The Relationship of Dietary ω-3 Long-Chain Polyunsaturated Fatty Acid Intake With Incident Age-Related Macular Degeneration: AREDS Report No. 23

Objective  To examine the association of dietary ω-3 long-chain polyunsaturated fatty acid and fish intake with incident neovascular age-related macular degeneration (AMD) and central geographic atrophy (CGA).

Methods  Multicenter clinic-based prospective cohort study from a clinical trial including Age-Related Eye Disease Study (AREDS) participants with bilateral drusen at enrollment. Main outcome measures were incident neovascular AMD and CGA, ascertained from annual stereoscopic color fundus photographs (median follow-up, 6.3 years). We estimated nutrient and food intake from a validated food frequency questionnaire (FFQ) at baseline, with intake of docosahexaenoic acid (DHA), eicosapentaenoic acid (EPA), c ombined EPA and DHA, and fish as primary exposures.

Results  After controlling for known covariates, we observed a reduced likelihood of progression from bilateral drusen to CGA among people who reported the highest levels of EPA (odds ratio [OR], 0.44; 95% confidence interval [CI], 0.23-0.87) and EPA+DHA (OR, 0.45; 95% CI, 0.23-0.90) consumption. Levels of DHA were associated with CGA in age-, sex-, and calorie-adjusted models (OR, 0.51; 95% CI, 0.26-1.00); however, this statistical relationship did not persist in multivariable models.

Conclusions  Dietary lipid intake is a modifiable factor that may influence the likelihood of developing sight-threatening forms of AMD. Our findings suggest that dietary ω-3 long-chain polyunsaturated fatty acid intake is associated with a decreased risk of progression from bilateral drusen to CGA.

Age-related macular degeneration (AMD) is a leading cause of irreversible vision loss in elderly persons.^1-3Docosahexaenoic acid (DHA) (C22:6 ω-3), the major dietary and structural ω-3 long-chain polyunsaturated fatty acid (LCPUFA) of the retina, has the capacity to modulate processes implicated in AMD pathogenesis.⁴Eicosapentaenoic acid (EPA) (C20:5 ω-3), the precursor to DHA and the other major dietary ω-3 LCPUFA, can exert similar actions to DHA.⁴Report 20 from the Age-Related Eye Disease Study (AREDS) describes a 40% to 50% reduced likelihood of having neovascular (NV) AMD among participants who reported the highest levels of ω-3 LCPUFA consumption (odds ratio [OR], 0.61; 95% confidence interval [CI], 0.41-0.90)⁵; these findings are concordant with those from other studies examining relationships of ω-3 LCPUFAs^6-9and ω-3 LCPUFA–rich food intake^6-8,10-13with various stages of incident^8,9,14,15and prevalent^6,7,10-13AMD. Although statistical relationships did not always exist in these studies, measures of association were consistently in the direction of benefit.In addition to its value for cross-sectional analyses, AREDS provides a large and carefully phenotyped prospective sample of participants examined for progression to advanced AMD. Higher levels of ω-3 LCPUFA and fish intake were associated with a decreased likelihood of progression to advanced AMD in subgroups from the 2 published studies examining this issue.^8,14In this report, we document an inverse relationship of EPA and ω-3 LCPUFAs with progression to central geographic atrophy (CGA) among AREDS participants at moderate risk of this event.

The AREDS protocol was approved by a Data and Safety Monitoring Committee and by the institutional review board of each clinical center. Informed consent was obtained from all participants before enrollment. Details of the AREDS design and methods have been given in earlier publications.^5,16,17This report is a prospective study of the AREDS clinical trial cohort.

We analyzed the 2132 AREDS category 2 and 3a participants who responded to the AREDS Food Frequency Questionnaire (FFQ) at baseline and reported caloric intake values between the 1st and 99th percentiles of those from a nationally representative probability sample of people aged at least 71 years (Continuing Survey of Food Intakes by Individuals [CSFII], 1994-1996 and 1998).^18,19AREDS report No. 19¹⁷contains detailed definitions of categories 2 and 3a; participants in these categories had bilateral visual acuity of 20/32 or better (≥75 letters read correctly on the Early Treatment of Diabetic Retinopathy Study chart), measured by a standard protocol, and media clear enough to obtain good-quality fundus photographs.¹⁶Category 2 participants had mild or borderline age-related macular changes (multiple small drusen [≤63 μm] or intermediate drusen [>63 μm and <125 μm] and/or pigment abnormalities). Category 3a participants had large drusen (≥125 μm), extensive intermediate drusen, and/or geographic atrophy (GA) that did not involve the center of the macula. Participants were included if these macular changes were seen in both eyes at baseline and are defined as having bilateral drusen. Our study population was chosen to represent participants at mild-to-moderate risk of progression to advanced AMD in both eyes.

We administered detailed questionnaires to obtain demographic information, history of smoking and sunlight exposure, medical history, history of specific prescription drug and nonprescription medication use, and history of vitamin and mineral use. Participants completed a validated self-administered, 90-item, semiquantitative FFQ at baseline.^5,20,21The food list for the AREDS FFQ contained items rich in a variety of nutrients that have putative associations with AMD, such as lipids, macular xanthophylls (lutein/zeaxanthin), pro–vitamin A carotenoids, and vitamins and minerals with antioxidant properties. Participants were asked how often, on average, they had consumed each food or beverage item during the past year. Average frequency of consumption was recorded across 9 levels that ranged from “never or less than once per month” to “2+ per day.” Average serving size was recorded as “small,” “medium,” or “large,” with respect to standard examples. We assessed fish intake with 5 separate items: (1) tuna, tuna salad, or tuna casserole; (2) fried fish or fish sandwich; (3) other fish (baked/broiled); (4) oysters; and (5) other shellfish (eg, shrimp, crab, lobster). The AREDS FFQ was based on the 1987 National Cancer Institute Health Habits and History Questionnaire version 2.1, that was modified for use in AREDS with data obtained from 2-day food records sampled from 78 study-eligible persons selected from the 11 AREDS clinics. The instrument was validated using a telephone-administered 24-hour dietary recall at 3 and 6 months after enrollment in 197 randomly selected participants. Dietary intake data were processed with DIETSys software (version 3.0, National Cancer Institute, Information Management Services, Inc, Block Dietary Data Systems, Bethesda) at the University of Minnesota Nutrition Coordinating Center. The DIETSys system produced daily nutrient intake estimates for each subject by first multiplying the average age- and sex-adjusted portion size (derived from data gathered in the second National Health and Nutrition Examination [NHANES II], by the subject's reported serving size. Cumulative estimates for each nutrient were computed by summation of nutrient values across all foods and items. Correlations of 24-hour recall data with the AREDS FFQ were corrected for attenuation using the method of Rosner and Willett.²²Values for correlation coefficients were 0.35 for EPA and 0.32 for DHA. We used the University of Minnesota Nutrition Coordinating Center Food Composition Database (version 31, November 2000) with the estimated quantity of intake to derive individual nutrient values for each questionnaire item.

General physical and ophthalmic examinations included height, weight, blood pressure, manifest refraction, best-corrected visual acuity, intraocular pressure, slitlamp biomicroscopy, and ophthalmoscopy. Slitlamp photographs (Topcon Corporation, Tokyo, Japan) and Neitz photographs (Neitz Instruments Co Ltd, Tokyo) of the lens were taken, along with stereoscopic fundus photographs of the macula and red reflex lens photographs. These were graded at a photograph reading center, where the various lesions associated with AMD and the degree of lens opacities by type were assessed through standardized grading procedures.¹⁶

Progression to NV AMD for a study eye was based on clinical center reports of photocoagulation for choroidal NV or photographic documentation at the reading center of any of the following: nondrusenoid retinal pigment epithelial detachment, serous or hemorrhagic retinal detachment, hemorrhage under the retina or the retinal pigment epithelium, and/or subretinal fibrosis.¹⁶The analysis of progression to CGA (GA definitely involving the center of the macula, or questionably involving the center but definitely present proximally, based on reading center reports) did not count CGA when it occurred in an eye exhibiting subretinal fibrosis at the same visit. With this 1 exception, analyses of progression to NV AMD or CGA were without regard to progression to the other. Analyses involved progression within a participant, regardless of whether 1 or 2 eyes showed progression.

We applied the analytic framework described in AREDS report No. 19.¹⁷In brief, we used a staged modeling technique with generalized estimating equations in repeated-measures logistic regression. The technique allowed us to consider the occurrence of outcomes at each visit for each participant. We adjusted our initial models for age, sex, and AREDS treatment. Additional factors then were specified through a model simplification process evaluating χ²tests of change in deviance. This process consisted of identifying nominally nonsignificant (P > .1) coefficients from stage 2 and removing those variables from the model. Model simplification continued until the reduced model yielded a significant (P < .05) worsening of fit. Regression-fit diagnostics with the likelihood ratio criterion were conducted to construct parsimonious multivariable models. Final models from AREDS report No. 19¹⁷were the starting point for our multivariable analyses.

Habitual dietary intake of DHA and EPA were the primary independent variables, as represented with a nutrient density score (milligrams of intake per day/total daily caloric intake); details of dietary intake assessment with the AREDS FFQ and nutritional data analysis exist in AREDS report No. 20⁵and report No. 22.²¹Ordinal intake categories were computed from the distributions of data collected from all AREDS participants submitting a valid FFQ at the baseline visit. We added terms for total energy intake (TEI) to all of the models. The main sources of EPA and DHA in the retina are from preformed molecules (and not biosynthesis from their precursors). We report results for EPA and DHA separately because the following physiological bases for doing so exist: (1) DHA is a key structural molecule in the retina and also acts as the precursor to bioactive compounds (docosanoids)²³with immunoregulatory and cytoprotective properties and (2) EPA acts as the precursor to a number of bioactive compounds (eicosanoids)²⁴with vasoregulatory and immunoregulatory properties.We suspected that analyses of EPA and DHA would yield similar point estimates owing to their similar distribution in foods and capacity to affect similar biological processes. However, because there are no published estimates for these nutrients for our outcomes, we believe it is important to give them in this report.

Our analyses included 2132 participants whose total energy intake was reported to be between the 1st percentile (677 kcal/d for women and 794 kcal/d for men) and 99th percentile (1994 kcal/d for women and 2771 kcal/d for men) of the CSFII.¹⁸We considered these participants most likely to submit accurate dietary intake estimates. The 232 participants excluded from the analysis were of similar age and race and had a similar history of smoking, use of antacids, and AREDS clinical trial treatment assignment as those in the final sample. They were more likely to be female (73% vs 56%; P < .001) and less likely to have at least 16 years of formal education (29% vs 36%; P < .01). Progression to CGA and NV AMD were the primary outcomes. AREDS report No. 19¹⁷contains details on outcome ascertainment and classification.

Baseline characteristics of the sample are shown in eTable 1). Median follow-up was 6.3 (range, 1-8) years. Most participants were 70 years or older (43.3%), female (56.1%), and white (95.8%). Median intake of DHA (in percentage of TEI) was 0.010% for quintile 1, 0.018% for quintile 2, 0.026% for quintile 3, 0.037% for quintile 4, and 0.061% for quintile 5. The respective intake values of EPA are 0.000%, 0.009%, 0.015%, 0.024%, and 0.044%. The median percentage of TEI for DHA quintile 3 in AREDS was 0.026%, which is similar to that reported as the overall mean percentage of TEI from the Third National Health and Nutrition Examination (NHANES III) (0.03%) and the 1994-1996 and 1998 CSFII (0.025%). The AREDS median percentage of TEI for quintile 3 of EPA was 0.015%; NHANES III reported 0.02%, and CSFII reported 0.013%. To convert these values to approximate grams per day for a person consuming a 2000-kcal/d diet, multiply the percentage of TEI times 2.185.

Table 1contains results of the final multivariable models, and elements of the full model also appear in that table. All CGA and NV AMD models included terms for age, sex, AREDS treatment, smoking history, antacid use, and total calories (modeled as a continuous variable). The CGA model also included education and body mass index In addition to the common risk factors, the NV AMD model included race.

Table 2shows results of the final multivariable models for fish intake. Covariates for the models in Table 2are the same as in Table 1. Fish intake contributed 93% of EPA and 71% of DHA intake in the AREDS population.

Central GA developed in 113 participants.In age-, sex-, and calorie-adjusted models, participants who reported the highest levels of DHA intake were half as likely to experience progression to CGA as those who reported the lowest intake (OR, 0.51; 95% CI, 0.26-1.00). There was a negligible change in the magnitude of this estimate after statistical adjustment for the set of nonnutritional predictors and correlates of CGA in AREDS report No. 19¹⁷; however, the OR did not attain significance (OR, 0.55; 95% CI, 0.28-1.10). Participants who reported the highest EPA intake (OR, 0.41; 95% CI, 0.21-0.78) and EPA+DHA intake (OR, 0.41; 95% CI, 0.21-0.80) were 60% less likely to progress to CGA in the age-, sex-, and calorie-adjusted models. These relationships persisted in multivariable models and, as observed with DHA, did not change appreciably by adding covariates. The respective ORs for EPA intake and EPA+DHA intake were 0.44 (95% CI, 0.23-0.87) and 0.45 (95% CI, 0.23-0.90). The ORs for fish intake were in the same range but did not attain statistical significance.

Neovascular AMD developed in 198 participants. We did not observe relationships of EPA, DHA, or EPA+DHA intake with incident NV AMD.People who reported the most frequent tuna intake were approximately half as likely to experience progression to NV AMD as those who reported the least frequent intake (OR, 0.48; 95% CI, 0.24-0.95).

This report extends the findings of AREDS report No. 19¹⁷to include ω-3 LCPUFA and ω-3 LCPUFA–rich food-based factors. To our knowledge, this work is the first to report separate outcomes on progression to NV AMD and CGA as they may relate to ω-3 LCPUFA intake. Associations of nutrients and AMD attained statistical significance for CGA only. The ω-3 LCPUFAs and their metabolites show cytoprotective effects under conditions of oxidant stress in primary cell cultures of rat photoreceptors^25,26and human retinal pigment epithelial cells²⁷through their effect on the expression of apoptotic regulatory proteins. Studies of ω-3 LCPUFA feeding in young rodents resulted in a decreased volume of abnormal neovascular retinal²⁸and choroidal²⁹growth in response to angiogenic stimuli via modulation of inflammatory factors and processes. Our null results on NV AMD are unexpected in the context of emerging evidence, and we believe it is best to obtain confirmation from independent samples before applying strong inference. AREDS2 (http://www.areds2.org) is a 4000-person randomized clinical trial designed to examine the efficacy of ω-3 LCPUFA supplementation on prevention of progression to CGA and NV AMD and should offer information on issues discussed in this report.

We acknowledge a number of cautions concerning the validity of inference in our study. These include differential and random misclassification of nutrient intake based on disease status and psychometric factors (FFQ measurement and interpretation issues). A healthy diet usually reflects a healthy lifestyle; this is a factor that may confound our inferences. We did not have measures of physical activity to aid in the assessment of this possibility.These concerns for validity are to be considered with the pertinent strengths of the AREDS research design. Strengths include the standardized methods of exposure and outcome ascertainment, an extensive set of directly measured health-related variables, annual ophthalmic examinations with fundus photographs applied for outcome classification, and low participant attrition rates. Because dietary intake data were obtained before the development of NV AMD or CGA, it is unlikely that AMD status systematically biased reporting. If misclassification (in this case, inaccuracy) of nutrient intake existed, it was most probably random in nature, thus affecting all participants and attenuating measures of association. Because relationships of CGA with EPA, DHA, or EPA+DHA were always strongest in comparison of extreme exposure categories (quintile 1 vs quintile 5 or quintile 4), the strength of associations was not changed appreciably after adjustment for age, sex, smoking, and other covariates. We acknowledge that unmeasured factors may have operated differentially within the participants who experienced progression to CGA, and we are unable to conclusively refute the idea that a factor or a process strongly associated with ω-3 LCPUFA intake may have confounded our inferences. To have an appreciable effect on our results, the actions of this factor or process would need to vary with both ω-3 LCPUFA intake and AMD. The ORs suggest a potential relationship between tuna intake and NV AMD, but the lack of a strong finding for relationships of EPA and DHA with NV AMD may indicate that fish intake is a proxy for a healthy lifestyle. Although concentrations of DHA and EPA vary with the type and habitat of fish, the structure of AREDS FFQ items did not provide information on these factors.

Emerging evidence suggests that ω-3 LCPUFA intake may reduce the likelihood of progression to CGA or NV AMD. Seddon et al¹⁴applied a prospective design to report a 60% reduced risk of progression to advanced AMD (defined as NV AMD or GA) among people who reported the highest fish consumption (≥2 times vs <1 time per week) and low linoleic acid intake (relative risk, 0.36; 95% CI, 0.14-0.95). We did not observe any effect modification of linoleic acid on ω-3 LCPUFAs for either outcome. A 5-year prospective sample from the Blue Mountains Eye Study⁸(with the end point defined as NV AMD or GA) yielded ORs of 0.18 (95% CI, 0.02-1.38) for comparison of the highest quintile vs quintiles 2, 3, and 4 and 0.25 (95% CI, 0.06-1.00) for the highest quintile vs the lowest quintile of fish consumption. Neither work reported separate findings for NV AMD or GA.

Results from participants who reported caloric intake values between the 5th and 95th percentiles from the CSFII are found in eTable 2; these AREDS participants were most likely to report dietary intake accurately. In this cohort, association of CGA with the highest level of DHA intake emerged (OR, 0.45; 95% CI, 0.21-0.95), and the reduced likelihood of progression to CGA remained at 60% (OR, 0.41; 95% CI, 0.20-0.85) for EPA and EPA+DHA.

This AREDS report provides evidence that participants with bilateral drusen who reported the highest levels of EPA and EPA+DHA intake had a 50% decreased likelihood of progression to CGA relative to their peers, after considering the impact of known risk factors for AMD. Carefully designed observational analytic designs or clinical trials would provide necessary information on whether dietary intervention or supplementation with ω-3 LCPUFAs may help prevent the development of advanced AMD.

Correspondence:AREDS Coordinating Center, The EMMES Corporation, 401 N Washington St, Ste 700, Rockville, MD 20850-1707 (aredspub@emmes.com).

Submitted for Publication:March 23, 2007; final revision received January 28, 2008; accepted February 26, 2008.

Financial Disclosure:None reported.

Funding/Support:This study was supported by contracts from the National Eye Institute, National Institutes of Health, Department of Health and Human Services.