Diet patterns and the incidence of age-related macular degeneration in the Atherosclerosis Risk in Communities (ARIC) study

Abstract

Background

Age-related macular degeneration (AMD) is the leading cause of irreversible vision loss among the elderly.

Objective

This study aimed to determine the association between dietary patterns and food groups (used to make them) with the 18-year incidence of AMD.

Methods

ARIC (Atherosclerosis Risk in Communities) participants who showed change in AMD lesions between retinal photographs taken at visit 3 and visit 5 were graded side by side to determine incident AMD (any=144; early=117; late=27). A 66-line item food frequency questionnaire, administered at visit 1 and visit 3, was used to identify 29 food groups. Principal component analysis was used to derive dietary patterns from average food group servings. Logistic regression was used to estimate ORs and 95% CIs for incident AMD (any, early and late) by tertiles of dietary pattern scores, adjusted for age, race, education, total calories and smoking status. P-trend was estimated using continuous scores.

Results

Western (unhealthy) and Prudent (healthy) dietary patterns were identified. No significant associations were observed between either dietary pattern and incident any or incident early AMD. However, a threefold higher incidence of late AMD was observed among participants with a Western pattern score above, as compared with below, the median (OR=3.44 (95% CI 1.33 to 8.87), p-trend=0.014). The risk of developing late AMD was decreased, but not statistically significant, among participants with a Prudent pattern score above, as compared with below, the median (OR=0.51 (95% CI 0.22 to 1.18), p-trend=0.054).

Conclusions

Diet patterns were not significantly associated with incident any or incident early AMD. However, consumption of a Western pattern diet may be a risk factor for development of late AMD.

INTRODUCTION

Age-related macular degeneration (AMD) is the leading cause of irreversible vision loss among the elderly.¹ This vision-threatening eye condition affects the retina, particularly the macular region. Early AMD is characterised by pigmentary changes and accumulation of drusen, which are proteinaceous deposits.² Early AMD can progress to late AMD, which shows atrophy and growth of new blood vessels (neovascularisation) in the retina. Poor vision poses a challenge to one’s physical, mental and social well-being, leading to increased falls, fractures and a higher prevalence of depression.^3–5 Current treatments for AMD are intraocular injections and lasers,⁶ which are not only invasive and expensive, but available only for late stages. Therefore, it is important to identify modifiable risk factors like diet, to preserve vision and autonomy, by preventing the development and progression of conditions like AMD.

AMD clinical trials have demonstrated a protective role of antioxidants and supplements like vitamin C, E, zinc⁷ and carotenoids such as lutein and zeaxanthin.⁸ Observational studies have shown protective associations with nutrients like omega-3 fatty acids,⁹ vitamin D,^10,11 and low glycaemic index foods.¹² However, foods or nutrients are not consumed in isolation, instead as combinations, constituting meals. Thus, an integrated approach to investigate the impact of overall diet in the form of dietary pattern analysis may be a more relevant predictor of AMD risk than single nutrients or foods.

Previous cross-sectional and case–control studies have observed protective associations between a healthy diet pattern and risk of AMD, but temporality cannot be established in such study designs.^12–18 Existing prospective analyses have examined the association between a Mediterranean diet pattern and late, but not early, AMD.^19,20 We used data from the Atherosclerosis Risk in Communities (ARIC) study to examine the association between baseline dietary patterns and the 18-year incidence of early and late AMD. We also explored the association between the consumption frequency of individual food groups (used to create the diet patterns) and incident AMD.

METHODS

Study sample

The ARIC cohort, a prospective, population-based epidemiological study,²¹ recruited 15 792 participants at four field centres (Washington County, Maryland; Forsyth County, North Carolina; Jackson, Mississippi; and Minneapolis, Minnesota) (figure 1). Six study visits were completed between 1987 and 2017, with the seventh ongoing.²² Participants answered questionnaires related to lifestyle habits, demographics and medical histories.^21,23 Clinical examinations were also conducted.^21,23 Data on age, sex, race, body mass index (BMI), education, smoking status, history of hypertension, diabetes, physical activity and diet were collected. Participants provided consent and participating institutional review boards approved the ARIC protocol.

Figure 1.

Diagrammatic representation of the ARIC cohort. ARIC, Atherosclerosis Risk in Communities; AMD, age-related macular degeneration.

At visit 3 (1993–95), 12 887 participants had retinal photographs, of which 11 863 were gradeable.²⁴ Participants consenting for cardiovascular disease research only were excluded (n=796). Among the 6538 visit 5²⁵ participants, 2629 had retinal photographs taken. Of these, 1891 had gradeable photographs at both visits 3 and 5, but only 1326 belonged to the same eye, which allowed for a comparison of change in retinal lesions over time.

One hundred twenty-two participants went from no AMD to early AMD, 28 went from no or early to late AMD, 1159 participants remained free of AMD and 16 remained with early AMD. One participant with late AMD at visit 3 was excluded due to ineligibility to progress. Participants identified as outliers for energy (total calorie intake <500 or >3500 kcal, n=47) were excluded, leaving a final sample of n=1278, of which 117 were incident early AMD cases and 27 were incident late AMD cases (20 participants who went from no AMD to late AMD and 7 who progressed from early AMD to late AMD) (figure 1).

Outcome assessment

Graders, masked to previous outcomes, graded all retinal photographs by the Wisconsin Age-Related Maculopathy Grading System.²⁶ Non-mydriatic film photographs, centred between the disc and fovea, were taken of one randomly chosen eye at visit 3.²⁴ Non-mydriatic digital photographs of both eyes were taken at visit 5.²⁵ Film and digital photographs are comparable.²⁷ Eye photographs showing change between visits, like appearance/disappearance of lesions, underwent a side-by-side review for confirmation, thus determining incident early and incident late AMD. Participants were categorised as having incident early AMD if they had any of the following: (1) soft drusen (≥63 μm circle) within a grid area >500 μm circle, (2) any soft drusen ≥125 μm circle (either distinct or indistinct) present in the grid along with the presence of any pigmentary abnormality (increased/decreased pigmentation in the grid) or (3) large (≥125 μm circle) soft indistinct drusen present. Presence of either geographic atrophy, retinal pigment epithelial detachments, subretinal haemorrhage, subretinal fibrous scar, subretinal new vessels or history of treatment (laser, intravitreal injections) was categorised as late AMD.²⁶

Exposure assessment

Diet was assessed at visits 1 and 3 using a modified Willet’s 66-line item Food Frequency Questionnaire (FFQ).²⁸ Usual dietary intake over the past year was assessed in nine categories ranging from never to ≥6 times a day, assuming standard portion sizes. Average consumption frequencies reported at visit 1 and visit 3 were used to estimate dietary exposure over the previous ~6 years. Since diet may have changed between visits, a sensitivity analysis was conducted using visit 3 FFQ only, as this time point was closer to outcome ascertainment.

Genetic Data

The ARIC study has data on two high-risk AMD genes and single nucleotide polymorphisms (SNPs), CFH Y402H [rs1061170] and ARMS2 A69S [rs10490924], known to be associated with increased risk of early²⁹ and late³⁰ AMD. Affymetrix Genome-Wide Human SNP Array 6.0, Hap Map was used for genotyping and 1000 Genomes reference panels were used to impute CFH Y402H [rs1061170]. The imputation quality score was >0.80.^31,32 Hardy Weinberg equilibrium was met.

Statistical analysis

Similar to previous ARIC studies, FFQ line items were collapsed together to make 29 food groups.^33,34 Principal components analysis (PCA)³⁵ was used to derive factor loadings and subsequent dietary patterns, after orthogonal rotation of factors. Dietary patterns with eigenvalues above 2 were retained and were named the ‘Western/unhealthy’ and the ‘Prudent/healthy’ patterns, based on the food groups that loaded high on each (online supplementary table 1). Dietary pattern scores were categorised into tertiles. Differences in participant characteristics, including dietary and non-dietary factors, were compared by visit 5 AMD status, using χ² and two sample t-tests or analysis of variance for categorical and continuous variables, respectively. Logistic regression was used to estimate ORs and 95% CIs for the odds of incident any, incident early and incident late AMD among participants in dietary pattern tertiles 2 and 3 compared with tertile 1. P-trend was estimated using continuous dietary pattern scores. Models were adjusted for age and energy intake as these covariates independently increased the crude OR more than 10%. A stepwise approach was used to further examine the influence of variables like sex, race, education, smoking status, history of hypertension, diabetes and hormonal therapy. We chose to also adjust for education and the established risk factors of race and smoking.

The primary analyses were repeated using visit 3 FFQ data only. Analyses were also repeated stratifying by race. Associations between risk of any, early and late AMD among participants above compared with below the median frequency of consumption of each food group was examined.

Some participants with retinal photographs at visit 3 did not return for visit 5. To better understand the impact of this loss to follow-up, analyses were repeated accounting for inverse probability weights (IPW) (S2SAMWT51), designed to adjust for loss to follow-up bias from non-participation, death or photograph selection at visit 5,³⁶ as previously done in this cohort.³⁷ We used p value ≤0.05 for two-tailed test, which was considered statistically significant for all analyses. Analyses were conducted using SAS V.9.4.

RESULTS

Over a mean 18 years of follow-up, 117 (9%) participants developed incident AMD out of 1256 without any AMD at visit 3. There were 27 (2%) participants who developed incident late AMD out of 1256 without AMD and 22 with early AMD at visit 3. We identified two principal components, with eigenvalues3.19 and 2.91, explaining 11% and 10% variability, respectively. Food group factor loadings on both dietary patterns are presented in online supplementary table 1. Both Western and Prudent dietary pattern scores were normally distributed. Processed meat, fried food, dessert, eggs, refined grains, high fat dairy and sugar sweetened beverages loaded higher on the Western pattern. Cruciferous, carotene, dark green leafy and other vegetables, poultry, fresh fruits, fish and sea foods loaded higher on the Prudent pattern.

Overall, participants with AMD were older, more likely white, less active and with high-risk genotypes (table 1). No significant differences were noted by sex, BMI, education, smoking, history of hypertension or history of diabetes. Lutein and zeaxanthin intake per 1000 kcal and mean servings of fresh fruits, dark leafy vegetables, red meat and fish intake differed statistically significantly by AMD status (online supplementary table 2). Participant characteristics by diet pattern scores are also presented (online supplementary table 3 A and B).

Table 1.

Visit 3 characteristics (mean (SD) or n (%)) by incident any, early and late AMD at visit 5 (n=1278)

	Total n*	No AMD (n=1134)	Any AMD (n=144)	Early AMD (n=117)	Late AMD (n=27)	P value† No vs any No vs early No vs late
Age (years)	1278	58.4 (5.0)	61.3 (4.7)	61.0 (4.8)	62.8 (4.2)	<0.001
						<0.001
						<0.001
Sex	1278					0.572
Female	738	658 (58%)	80 (56%)	65 (56%)	15 (56%)	0.605
Male	540	476 (42%)	64 (44%)	52 (44%)	12 (44%)	0.815
Race	1278					<0.001
Whites	1076	937 (83%)	139 (97%)	112 (96%)	27 (100%)	<0.001
African–American	202	197 (17%)	5 (3%)	5 (4%)	0 (0%)	0.023
BMI (kg/m2)	1193
Normal weight (<25.0)	342	301 (29%)	41 (29%)	35 (31%)	6 (23%)	0.760
Overweight. (25.0–30.0)	528	464 (44%)	64 (46%)	51 (45%)	13 (50%)	0.711
Obese (≥30.0)	323	289 (27%)	34 (25%)	27 (24%)	7 (27%)	0.784
Highest education	1278
Basic	156	140 (12%)	16 (11%)	14 (12%)	2 (7%)	0.856
Intermediate	553	488 (44%)	65 (45%)	49 (42%)	16 (59%)	0.925
Advanced	569	506 (45%)	63 (44%)	54 (46%)	9 (33%)	0.232
Smoking status	1278
Current	138	120 (11%)	18 (13%)	13 (10%)	5 (19%)	0.063
Former	534	463 (41%)	71 (49%)	60 (51%)	11 (40%)	0.081
Never	606	551 (49%)	55 (38%)	44 (38%)	11 (41%)	0.409
Physical Activity Composite Score‡	1236
0–2	460	391 (36%)	69 (50%)	56 (50%)	13 (50%)	0.004
2–4	563	515 (47%)	48 (35%)	38 (34%)	10 (38%)	0.010
4–6	213	192 (17%)	21 (15%)	18 (16%)	3 (12%)	0.376
History of hypertension	1274					0.493
No	934	825 (73%)	109 (76%)	90 (77%)	19 (70%)	0.355
Yes	340	305 (27%)	35 (24%)	27 (23%)	8 (30%)	0.727
History of diabetes	1272					0.399
No	1220	1080 (96%)	140 (97%)	113 (97%)	27 (100%)	0.673
Yes	52	48 (4%)	4 (3%)	4 (3%)	0 (0%)	0.278
CFH genotype [rs 1061170]	1077
TT No high-risk allele	428	391 (41%)	37 (29%)	31 (31%)	6 (22%)	<0.001
CT One high-risk allele	499	444 (47%)	55 (43%)	43 (43%)	12 (44%)	<0.001
CC Two high-risk alleles	150	114 (12%)	36 (28%)	27 (27%)	9 (33%)	0.008
ARMS2 genotype [rs 10490924]	1093
GG No high-risk alleles	658	595 (62%)	63 (49%)	54 (53%)	9 (33%)	0.002
TG One high-risk allele	375	323 (33%)	52 (40%)	39 (38%)	13 (48%)	0.120
TT Two high-risk alleles	60	46 (5%)	14 (11%)	9 (9%)	5 (19%)	0.001

^*Total n<1278 is due to missing data for this variable.

^†Two sample t-test to test for differences by continuous variables. χ² tests to test for differences by categorical variables.

^‡Composite Physical Activity Score (ranging from 0 (low physical activity) to 6 (high physical activity)) at visit 3 was estimated using questionnaire scores for leisure, work and sports-related activity.

AMD, age-related macular degeneration; BMI, body mass index.

Western and Prudent diet patterns

There was no association between incident any or incident early AMD and the Western pattern score (table 2). An increased odds of incident late AMD was seen among participants who had a Western pattern score above compared with below the median (Model 1: OR=3.44 (1.33 to 8.87); p-trend=0.014). No statistically significant associations were observed between the Prudent pattern score and incident any, incident early or incident late AMD (table 2). To examine the influence of loss-to follow-up, we applied IPWs (Model 2). Adding these weights mostly widened the 95% CIs, but the significant association with the Western pattern score and incident late AMD remained, and the p-trend for the Prudent pattern score became statistically significant (p-trend=0.012) (table 2).

Table 2.

Adjusted ORs and 95% CIs for incident any, early and late AMD at visit 5, by tertiles of Western and prudent dietary pattern scores using average visit 1 and visit 3 diet data (n=1278)

Any AMD	Tertile 1	Tertile 2	Tertile 3	P trend*
Western pattern	n=426	n=426	n=426
Events n†	49	50	45
Model 1‡	Reference	1.19 (0.76 to 1.87)	1.43 (0.83 to 2.46)	0.287
Model 2§	Reference	1.31 (0.78 to 2.20)	1.69 (0.88 to 3.25)	0.144
Prudent pattern	n=426	n=426	n=426
Events n	48	54	42
Model 1	Reference	1.11 (0.72 to 1.72)	0.90 (0.56 to 1.46)	0.623
Model 2	Reference	1.24 (0.73 to 2.11)	0.84 (0.49 to 1.46)	0.555
Incident Early AMD¶	Tertile 1	Tertile 2	Tertile 3	P trend
Western pattern	n=417	n=412	n=407
Events n	44	40	33
Model 1	Reference	1.03 (0.64 to 1.66)	1.10 (0.61 to 2.00)	0.882
Model 2	Reference	1.24 (0.71 to 2.14)	1.41 (0.69 to 2.88)	0.493
Prudent pattern	n=411	n=410	n=415
Events n	37	44	36
Model 1	Reference	1.17 (0.73 to 1.89)	1.00 (0.59 to 1.68)	0.858
Model 2	Reference	1.38 (0.76 to 2.52)	0.97 (0.53 to 1.78)	0.996
Incident late AMD	<Median	≥Median	-	P trend
Western pattern	n=639	n=639	-
Events n	9	18	-
Model 1**	Reference	3.44 (1.33 to 8.87)	-	0.014
Model 2	Reference	3.00 (1.06 to 8.47)	-	0.005
Prudent pattern	n=639	n=639	-
Events n	17	10	-
Model 1**	Reference	0.51 (0.22 to 1.18)	-	0.054
Model 2	Reference	0.43 (0.18 to 1.05)	-	0.012

^*P-trend was computed using the continuous diet patterns score.

^†Events=Total cases in the tertile group.

^‡Model 1: Adjusted for age, race, education, daily energy intake (kcal) and smoking status.

^§Model 2: Model 1 further adjusted for inverse probability weights.

^¶The n=1236 because participants with early AMD at visit 3 and/or those who developed incident late AMD by visit 5 were excluded from this analysis.

^**There were no cases of African–American participants with late AMD; therefore, the model for risk of late AMD could not be adjusted for race.

AMD, age-related macular degeneration.

Analyses repeated with visit 3 diet, instead of the average diet for intake at visit 1 and 3, yielded similar results (online supple mentary table 4). We also stratified our analyses by race (online supplementary table 5). Among whites, a similar threefold increased odds of incident late AMD was observed among participants with a higher compared with lower Western pattern score. Among African–Americans, no associations were observed between diet patterns and incident early AMD. Lack of incident late AMD in African–Americans prevented estimates of associations with diet pattern scores.

Food groups

We also examined the odds of incident any, incident early and incident late AMD among participants in high compared with low intake of each food group as defined by the median cut-point in servings/month. We observed no significant associations with incident any AMD (data not shown). For incident early AMD, we observed an increased odds of early AMD among participants with high compared with low frequency of consumption of fried food (as defined by the sample median) (figure 2). We observed a decreased odds of late AMD with high compared with low frequency of consumption of fresh fruit, legumes and low fat dairy, and an increased odds of late AMD with high compared with low frequency of consumption of red meat, eggs and fat (defined as servings of margarine or butter) (figure 2).

Figure 2.

Forest plot showing adjusted ORs and 95% CIs for early and late age-related macular degeneration (AMD) at visit 5 by servings of the 29 food groups used to make the pattern scores (food group servings/month from the average of intake at visits 1 and 3).

DISCUSSION

In this prospective study, dietary patterns were not significantly associated with incident any or incident early AMD. We did observe an increased risk of incident late AMD with high compared with low consumption of a Western (unhealthy) diet pattern. Our results complement the findings from two recent prospective studies,^19,20 which found a reduced risk of incident late AMD, among people with high compared with low ranking on a Mediterranean (healthy) diet score. Together, these findings constitute some of the first evidence that diet patterns are related to risk of developing late AMD over time. Most previous studies have presented associations between diet and AMD outcomes (prevalent AMD) measured at the same point in time, with the inability to establish temporality.

In our study, the Western pattern consisted of high intake of processed and red meat, fried food, dessert, eggs, refined grains, high fat dairy and sugar sweetened beverages, and low intake of yoghurt, low fat dairy, fresh fruit, cruciferous vegetable, whole grains and carotene vegetables. The Prudent pattern was inversely correlated with the Western pattern (Pearson r=−0.15, p<0.001). The Prudent pattern was defined by high consumption of all vegetables, including dark leafy vegetables, legumes, poultry, fish and seafood and by low intake of sugar sweetened beverages, fried food, coffee, processed meat, sweets and candy, eggs, ice cream and desserts. Our study’s findings support the hypothesis that low of intake of foods rich in nutrients like lutein, zeaxanthin (leafy and carotene vegetables), and long chain polyunsaturated fatty acids like ecosapentanoiec acid and doxosahexanoiec acid (fish and seafoods), increase the risk of developing late AMD.^38–42 Such nutrients are thought to promote a healthy immune system, reduce oxidative stress due to free radicals and inflammatory changes in the retina.^2,43–45 Further, our study’s findings support the hypothesis that high consumption of processed and red meats, refined grains and foods with added sugars is associated with increased risk for incident late AMD.

Previous studies of dietary patterns and AMD have used both data-driven techniques like PCA^14,15,18 as well as dietary indices.^{12,13,16,17,19,20} Despite the method used to determine dietary patterns, the results from all studies examining associations between dietary patterns and AMD have been consistent; that a poor diet may increases risk of late AMD. The Melbourne Collaborative Cohort Study found data-driven dietary patterns defined by high intake of grains, fish, steamed chicken, nuts and some vegetables to be associated with lower odds of prevalent late AMD, and data-driven dietary patterns defined by high intake of red meat to be associated with a higher prevalence of late AMD.¹⁵ No associations were observed between these dietary patterns and early AMD. In AREDS, using a data-driven approach, it was observed that an Oriental dietary pattern was associated with decreased odds of early and late AMD, and that a Western dietary pattern was associated with increased odds of early and late AMD.¹⁴ Mares et al reported that women in the highest compared with the lowest quintile of a modified Healthy Eating Index had 46% reduced odds of early AMD.¹³ A case–control study reported a 46% reduced odds of late AMD, comparing participants in the fourth to first quartile for the alternative Heathy Eating Index, but no association was seen with early AMD.¹⁷ The European Eye Study compared participants with Mediterranean diet score (MDS) above 6 with that ≤4, and found no association with early, but a protective trend for neovascular AMD.¹⁶ We did not stratify by late AMD types (neovascular/geographic atrophy) due to limited case numbers.

In addition to overall diet, we also assessed how consumption of individual food groups influenced AMD incidence. Previous studies suggest that foods like fish and nuts³⁹ are protective, while higher red meat⁴⁶ intake is associated with increased AMD risk. Although we did find associations with some food groups that rank high by their factor loadings on one of the dietary patterns, not all foods which ranked high or low on the dietary patterns were consistently associated with our AMD outcomes. This may suggest that the synergy of looking at multiple foods together in a dietary patterns tells us more about the influence of diet on AMD than one food or nutrient alone.

We acknowledge that our findings for late AMD are based on small sample sizes, and thus should be interpreted with this limitation in mind. All self-reported FFQs are subject to bias and exposure misclassification, but an analysis with an expanded version of the FFQ used in this study has shown that it is reproducible and valid for assessment of dietary patterns.⁴⁷ It is possible that participants changed their diets in the interim between dietary assessment and retinal photographs at visit 5 which could have influenced our findings. We used a 66-line item FFQ which has fewer foods than other FFQs used to assess associations between diet patterns and AMD, and this may have failed to capture all dietary exposures. However, our results are consistent with most previously published work in this area. We also acknowledge that lifestyle factors are difficult to disentangle and residual confounding likely still exists. Additionally, PCA, a population-specific technique, limits our ability to compare our findings with other cohorts. Lastly, it is possible that bias from loss to follow could have influenced our results, but this seems unlikely as our findings were similar regardless of whether or not the IPW weights were applied.

Despite these limitations, we still observed that consuming an unhealthy Western pattern diet was associated with increased risk of late AMD. Our study was not without strengths, such as its prospective design. ARIC is a well-established cohort and implements good quality control measures. Multiple trained professionals were used to grade retinal photographs, reducing the chance of false positives or negatives for AMD incidence. We also were able to average dietary intake from two time points, capturing a longer period of dietary exposure than one assessment alone. In conclusion, among older adults, consuming a Western (unhealthy) diet may increase the risk of progression to incident late AMD over time. We believe that these results, along with the current knowledge regarding chronic disease prevention and findings from the literature, will aid in making recommendations regarding a vision-healthy diet inclusive of plant foods and limited in processed meats, red meats, foods rich in saturated fats, simple sugars and refined grains. Pooling data from existing prospective studies with diverse populations should be conducted. This will increase the between-person variation in dietary intake and provide the power to examine effect modification by genetics and other factors.