Plasma Phospholipid Long-Chain ω-3 Fatty Acids and Total and Cause-Specific Mortality in Older Adults: A Cohort Study
Download the PDF here|
Ann Intern Med. 2 April 2013
Dariush Mozaffarian, MD, DrPH; Rozenn N. Lemaitre, PhD, MPH; Irena B. King, PhD; Xiaoling Song, PhD; Hongyan Huang, PhD; Frank M. Sacks, MD; Eric B. Rimm, ScD; Molin Wang, PhD; and David S. Siscovick, MD, MPH
Background: Long-chain ω-3 polyunsaturated fatty acids (ω3-PUFAs), including eicosapentaenoic acid (EPA) (20:5ω-3), docosapentaenoic acid (DPA) (22:5ω-3), and docosahexaenoic acid (DHA) (22:6ω-3), have been shown to reduce cardiovascular risk, but effects on cause-specific and total mortality and potential dose-responses remain controversial. Most observational studies have assessed self-reported dietary intake and most randomized trials have tested effects of adding supplements to dietary intake and evaluated secondary prevention, thus limiting inference for dietary ω3-PUFAs or primary prevention.
Objective: To investigate associations of plasma phospholipid EPA, DPA, DHA, and total ω3-PUFA levels with total and cause-specific mortality among healthy older adults not receiving supplements.
Design: Prospective cohort study.
Setting: 4 U.S. communities.
Participants: 2692 U.S. adults aged 74 years (±5 years) without prevalent coronary heart disease (CHD), stroke, or heart failure at baseline.
Measurements: Phospholipid fatty acid levels and cardiovascular risk factors were measured in 1992. Relationships with total and cause-specific mortality and incident fatal or nonfatal CHD and stroke through 2008 were assessed.
Results: During 30 829 person-years, 1625 deaths (including 570 cardiovascular deaths), 359 fatal and 371 nonfatal CHD events, and 130 fatal and 276 nonfatal strokes occurred. After adjustment, higher plasma levels of ω3-PUFA biomarkers were associated with lower total mortality, with extreme-quintile hazard ratios of 0.83 for EPA (95% CI, 0.71 to 0.98; P for trend = 0.005), 0.77 for DPA (CI, 0.66 to 0.90; P for trend = 0.008), 0.80 for DHA (CI, 0.67 to 0.94; P for trend = 0.006), and 0.73 for total ω3-PUFAs (CI, 0.61 to 0.86; P for trend < 0.001). Lower risk was largely attributable to fewer cardiovascular than noncardiovascular deaths. Individuals in the highest quintile of phospholipid ω3-PUFA level lived an average of 2.22 more years (CI, 0.75 to 3.13 years) after age 65 years than did those in the lowest quintile.
Limitation: Temporal changes in fatty acid levels and misclassification of causes of death may have resulted in underestimated associations, and unmeasured or imperfectly measured covariates may have caused residual confounding.
Conclusion: Higher circulating individual and total ω3-PUFA levels are associated with lower total mortality, especially CHD death, in older adults.
Primary Funding Source: National Institutes of Health.
Context The effects of dietary long-chain ω-3 polyunsaturated fatty acids (ω3-PUFAs) on total and cause-specific mortality are uncertain, as are the potential benefits for the primary prevention of cardiovascular disease.
In this cohort of individuals aged 65 years or older without known cardiovascular disease at baseline, higher baseline levels of specific individual and total ω3-PUFAs were associated with decreased total mortality, primarily due to fewer cardiovascular events.
Caution Fatty acid levels were obtained at baseline.
Implication Further study is required to evaluate whether certain ω3-PUFAs are beneficial for the primary prevention of cardiovascular disease and decreased mortality.
Experiments and clinical studies have shown physiologic benefits of long-chain ω-3 polyunsaturated fatty acids (ω3-PUFAs), which include eicosapentaenoic acid (EPA) (20:5ω-3), docosapentaenoic acid (DPA) (22:5ω-3), and docosahexaenoic acid (DHA) (22:6ω-3) (1). Yet, although observational studies have found inverse associations between dietary ω3-PUFA level and death from coronary heart disease (CHD) (1 - 2), randomized trials of ω3-PUFA supplementation have had mixed results (3).
Consequently, effects of ω3-PUFAs on cardiovascular disease (CVD) and total and cause-specific mortality remain controversial. Understanding the influence of ω3-PUFAs on CVD and mortality; whether such effects vary for EPA, DPA, or DHA; and their potential dose-response is crucial for scientific advancement and dietary guidance.
Most observational studies of ω3-PUFAs have assessed self-reported dietary intake rather than objective biomarkers, which may have led to measurement errors or bias. Conversely, most randomized trials have tested the effects of ω3-PUFA supplements among patients with established CVD or multiple risk factors, thus limiting inference for primary prevention. In addition, the trials evaluated ω3-PUFA supplements that were added to background dietary intake, which could reduce efficacy if the dose-response for ω3-PUFAs is nonlinear.
In particular, a potential threshold effect (4 - 5) could explain why moderate consumption is associated with benefits when compared with little or no consumption in observational studies, whereas adding higher supplement doses to already moderate background dietary intake produces smaller or no effects in trials. Differences could also be due to stronger effects of ω3-PUFAs on CHD death, which is often evaluated in observational studies (4 - 5), versus composite end points of total CHD or total CVD events in trials.
Whether potential cardiovascular benefits of ω3-PUFAs translate into lower total mortality or whether ω3-PUFAs influence noncardiovascular causes of death is also unclear. Competing risks from noncardiovascular conditions (for example, cancer or lung disease) may be unaffected by ω3-PUFAs (6), thus minimizing effects on total mortality, particularly later in life. In meta-analyses of trials, ω3-PUFA supplementation produced non-statistically significant trends toward lower total mortality (3). However, these trials typically evaluated higher-dose fish oil supplements in high-risk patients, many of whom were already consuming fish. Several prospective cohort studies of generally healthy populations have found non-statistically significant inverse trends between self-reported dietary ω3-PUFA intake and total mortality (7 - 9). Self-reported diet also cannot reliably distinguish among specific long-chain ω3-PUFAs (EPA, DPA, or DHA), which may partially differ in physiologic effects (10).
Circulating ω3-PUFA biomarkers objectively reflect dietary consumption and biologically relevant processes (for example, absorption, incorporation, or metabolism) that influence tissue levels. Metabolic influences seem especially relevant for DPA, which is elongated from and retroconverted to EPA (10). Biomarkers also permit direct evaluation of individual ω-3 fatty acids, which may have different effects on certain biological pathways or clinical end points (10). However, to our knowledge no prior studies have evaluated how circulating ω3-PUFA biomarkers relate to total mortality and diverse CVD subtypes in generally healthy populations.
To address these gaps, we prospectively designed and implemented an investigation of ω3-PUFA biomarkers, including EPA, DPA, and DHA, and risk for CVD (CHD or stroke) and total and cause-specific mortality in a large, community-based cohort of older U.S. adults. We hypothesized, on the basis of mechanistic studies and physiologic effects (1,10), that ω3-PUFA levels would be associated with decreases in cardiovascular mortality (especially CHD death) but not noncardiovascular mortality. We also hypothesized that among individual ω3-PUFAs, DHA would be most strongly associated with arrhythmic CHD death, and EPA and DPA would be most strongly associated with nonfatal CHD.
In this prospective study of older adults, circulating individual and total ω3-PUFA levels were associated with lower total mortality, with a 27% lower risk across total ω3-PUFA quintiles. Associations seemed strongest for cardiovascular deaths, especially arrhythmic CHD deaths, with nearly 50% lower risk across quintiles. The observed mortality differences corresponded to approximately 2.2 more years of remaining life after age 65 years in persons with higher ω3-PUFA levels than in those with lower levels.
Because these biomarkers were measured among older adults, our findings suggest that dietary ω3-PUFAs late in life may reduce total mortality. Alternatively, these associations could reflect an influence of life-long dietary habits. Specificity for CVD events, especially arrhythmic CHD death, and magnitudes of the latter association argue against residual confounding as the sole explanation for our results. Cardiovascular benefits of ω3-PUFAs are supported by in vitro studies, animal models, and placebo-controlled trials showing physiologic benefits (1). Effects include reduced heart rate, lower blood pressure, improved myocardial efficiency and diastolic function, and lower hepatic triglyceride production and may also include improved autonomic and endothelial function, antithrombotic effects, and antiarrhythmic effects (1). In addition, ω3-PUFAs are precursors to recently identified resolvins, protectins, maresins, and monoepoxides, which are synthesized by cyclooxygenase, lipoxygenase, and cytochrome-P450 pathways and seem crucial for restoring homeostasis after tissue injury or inflammation (32 - 33). Although many of these physiologic effects are modest, their combined benefits could plausibly reduce mortality, especially deaths related to CVD.
Although associations of circulating DPA and DHA levels with mortality seemed relatively linear, relationships of dietary versus circulating ω3-PUFA levels did not, with steepest dose-responses up to consumption of approximately 400 mg/d. Other circulating nutrients show similar dietary dose-responses, with concentrations increasing steeply at lower consumption levels and relatively saturating thereafter (34 - 35). A meta-analysis of cohort studies and randomized trials found a statistically significant, nonlinear threshold relationship between dietary ω3-PUFA level and CHD mortality, with greatest benefits up to consumption of approximately 250 mg/d (5). In light of these prior studies, the present findings for ω3-PUFA biomarkers suggest that circulating ω3-PUFAs (especially DHA) may linearly reduce CHD death within ranges determined by dietary intake and that previously observed nonlinear (threshold) relations of dietary ω3-PUFA level with CHD death may partly relate to a nonlinear dose-response of circulating levels to dietary consumption. The observed dose-response between dietary and circulating ω3-PUFAs represents an average, and individual variation will exist.
Nonetheless, the present findings support an average target dietary range of 250 to 400 mg of EPA plus DHA per day. Relatively few interventions substantially alter total mortality later in life, and these results highlight potential benefits of modest ω3-PUFA consumption, compared with little or none, for primary prevention in older adults.
In contrast to self-reported diet, circulating biomarkers provide objective measures of exposure, allow evaluation of individual fatty acids, and account for potential nondietary processes that might influence disease risk. Nondietary processes might be most relevant for DPA, which was not correlated with dietary fish intake and at least partly derives from metabolic interconversion with EPA (10). Conversely, diet clearly influences circulating EPA and DHA levels, which were correlated with fish consumption and each other.
Adjustment for self-reported fish consumption did not substantially alter the results, and this is concordant with prior analyses of circulating ω3-PUFA biomarkers and CVD outcomes (11,36 - 37). If circulating ω3-PUFA levels are a key causal mediator of cardiovascular effects of fish consumption, self-reported fish consumption and its correlates should not confound the associations. In addition, these results might suggest that other nondietary metabolic influences on circulating ω3-PUFA levels are also relevant for disease risk.
A strength of this investigation was its ability to evaluate each long-chain ω3-PUFA separately. Docosahexaenoic acid was most strongly associated with fatal CHD and arrhythmic CHD death. In light of known higher myocardial concentrations of DHA (38) and prior studies showing inverse associations of circulating DHA (but not EPA or DPA) with incident atrial fibrillation (37,39), our results suggest that DHA might be especially relevant for cardiac arrhythmias (10). Conversely, only EPA was statistically significantly associated with nonfatal MI. In a large randomized trial, treatment consisting of EPA combined with a statin reduced nonfatal coronary events compared with statin treatment alone (40), and recent prospective studies found that circulating EPA or DPA level was more strongly associated with nonfatal cardiac outcomes than was DHA level (11,31). In the present investigation, the mutually adjusted results support greater specificity of DHA level for fatal CHD and of EPA level for nonfatal MI. Yet, circulating concentrations of these fatty acids are causally interrelated due to common dietary sources, metabolic interconversion, or both, so biological relevance of mutually adjusted results should be interpreted cautiously. We also found that DHA, but not EPA, was associated with less ischemic stroke-an intriguing finding, given experimental studies suggesting that DHA reduces hypoxic brain injury and apoptosis (41).
However, this association was no longer statistically significant after multivariate measurement error correction, and in randomized trials with durations ranging from approximately 1 to 5 years, fish oil supplements have not reduced stroke (3). Our overall findings, together with other mechanistic studies, support the hypothesis that each long-chain ω3-PUFA may have partially differing, complementary effects on pathways of cardiovascular risk (10).
Our findings do not support major effects of circulating ω3-PUFA levels on mortality from non-CVD conditions later in life. Evidence for effects of ω3-PUFA on risk for cancer, dementia, or chronic inflammatory conditions has been inconsistent (42 - 44). The present results do not exclude potential benefits on incidence or severity of these conditions or on mortality due to more specific subtypes of these diseases. The observed inverse association of DPA level with cancer mortality warrants further study; higher DPA levels could have independent beneficial effects or be a marker for healthier underlying physiology and metabolism. The observed lower risk for death from infection was unexpected but supported by protective effects against infection in animal studies (45 - 46) and beneficial effects in some but not all trials of ω3-PUFAs in severe acute lung injury (47 - 49). Our results support the need for evaluation of ω3-PUFAs in less severe infections, such as community-acquired pneumonia, in older adults. Because of the absence of a priori hypotheses related to cancer or infection in this analysis, these findings should be considered exploratory.
Few observational studies have evaluated fish or ω3-PUFA consumption and total mortality in generally healthy populations (7 - 9). Most found only non-statistically significant inverse trends and may have been limited by smaller numbers of events or reliance on self-reported diet. Several but not all prior reports (50) found inverse associations of ω3-PUFA biomarkers with total mortality in patients with CHD (51 - 53) and in hospitalized patients (54); associations with cause-specific mortality were generally not reported. To our knowledge, no prior study has evaluated how objectively measured ω3-PUFA biomarkers relate to total mortality in generally healthy populations, such as ours.
Four large randomized trials among patients who either have or are at high risk for CVD showed that fish consumption or fish oil supplementation reduced coronary events (3). However, more recent trials have not confirmed these findings (3). A meta-analysis found that ω3-PUFA supplementation reduced cardiac death (relative risk, 0.91 [CI, 0.85 to 0.98]) but not all-cause mortality (relative risk, 0.96 [CI, 0.91 to 1.02]) (3). Such meta-analyses have not accounted for differences in background dietary fish consumption or potential nonlinear effects of supplemental ω3-PUFAs. Our dose-response analysis of diet and circulating levels suggests that dietary or supplemental ω3-PUFAs may be most beneficial for persons with little or no consumption.
No controlled trials have reported effects of ω3-PUFAs on total mortality in generally healthy populations; 1 primary prevention trial is enrolling (55).
Adding a supplement to background consumption of approximately 150 mg of EPA plus DHA per day (the approximate mean consumption in the United States and many European countries) would be calculated, on the basis of nonlinear effects in observational studies (5), to produce a reduction in CHD mortality of approximately 15% (CI, 8% to 21%). Such a modest effect, while clinically revelant, may be challenging to detect in new trials. In comparison, increasing from a baseline of low intake to at least moderate consumption (>250 mg/d)-or similarly, as in our present investigation, comparing low and high circulating levels-would be predicted to have larger benefits, consistent with our observations. In our analysis, we evaluated biomarkers of ω3-PUFAs, measured late in life, that would be generally derived from dietary seafood and perhaps partly from endogenous metabolism (for example, DPA) rather than from supplements. Ranges of dietary exposure and absolute levels in the reference group were also generally much lower than would be seen in a supplement trial.
Our analysis has strengths. Information on demographic characteristics, risk factors, and lifestyle was prospectively collected in a well-established multicenter cohort with little loss to follow-up. We adjusted for multiple covariates, thus minimizing confounding. The cohort focused on older adults, in whom mortality and competing causes of death are common. Circulating biomarkers provided objective measures of individual fatty acids. Total and cause-specific mortality were prospectively adjudicated using medical records, and large numbers of events provided statistical power. Population-based enrollment from several U.S. communities increased generalizability. In contrast to randomized trials, our investigation allowed evaluation of generally healthy adults, a larger number of mortality and CVD events, ω3-PUFA exposures related to usual dietary habits rather than supplements, different ω-3 fatty acids separately, and a wide range of dose-response (very low to high).
Our study also has limitations. Fatty acid levels were measured at baseline, and dietary and metabolic fluctuations over time would increase exposure misclassification during follow-up, thus causing underestimation of true relationships with mortality. Although events were centrally adjudicated, some deaths may have been misclassified; such errors would probably be random with respect to baseline circulating ω3-PUFA levels, again causing attenuation of true relationships. The cohort included older men and women, and results may not be generalizable to younger populations. Relatively few hemorrhagic strokes occurred, limiting statistical power for this end point. The observational design precludes the exclusion of residual confounding by unknown or unmeasured factors. Yet, results were robust to adjustment for multiple major risk factors.
Also, varying relationships were present between each fatty acid and different potential confounders. For example, DPA was unassociated with education or fish intake, limiting potential confounding for this fatty acid due to these factors or their correlates.
In summary, our findings suggest that circulating ω3-PUFA levels are linked to lower total mortality among generally healthy adults later in life, with the potentially greatest associations with cardiovascular events and, especially, arrhythmic cardiac death.
At baseline, 63.7% of participants were women, and the mean age was 74 years. Most (87.8%) were white; approximately 1 in 8 (11.7%) were African American. In unadjusted comparisons, plasma phospholipid EPA, DPA, and DHA levels had dissimilar relationships with several baseline characteristics that might be key confounders, such as age, sex, race, education, and alcohol use (Appendix Table 1). As seen in other cohorts (31), fish consumption was associated with levels of EPA and DHA, but not DPA. Levels of EPA and DHA (Spearman r = 0.43) and EPA and DPA (Spearman r = 0.51) were modestly intercorrelated; levels of DPA and DHA were less so (Spearman r = 0.13).
During 30 829 person-years, 1625 deaths occurred (5.3/100 person-years). After adjustment for demographic, cardiovascular, lifestyle, and dietary factors (including fish intake), both individual and combined levels of EPA, DPA, and DHA were associated with lower total mortality (Table 1). Across quintiles, individuals with higher EPA, DPA, and DHA levels had 17%, 23%, and 20% lower risk, respectively (P for trend = 0.005, 0.008, and 0.006, respectively), and those with higher total ω3-PUFA levels had 27% lower risk (P for trend < 0.001). Further adjustment for other dietary factors or use of aspirin, lipid-lowering drugs, or other medications had no appreciable effects (data not shown).
For cause-specific deaths, all 3 ω3-PUFAs were associated with lower CVD mortality and their combined levels were associated with 35% lower risk across quintiles (P for trend < 0.001) (Table 2). Among CVD subtypes, DHA seemed most strongly related to CHD death (40% lower risk), especially arrhythmic CHD death (45% lower risk), whereas DPA was most strongly related to stroke death (47% lower risk).
As hypothesized, ω3-PUFA concentrations were generally unassociated with non-CVD mortality (Appendix Table 2). Exceptions included inverse associations between DPA level and cancer mortality (P for trend = 0.032) and between total ω3-PUFA level and deaths from infection (P for trend = 0.010).
Levels of EPA, DHA, and total ω3-PUFA were each associated with lower incidence of total (fatal plus nonfatal) CHD (Table 2). For both DHA and total ω3-PUFA levels, this seemed to be predominantly driven by lower risk for fatal CHD. Neither EPA nor DPA was statistically significantly associated with fatal CHD, and DPA and DHA were not associated with nonfatal myocardial infarction (MI). Non-statistically significant trends toward modestly lower risk could not be excluded. Levels of DHA and total ω3-PUFA showed nominal inverse associations with incident ischemic stroke. Statistically significant associations were not seen for total or hemorrhagic stroke.
In semiparametric analyses, associations of circulating EPA, DPA, and DHA levels with total mortality seemed generally linear (Figure 1). A possible threshold effect for EPA was visually suggested but not statistically significant (P for nonlinearity = 0.142). To understand how diet was related to circulating biomarker levels, we evaluated the dose-response relation between estimated dietary EPA plus DHA consumption and phospholipid EPA plus DHA level (Figure 2). The association was strongly nonlinear (P for nonlinearity < 0.001), with steepest dose-responses up to dietary intakes of about 400 mg/d and smaller increases in circulating levels thereafter.
Relations of ω3-PUFA level with mortality and CVD were similar when deaths were excluded during the first 2 years or censored at mid-follow-up (data not shown). Adjustment for regression dilution bias in ω3-PUFA level strengthened all risk estimates and widened CIs (Appendix Table 3). After additional multivariate measurement error correction for covariates, associations of EPA, DPA, DHA, and total ω3-PUFA levels with total mortality, CVD mortality, and CHD mortality were each strengthened (Appendix Table 4). For arrhythmic CHD death, 69% lower risk was evident across total ω3-PUFA quintiles (multivariate measurement error-corrected hazard ratio, 0.31 [95% CI, 0.12 to 0.78]; P for trend = 0.009). In comparison, for nonarrhythmic CHD death, the corresponding hazard ratio was 0.60 (CI, 0.22 to 1.59; P for trend = 0.134). After multivariate measurement error correction, the magnitude of association of total ω3-PUFA level with ischemic stroke was unchanged and, due to greater uncertainty, no longer statistically significant.
Simultaneous adjustment for EPA, DPA, and DHA levels attenuated the inverse associations of DPA and DHA levels with total mortality and the association of EPA level with nonfatal MI (Appendix Table 5). The association between DHA level and CHD mortality was not substantially altered by adjustment for EPA and DPA levels, whereas the association between EPA level and total mortality was no longer present after adjustment for DPA and DHA levels. There was little evidence that relationships of EPA, DPA, or DHA levels with total mortality varied by age, sex, or education (Bonferroni-corrected P > 0.0056 for each).
To inform potential personal and public health relevance of these associations, we calculated the multivariate-adjusted differences in remaining years of life after age 65 years among persons with higher or lower ω3-PUFA levels. Individuals with higher levels had significantly greater longevity after age 65 years than those with lower levels (Table 3). For total ω3-PUFA level, representative persons in the highest quintile lived an average of 2.22 more years (CI, 0.75 to 3.13 years) after age 65 years. Differences in life expectancy for other representative persons with varying baseline characteristics were similar (Table 4).
Design and Population
The CHS (Cardiovascular Health Study) is a multicenter prospective cohort of older U.S. adults. In 1989-1990, 5201 ambulatory, noninstitutionalized adults aged 65 years or older were randomly selected and enrolled from Medicare eligibility lists in 4 communities: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Allegheny County, Pennsylvania. An additional 687 black participants were similarly recruited and enrolled in 1992-1993. Among all eligible adults contacted, 57% agreed to participate; these adults were slightly healthier than those who declined. Trained personnel performed annual study clinic evaluations, which included physical examination; diagnostic testing; blood sampling; and questionnaires on health status, medical history, and lifestyle. Each center's institutional review committee approved the study, and all participants provided informed written consent.
We measured fatty acids in 3941 of the 5565 living cohort participants who had blood samples taken at the 1992-1993 study visit, which we considered the baseline for this analysis. Details of cohort sampling and fatty acid measurements have been described elsewhere (11) and are provided in the Appendix. The assessment of EPA, DPA, and DHA levels and incident CVD and mortality was a prespecified aim of the research. After exclusion of 1113 participants with prevalent CVD and 136 receiving fish oil supplements at the time of blood sampling, 2692 participants were included in this analysis. At the 1992-1993 visit, cardiovascular risk factors, anthropometric values, blood pressure, and laboratory values were measured using standardized procedures, and alcohol use and physical activity were assessed using validated questionnaires (12 - 17). Dietary habits were assessed 3 years earlier (1989-1990) using a validated semiquantitative food-frequency questionnaire (18), from which dietary EPA plus DHA consumption was estimated as previously described (19).
Participants were followed by means of alternating study clinic examinations and telephone contacts every 6 months through 2000 and biannual telephone contacts thereafter. Vital status follow-up was 100% complete; less than 1% of all person-time was otherwise missing and censored early. All-cause and cause-specific mortality, as well as all suspected cases of incident (fatal or nonfatal) CHD and stroke, were assessed and adjudicated by a centralized events committee using available data from interviews; next of kin; death certificates; and medical records, including diagnostic tests and consultations. Algorithms and methods for follow-up; confirmation; and classification of deaths, CHD, and stroke have been described (20 - 22). Cardiovascular disease mortality included deaths due to CHD, stroke, other atherosclerotic disease, and other CVD. Non-CVD mortality included deaths due to cancer, pulmonary diseases, infection, dementia, fractures or trauma, and other causes. Arrhythmic CHD deaths were also adjudicated (21); sensitivity and specificity in comparison to a Hinkle classification were found to be 93% and 95%, respectively.
We evaluated ω3-PUFA levels in quintiles as indicator variables. To evaluate trends, we assessed quintiles as a continuous variable after participants were assigned the median value in each quintile. We estimated the hazard ratio using a Cox proportional hazards model (stcox command in Stata, release 12.0 [StataCorp, College Station, Texas]) with time at risk until first event, other deaths in cause-specific mortality analyses, or the latest date of adjudicated follow-up. The proportional hazards assumption was not violated on the basis of Schoenfeld residuals. Covariates were selected on the basis of biological interest, well-established relations with mortality in older adults, or associations with exposures (Appendix Table 1). We imputed missing covariates (0.18% to 0.72% for most factors and 7.79% to 12.30% for dietary factors) by best-subset regression (impute command in Stata) using multiple demographic and risk variables; results were similar when missing values were excluded. Potential nonlinear associations were evaluated semiparametrically using restricted cubic splines (mkspline command in Stata) (23). We estimated absolute years of remaining life gained or lost for each quintile of ω3-PUFAs by using both semiparametric and parametric approaches (24 - 26) (Appendix).
Sensitivity analyses were adjusted for regression dilution bias in ω3-PUFA levels (27 - 29) and measurement error in covariates (30) (Appendix), were limited to mid-follow-up (8 years) to minimize misclassification of exposures and covariates over time, and excluded deaths within the first 2 years to minimize effects of unrecognized subclinical disease on fatty acid levels. We defined statistical significance as a 2-tailed α level of 0.05. Exploratory analyses evaluated whether age, sex, or education affected relationships of EPA, DPA, and DHA levels with total mortality, with a Bonferroni-corrected, 2-tailed α level of 0.0056 (9 exploratory comparisons). Analyses were performed using Stata, release 12.0, and SAS, version 9.2 (SAS Institute, Cary, North Carolina).
Role of the Funding Source
The study was funded by the National Heart, Lung, and Blood Institute and the Office of Dietary Supplements of the National Institutes of Health. The funders had no role in the design or conduct of the study; collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript.