fatty acids by nuclear magnetic resonance (NMR; Nightingale Health Plc). ¹¹ We note that in a recent interlaboratory experiment comparing fatty acid determinations by NMR and gas chromatography, ¹² the strongest predictor of the non‐DHA omega‐3% was EPA % (R ²=47%). Equations to convert serum omega‐3 metrics measured by NMR to plasma and red blood cell metrics measured by gas chromatography, based on Schuchardt et al, ¹² are provided in Table S3. For the purpose of updating the prior meta‐analysis by Qian et al ² for the EPA, DPA, and EPA + DHA associations with risk for AF, the estimated plasma (not red blood cell count) levels of these 3 fatty acid metrics were computed from the NMR data. Omega‐3 biomarkers from the NMR analysis were analyzed as individual quintiles or after standardization by the interquintile range (IQ₅R) of the fatty acid (eg, dividing each fatty acid metric by the difference between the 90th and 10th percentile). In separate analyses, we considered self‐reported FOS use (yes/no; UK Biobank Variable Identification: 6179) as a fourth exposure.

Covariate Assessment

In our multivariable models for the analysis of risk for AF by serum omega‐3 levels, we adjusted for the following demographic, behavioral, biomarker, and health‐related variables: biological sex (male/female); body mass index; self‐reported race or ethnicity (Asian, Black, White, other); education (college, high school, less than high school); self‐reported alcohol consumption (rarely, monthly, 1 or 2×/week, 3 or 4×/week, daily); self‐reported exercise (quartiles of moderate to vigorous exercise; as minutes per week); smoking status (never, previous, current); serum linoleic acid; other serum omega‐6 fatty acids (total omega‐6 fatty acids minus linoleic acid); self‐reported use of β‐blockers; self‐reported treatment for hypertension; self‐reported treatment for high cholesterol; prevalent diagnosis of diabetes; prevalent diagnosis of cardiovascular disease; and prevalent diagnosis of heart failure.

The covariates used in our replication of the FOS and risk for AF study reported by Zhang et al ⁹ were those originally used by Zhang and are listed in Table S4. These were similar to the covariates listed above for the serum omega‐3 versus AF risk analysis, with the exception that variables such as oily fish intake, the Townsend Deprivation Index, and a genetic risk score were included in the replication. Naturally, serum omega‐6 fatty acid levels were not included since they were not available on the full cohort used in this analysis.

Statistical Analyses

Descriptive sample characteristics are summarized using standard statistical methods. Cox proportional hazards models were used to estimate adjusted hazard ratio (HRs) and associated 95% CIs between incident AF (time to event or censoring) with DHA%, total omega‐3% or other omega‐3%, per quintile using quintile 1 as reference, as well as (in separate models) per IQ₅R, and using the covariates noted above. Tests of nonlinearity for omega‐3 fatty acids were conducted using cubic splines. Stratified analyses by FOS use (yes/no) were conducted. Fixed‐effects meta‐analysis was used to combine UK Biobank results with a recently conducted meta‐analysis ² using the rma() function in Rb (R Foundation for Statistical Computing). In parallel models examining the association between self‐reported FOS use (yes/no) and incident AF, we considered different ways of adjusting for age: (1) continuously (linear); (2) continuously (cubic splines); (3) dichotomously (65+ versus <65 years); or (4) categorically (<45, 45–49, 50–54, 55–59, 60–64, 65+ years). We used sensitivity analyses to explore the use of different covariate sets and different dates for data censoring (in order to replicate the prior studies on this topic). Statistical significance was set to 0.05 for all analyses, except for tests of nonlinearity, which used 0.01 due to the exploratory and multiple‐testing nature of those tests, with 2‐sided tests used in all cases. Proportional hazards assumptions were confirmed using the Grambsch–Therneau method ¹³ (all P>0.01), through the inclusion of sex, ethnicity, and β‐blockers as strata in the models. R (www.r‐project.org) was used for all analyses including the coxme and metafor packages.

RESULTS

Associations of Serum Omega‐3 Fatty Acids With Incident AF

Among the 261 108 individuals with serum fatty acid data available in the UK Biobank, the mean age was 56.5±8.1 years, 95% were of White race, and 55% were women, with an average body mass index of 27.4±4.8 kg/m2 (Table 1). The mean±SD serum total omega‐3 level was 4.97±1.64% in the FOS users and 4.12±1.43 in the nonusers (P<0.0001). After a median follow‐up of 12.7 years, all 3 omega‐3 fatty acid biomarkers showed statistically significant inverse associations with incident AF risk (Table 2). When comparing quintile 5 with quintile 1 of serum total omega‐3, the adjusted risk for AF was 14% lower, for DHA was 8% lower, and for other omega‐3s was 16% lower. Absolute risk for AF in quintile 1 total omega‐3 was 7.5% and in quintile 5 was 6.9%. Similar inverse associations were seen in the per‐IQ₅R analyses. There was no evidence of nonlinearity in the biomarker relationships (P>0.01 in all cases). We found a significant interaction by FOS use versus nonuse on the relationship between serum omega‐3 and AF risk (P=0.03) and therefore conducted a stratified analysis. The HR (per IQ₅R of serum total omega‐3) for FOS users was 0.93 (95% CI, 0.87–0.99) and that for nonusers was 0.85 (95% CI, 0.81–0.90). Thus, in both groups, risk for AF was lower with higher omega‐3 levels, but the relationship was stronger in the nonusers.

Table 1.

Patient Characteristics (N=261 108)

Variable	Percentage (no.) or mean (SD)
Age, y	56.5 (8.1)
Female sex	54.6% (142 617)
Race or ethnicity
White	95.4% (249 021)
Asian	1.9% (5013)
Black	1.4% (3543)
Other *	1.4% (3531)
BMI, kg/m2	27.4 (4.8)
Education
College	60.2% (157 223)
High school	22.5% (58 737)
Less than high school	17.3% (45 148)
Exercise
Lowest quartile	22.3% (58 329)
Second lowest quartile	23.0% (59 970)
Second highest quartile	23.1% (60 344)
Highest quartile	23.3% (60 914)
Unavailable	8.3% (21 551)
Smoking status
Never	54.8% (143 156)
Previous	34.7% (90 516)
Current	10.5% (27 436)
Alcohol consumption
Rarely	30.2% (78 733)
1 or 2×/wk	26.2% (68 446)
3 or 4×/wk	23.4% (61 139)
Daily	20.2% (52 790)
Linoleic acid (% composition)	28.9 (3.4)
Nonlinoleic omega‐6s (% composition)	8.9 (1.9)
Taking β‐blockers	6.0% (15 637)
Treated for hypertension	20.4% (53 368)
Treated for high cholesterol	17.2% (44 966)
Diagnosed with diabetes	5.1% (13 317)
Diagnosed with cardiovascular disease	6.3% (16 469)
Diagnosed with HF	0.4% (995)

BMI indicates body mass index; and HF, heart failure.

^*Other in the ethnicity category is simply not Asian, Black or White. The possible subcategories of other for the UK are multiple.

Table 2.

Association of Plasma Omega‐3 Quintiles With Risk for Incident AF in the UK Biobank

Fatty acids	Median fatty acids level (%)	Sample size	No. of incident AF cases	Adjusted HR (95% CI)*
DHA
IQ5R	1.92	261 108	18 264	0.93 (0.89–0.97)
Quintile 1 (<1.46%)†	1.24	51 642	3888	1.00
Quintile 2 (1.46%<1.77%)	1.63	52 085	3647	0.95 (0.90–0.99)
Quintile 3 (1.77%<2.07%)	1.92	52 399	3582	0.93 (0.89–0.98)
Quintile 4 (2.07%<2.47%)	2.25	52 516	3504	0.91 (0.86–0.96)
Quintile 5 (>2.47%)	2.83	52 466	3643	0.92 (0.87–0.97)
Total omega‐3
IQ5R	4.17	261 108	18 264	0.89 (0.86–0.93)
Quintile 1 (<3.15%)	2.68	51 877	3457	1.00
Quintile 2 (3.15%–3.84%)	3.52	52 261	3558	0.93 (0.89–0.97)
Quintile 3 (3.84%–4.52%)	4.17	52 281	3711	0.90 (0.86–0.94)
Quintile 4 (4.52%–5.45%)	4.92	52 315	3709	0.86 (0.82–0.91)
Quintile 5 (>5.45%)	6.32	52 374	3829	0.86 (0.82–0.90)
Other omega‐3
IQ5R	2.30	261 108	18 264	0.86 (0.83–0.90)
Quintile 1 (<1.57%)	1.20	52 135	3223	1.00
Quintile 2 (1.57%–2.08%)	1.84	52 255	3519	0.92 (0.88–0.97)
Quintile 3 (2.08%–2.53%)	2.30	52 300	3785	0.90 (0.86–0.95)
Quintile 4 (2.53%–3.13%)	2.80	52 156	3817	0.85 (0.82–0.90)
Quintile 5 (>3.13%)	3.65	52 262	3920	0.84 (0.80–0.88)

AF indicates atrial fibrillation; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; HR, hazard ratio; and IQ₅R, interquintile range.

^*Adjusted for all variables in Table 1.

^† The corresponding estimated omega‐3 index (red blood cell EPA+DHA) values computed from nuclear magnetic resonance: DHA levels are 4.40%, 5.36%, 6.30%, and 7.54%, respectively.

Update of Qian et al Meta‐Analysis With UK Biobank

Qian et al ² analyzed 17 prospective cohort studies, encompassing 55 214 individuals, among whom there were 7720 incident cases of AF during a median follow‐up of 13.3 years. ² Since the current analysis used the same covariate set as Qian et al, ² we were able to harmonize and update the meta‐analysis with the present results from the UK Biobank (Table 3, ² , ¹² ) now including 316 322 persons. Originally, Qian et al ² found inverse relationships between DHA, DPA, and EPA + DHA and risk of AF. The findings were largely similar in the updated meta‐analysis in which we used the equations in Table S3 to predict plasma DPA, EPA, and EPA + DHA levels (by gas chromatography) from the UK Biobank NMR data (the correlation coefficients were 0.84 for EPA, 0.91 for EPA + DHA, and 0.66 for DPA). Compared with the original Qian meta‐analysis, the only difference in the updated analysis was for EPA. Qian et al ² found no association between blood EPA and incident AF, while the UK Biobank analysis showed a significant inverse association. The inclusion of UK Biobank data did not substantially change the observed heterogeneities in the meta‐analysis (52.2%–59.7%).

Table 3.

Meta‐Analysis From Qian et al ² Updated With UK Biobank Findings

Fatty acid	Analysis	Adjusted HR per IQ5R*	Heterogeneity I 2 (P value)
DHA	Qian et al 2 meta‐analysis	0.90 (0.85–0.96) P<0.01	47.5% (P=0.016)
	UK Biobank alone	0.93 (0.89–0.97) P<0.001	…
	Updated meta‐analysis†	0.92 (0.89–0.96) P<0.001	43.4% (P=0.026)
EPA‡	Qian et al 2 meta‐analysis	1.00 (0.95–1.05)	52.2% (P=0.008)
	UK Biobank alone	0.92 (0.88–0.95) P<0.001	…
	Updated meta‐analysis†	0.94 (0.92–0.97) P<0.001	59.7% (P=0.0009)
DPA‡	Qian et al 2 meta‐analysis	0.89 (0.83–0.95) P<0.001	0.0% (P=0.51)
	UK Biobank alone	0.91 (0.87–0.94) P<0.001	…
	Updated meta‐analysis†	0.90 (0.88–0.93) P<0.001	0.0% (P=0.56)
EPA+DHA‡	Qian et al 2 meta‐analysis	0.93 (0.87–0.99) P<0.05	60.7% (P=0.001)
	UK Biobank alone	0.92 (0.88–0.95) P<0.001	…
	Updated meta‐analysis†	0.92 (0.90–0.95) P<0.001	58.5% (P=0.0013)

^*Adjusted for all variables in Table 1, which identically matches the de novo analysis presented in Qian et al. ²

^† Weight of the UK Biobank sample ranged from 67.3% to 71.7% in the updated meta‐analysis based on its large sample size.

^‡ Adjusted HRs for these fatty acids were computed based on their estimated values in the UK Biobank sample using the method of Schuchardt et al ¹² and the equations shown in Table S3.

DHA indicates docosahexaenoic acid; DPA, docosapentaenoic acid; EPA, eicosapentaenoic acid; HR, hazard ratio; and IQ₅R, interquintile range.

Reevaluation of Prior Findings on FOS Use and Risk of AF in the UK Biobank

We were able to replicate the findings of 2 recent papers on the risk of incident AF and self‐reported FOS use in the UK Biobank, but only by adjusting for age as a dichotomous variable. When age was treated as a continuous variable, the association between FOS use and incident AF was lost (adjusted HR, 1.00 [95% CI, 0.97–1.02]) (Table 4, ⁹ ). Other approaches of adjusting for age‐related risk (splines, more age categories) yielded findings similar to those noted when age was adjusted for as a continuous variable (Table 4). In addition, models adjusting for age in a more continuous manner showed improved concordance (ie, predictive ability) compared with those using dichotomous age (P<0.0001), providing additional confirmation of the superiority of more granular age adjustments. This approach was further confirmed in our sensitivity analyses (Table 4), which considered different follow‐up periods and different covariate sets (see Table S4 for distributions of these variables).

Table 4.

Reevaluation of the Association Between FOS Use and Risk for AF in Zhang et al ⁹

	Sample size	Fish oil users	Total no. incident AF	Type of age adjustment	Adjusted HR*	Adjusted HR†	Model concordance
Previously published results‡	468 665	148 192	25 748	None/dichotomous	1.22 (1.19–1.25)§	1.10 (1.07–1.13) 5
Replication attempts—data censored at September 1, 2020||	454 568	143 736	26 667	None	1.20 (1.17 –1.23)	…
				Dichotomous	1.05 (1.02–1.08)	1.09 (1.06–1.12)	0.746
				Continuous—linear	0.92 (0.90–0.95)	0.99 (0.97–1.02)	0.773
				Continuous—spline	0.92 (0.90–0.95)	0.99 (0.97–1.02)	0.773
				5‐y age category	0.93 (0.91–0.95)	1.00 (0.97–1.02)	0.770
Replication attempts—data censored at December 31, 2022#	454 568	143 736	32 162	No age adjustment	1.21 (1.18–1.24)
				Dichotomous	1.06 (1.03–1.08)	1.10 (1.07–1.12)	0.742
				Continuous—linear	0.93 (0.91–0.95)	1.00 (0.97–1.02)	0.770
				Continuous—spline	0.93 (0.91–0.95)	1.00 (0.98–1.02)	0.770
				5‐y age category	0.94 (0.92–0.96)	1.00 (0.98–1.03)	0.768

AF indicates atrial fibrillation; FOS, fish oil supplement; and HR, hazard ratio.

^*Unadjusted or adjusted for age only—either dichotomous (65+ years vs <65 years), continuous (linear), 5‐year age categories (<45, 45–49, 50–54, 55–60, 60–65, 65+ years) or cubic spline.

^† Adjusted for age plus all variables listed in Table S4, which matches Zhang et al. ⁹

^‡ Results previously published in Zhang et al. ⁹

^§ Zhang reports an unadjusted model (no age adjustment), which we attempted to replicate.

^|| Original censoring date used in Zhang. Slight differences in sample size likely attributable to individuals who withdrew consent.

^# Relationship between FOS use and AF comparing Zhang’s analysis (age adjustment dichotomous) and ours with continuous age adjustment using longer follow‐up time (ie, more cases) than was available for Zhang.

DISCUSSION

The 3 principal findings of this study were: (1) higher circulating levels of omega‐3s were associated with reduced risk for AF in the UK Biobank, in both FOS users and nonusers; (2) when added to a previous meta‐analysis by Qian et al, ² their conclusion that higher omega‐3 levels were linked with lower risk for AF was strengthened; and (3) earlier reports from 2 prior UK Biobank studies that FOS users were at increased risk for AF were nullified when age was adjusted for continuously rather than dichotomously.

Fully adjusted risk reductions when comparing the highest with the lowest quintiles for total omega‐3, other omega‐3, and DHA were 14%, 16%, and 8%, respectively. These findings largely paralleled those from a recent meta‐analysis by Qian including ≈55 000 individuals who demonstrated a 12% decreased risk of AF for patients in the top quintile of DHA + EPA blood levels compared with the bottom quintile. ² After the present data were added into the aforementioned meta‐analysis, higher blood levels of DHA, EPA, DPA, and EPA + DHA were associated with a 6% to 10% lower risk of AF.

Previous meta‐analyses of observational studies that have focused on biomarkers instead of estimated intakes have shown that higher levels of marine omega‐3s, which are reflective of chronic dietary intake, correlate with lower risks of all‐cause and cardiovascular mortality, ³ , ¹⁴ stroke, ⁵ and most recently, AF. ² The Qian study, along with the findings presented here, sharply contrast with 2 previous reports, ⁸ , ⁹ which were both from the same research group, both using the same UK Biobank database, and both concluding that FOS use may increase the risk of AF. Since FOS use is directly related to serum omega‐3 fatty acid levels, ¹² the findings of these 2 previous studies are at odds with our observations. To explore this apparent contradiction, we undertook a reanalysis of the same UK Biobank data used by Zhang et al ⁹ and Chen et al. ⁸ In both of these studies, age was adjusted for as a dichotomous, not a continuous, variable. Since the incidence of AF and the use of FOSs both increase linearly with age in the UK Biobank (Figure 1), a simple dichotomous adjustment for age is insufficient. When age was treated as a continuous variable the apparent association between FOS use and AF disappeared.

Figure 1. Reported use of fish oil supplements at baseline and the incidence of atrial fibrillation by age (years) in the UK Biobank.

In a recent meta‐analysis of randomized controlled trials testing the effects of pharmaceutical omega‐3 products in 83 112 individuals, there was a 24% increase in relative risk of AF (P=0.0002). ⁷ This relationship appeared to be dose‐dependent, with a 12% higher relative risk of AF in the 5 trials testing <1000 mg/d of DHA + EPA, versus a 51% relative risk increase of AF in the 3 trials that used 1.8 to 4.0 g/d of DHA and/or EPA. The weighted absolute increase in risk of AF in these studies was 0.67% overall, 0.46% for lower‐dose omega‐3 trials, and 1.18% for the high‐dose trials (Table S5 including a new report ¹⁵ not included in the meta‐analysis by Jia et al ⁷ ). These relative risk increases have been questioned, however, by Samuel and Nattel, who noted that since the risk of dying in all but one of these trials was 3 to 5 times higher than the risk of developing AF, and because the individuals randomized to omega‐3 had a significantly lower risk of dying compared with placebo‐treated patients, a failure to adjust for competing risks may have inflated the relative risk estimates. ¹⁶

These apparently contradictory findings regarding AF risk between randomized trials and biomarker‐ and FOS use–based observational studies is a currently unresolved issue. Adding to the confusion is the fact that 20 years ago, there was sufficient evidence to mount a major trial in 663 patients with AF in which high‐dose omega‐3 (4–8 g/d) was given to reduce recurrent AF episodes. ¹⁷ The hypothesis was not supported by the data but risk for recurrent AF events was not increased by treatment. Thus, significant uncertainty remains about the role of omega‐3 and AF. New randomized controlled trials designed to address this important question are needed.

Mechanistically, a variety of possibilities have been suggested as to why higher omega‐3 fatty acid levels might be associated with susceptibility to AF. As omega‐3 fatty acids become incorporated into myocardial membranes, they may alter the conformation of Piezo‐type mechanosensitive ion channel components 1 and 2. ¹⁸ These membrane‐bound pressure‐sensing proteins may be activated when high levels of omega‐3s are in the membranes and become arrhythmic triggers. On the other hand, low levels of omega‐3s can predispose to heart failure, hypertension, atherosclerotic cardiovascular disease and other cardiac conditions that increase risk for AF. ¹⁹ Intermediate omega‐3 levels, such as those found in individuals consuming fish/seafood and/or low to moderate doses of FOSs, but not high‐dose pharmaceutical products, may be protective, resulting in a potential U‐shaped relationship between omega‐3 status and AF risk (Figure 2, ¹⁸ ).

Figure 2. U‐shaped relationship between DHA plus EPA intake and risk of AF.

The chronic risk of AF appears to be higher for individuals who consume low amounts of EPA+DHA (<250 mg/d) and large amounts (>1500 mg/d). The AF risk is lowest for those who consume moderate amounts of EPA+DHA (250–1500 mg/d). Figure adapted from Fatkin et al. ¹⁸ AF indicates atrial fibrillation; ASCVD, atherosclerotic cardiovascular disease; DHA, docosahexaenoic acid; EPA, eicosapentaenoic acid; and HF, heart failure.

Other possible explanations relate to vagal tone. ²⁰ Studies have consistently shown that omega‐3s produce a dose‐dependent increase in vagal tone that becomes apparent at relatively low doses of EPA + DHA (≈500 mg/d). ²¹ , ²² , ²³ Low‐level vagal nerve stimulation is antiarrhythmic and associated with a reduced risk of AF. ²⁴ In contrast, high‐level vagal nerve stimulation is an effective way to induce AF in animal models. ²⁵ This phenomenon suggests another possible explanation of how lower omega‐3 intake (ie, from diets/supplements) correlates with reduced risk of AF, whereas higher‐dose (pharmaceutical) omega‐3 use appears to be associated with increased risk of AF.

Prior dietary questionnaire‐based epidemiological studies have suggested that an intake of ≈500 to 700 mg/d of DHA + EPA may be the ideal dose for decreasing risk of AF. ²⁶ A large observational study found that consumption levels less than this were associated with ≈10% higher risk of AF, ²⁶ whereas omega‐3 intakes >1250 mg/d are also linked with increased risks of AF. ²⁷ The typical adult American eats <1 serving per week of fish or seafood, so the average intake of DHA + EPA in the United States is ≈100 mg/d. ²⁸ , ²⁹ This corresponds to an average DHA + EPA level in red blood cells (ie, the omega‐3 index) in the United States of ≈5.4%. ³⁰ The estimated median omega‐3 index in quintile 5 in the present study was 8.62%. The omega‐3 index level that is optimal for minimizing risk of major adverse cardiovascular events, stroke, and all‐cause mortality is ≥8%; to achieve this in the average adult American would require an increase of ≈1500 mg/d of DHA + EPA. ¹⁴ , ²⁸

However, for individuals at high risk for AF or those with a history of AF, a lower target of ≈500 to 700 mg/d of DHA + EPA, preferably from fish/seafood, appears to be a safer level of intake. ³¹ The 2021 American Heart Association (AHA) dietary guidance for improving cardiovascular health advocates for a diet that prioritizes the consumption of vegetables, fruits, and fish/seafood as a preferred source of protein. The AHA also specifically recommends at least 2 meals per week that feature fish or seafood, ³² which, in the case of salmon, would supply ≈300 mg of EPA + DHA per day.

Finally, it is of interest to note that the omega‐3 index in Japan and South Korea averages >8% compared with ≈5.5% in the United States, ³³ yet the age‐adjusted AF prevalence per 100 000 people in 2019 in Japan and Korea was 200 and 399, respectively, compared with >1200 in the United States. This ecological finding is at least consistent with the results of the current observational study: higher dietary intakes (thus blood levels) of omega‐3 are associated with lower risks of AF, although obviously causal inferences cannot be drawn.

Strengths and Limitations

One strength of this study was the use of objective measurements of omega‐3 serum levels rather than relying on dietary questionnaires. Furthermore, the large sample of >261 000 participants followed for 12 years gives us sufficient statistical power to undertake subgroup analyses. A limitation of the UK Biobank is the relative lack of ethnic diversity, as 95% are of White race. However, the UK Biobank data harmonized strongly with a recent meta‐analysis of 17 international cohorts evaluating omega‐3 blood levels and risk of AF, ² thus our findings are likely to be generalizable. DHA and other omega‐3 fatty acids and all covariates were measured only at baseline; however, omega‐3 blood levels have shown good reproducibility over time. ³⁴ Although the UK Biobank cohort is generally healthier than the UK population, the biomarker–disease associations observed in the UK Biobank cohort are considered to be generalizable. ³⁵ Because this is an observational study, causation cannot be established. Even though we made statistical adjustments for many relevant risk factors (eg, age, sex, occupation, education, physical activity, smoking, hyperlipidemia, and hypertension), residual confounding from other unmeasured variables is always a possibility.

CONCLUSIONS

In agreement with recent biomarker‐based meta‐analyses, higher circulating blood levels of omega‐3 were associated with reduced risk of AF in the UK Biobank. In addition, this study reassessed the relationship between FOS use and risk of AF in the UK Biobank and found no evidence of an association between FOS use and AF. Although virtually all observational data on omega‐3 dietary intake/FOS use or blood levels suggest that higher intakes/levels reduce the risk of AF, new randomized controlled trials will be needed to clarify the disparity between the results of current trials and these observational studies.

Sources of Funding

Evan O’Keefe: The National Heart, Lung, and Blood Institute of the National Institutes of Health under award number T32HL110837 supported the research reported in this publication. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Health. This project was partially supported by the Richard Galamba Foundation.

Disclosures

Dr O’Keefe is the Chief Medical Officer of Cardiotabs, a company that sells omega‐3 products. WS Harris owns stock in OmegaQuant Analytics, LLC, a laboratory that offers blood fatty acid testing for researchers, clinicians, and consumers.

Supporting information

Tables S1–S5

References 36, 37, 38, 39, 40, 41, 42, 43