The Coimbra Eye Study

The CES comprises the Epidemiological Study (NCT01298674, n = 2975 for Mira cohort; n = 3021 for Lousã cohort), the AMD Incidence Study (n = 1617 for Mira cohort) and the Lifestyle and Food Habits Questionnaire in the Portuguese Population aged 55 or more (n = 985 for Mira cohort; n = 1007 for Lousã cohort).

All studies were cross-sectional and population based. The Epidemiological Study determined the prevalence of AMD in two different populations: Mira (coastal) and Lousã (inland). In the the Epidemiological Study, participants were recruited from the Primary Healthcare Units of Mira and Lousã, provided they were at least 55 years old, sex- and age-matched between the two cohorts. The AMD Incidence Study determined the 6.5-year incidence in Mira. The Lifestyle and Food Habits Questionnaire in the Portuguese Population aged 55 or more study was performed at the same time as the AMD Incidence Study and assessed nutrition and behavioural habits of the two populations (Mira and Lousã). Both studies ran between 2016 and 2017. Details of the studies can be found elsewhere [5, 6, 22,23,24].

Patients signed the informed consent after explanation of study procedures and possible consequences. The studies procedures complied with the tenets of the Declaration of Helsinki and International Conference on Harmonization–Good Clinical Practice Guidelines. The studies obtained the Association for Innovation and Biomedical Research on Light and Image (AIBILI)’s Ethics Committee approval.

Procedures

This manuscript refers to the data acquired in the AMD Incidence Study (NCT027048824) and Lifestyle and Food Habits Questionnaire in the Portuguese Population Aged 55 or More (NCT01715870).

Shortly, subjects who participated in the AMD Incidence Study went through an ophthalmological assessment by an ophthalmologist, best-corrected visual acuity with Early Treatment Diabetic Retinopathy Study (ETDRS) charts, color fundus photography (CFP, Topcon® fundus camera, TRC-NW8; Topcon Corp., Tokyo, Japan), spectral-domain optical coherence tomography (SD-OCT), fundus autofluorescence (FAF), and infra-red (IR) imaging (Spectralis HRA + OCT Heidelberg Engineering, Heidelberg, Germany). Additionally, a trained study nurse collected data on clinical history and took blood samples from participants who consented for genetic analysis (n = 948 genotyped).

Lifestyle and food questionnaire

In the Lifestyle and Food Habits Study in the Portuguese Population aged 55 or more, participants answered a validated lifestyle and food questionnaire. In the lifestyle questionnaire, data on demographics, weight, height and waist circumference, education, smoking habits, physical activity (any type of physical activity performed, regardless the duration), and cardiovascular comorbidities were recorded.

The food questionnaire, validated for the Portuguese population [25], comprised 86 food items arranged by eight groups: (1) Dairies; (2) Eggs, meat, and fish; (3) Oils and fats; (4) Bread and cereals; (5) Sweets and pastries; (6) Legumes; (7) Fruits and (8) Beverages and miscellaneous. For each of the 86 food items, participants answered what the frequency of intake was in the last year (never or less than 1 time/month, 1–3 times/month, 1 time/week, 2–4 times/week, 5–6 times/week, 1 time/day, 2–3 times/day, 4–5 times/day, 6 or more times/day), as well as the average serving size (cups, plate, units, mL, g, number of items, tablespoons, teaspoons) and whether the food item was seasonal.

Each food item’s intake was calculated as the average daily consumption, adjusted for the serving size, in grams. In case of seasonal intake, a value of 0.25 was considered to weight for a period of three months.

Assessment of adherence to the Mediterranean diet (MediSCORE)

In order to assess the level of mediSCORE [26]. Food items from the food questionnaire were grouped into nine groups: vegetables, legumes, fruits, cereals, fish, meat, dairies, alcohol, and a ratio of monounsaturated lipids (mainly olive oil) to saturated lipids (fats). Each of these groups takes up a value of 0 or 1, compared with the median of sex-specific food item consumption (in grams) of the Mira cohort population calculated in the food questionnaire. For healthy groups (vegetables, legumes, fruits, cereals, and fish), consumption above the median was assigned the value 1. For the food groups considered harmful (meat and dairies), consumption above the median was assigned the value 0. Specifically, for alcohol, a value of 1 was attributed to consumptions between 10 and 50 g/day for men and between 5 and 25 g/day for women, as it was considered healthy. Also, for fat, a ratio of monounsaturated/saturated lipids was considered beneficial when above the sex-specific median consumption and, hence, assigned the value of 1 [26, 27]. The quantity of monounsaturated and saturated lipids that each food item contains is given by “The Portuguese Food Composition Database”, which is the reference document for the composition of food item in Portugal.

Additional file 1: Table S1 presents the cut-off values used to assess a beneficial or detrimental consumption of each food group of the mediSCORE.

An overall value was calculated for each participant, ranging from 0 to 9, using the sum of each nine food groups.

High adherence to the Mediterranean diet was expressed by a mediSCORE ≥ 6, which is close to the third tercile of the distribution of the mediSCORE in our sample (Fig. 1). Low adherence to the Mediterranean diet was defined by mediSCORE < 6.

Fig. 1

Distribution of the Mediterranean diet score (mediSCORE) in the Mira cohort (n = 612). mediSCORE, adherence to the Mediterranean diet score

AMD grading

A general grading (disease/no disease) and a detailed grading (AMD staging) were performed. In cases of bilateral disease, the AMD phenotype was determined by the grading of the most severe eye. In case of only one presenting the disease, this eye was considered for grading. Grading was performed at Coimbra Ophthalmology Reading Center (CORC, AIBILI, Portugal), by certified ophthalmologists. When available, the grading of the two eyes was considered for this analysis.

Definition of cases and controls and study subjects

Rotterdam Classification was used [28, 29]. Cases were defined as participants with stage 2a (soft, indistinct or reticular drusen), stage 2b (soft, distinct drusen with pigmentary irregularities), stage 3 (soft, indistinct drusen with pigmentary irregularities) and stage 4 geographic atrophy (GA) or choroidal neovascularization (CNV). Controls were participants with stage 0 (no AMD or drusen < 63 μm) above 60 years old, and stage 1 (soft, distinct drusen or pigmentary irregularities) above 70 years old, in order to minimize the chance of classifying participants who could still progress to AMD as controls.

Genotyping

Genotyping of patients was performed in collaboration with the The European Eye Epidemiology Consortium (E3), within the EYE-RISK project. Detailed information can be found elsewhere [30]. In brief, 69 SNPs described by the International AMD Genomics Consortium [19] were successfully genotyped. Genotyping was performed using single-molecule molecular inversion probes and next generation sequencing to target SNPs and coding and splice-site regions of 10 AMD-related genes (ARMS2, C3, C9, CD46, CFB, CFH, CFI, HTRA1, TIMP3 and SLC16A8) and three genes associated with inherited macular dystrophies (ABCA4, CTNNA1, and PRPH2). Ten SNPs were genotyped by Kompetitive allele-specific polymerase chain reaction (KASP) genotyping assays to ensure a full genotyping of the 52 variants identified by the International AMD Genomics Consortium (IAMDGC).

GRS

The GRS was calculated for each participant as the sum of the number of risk alleles \(_\), at each variant, multiplied by the risk allele effect size \(_ (log odds-ratio)\)from the genome-wide association studies (GWAS) of the IAMDGC fully conditioned analysis, according to the formula \(GRS=_^\left(_}_\right)\).

The 52 variants identified by Fritsche et al. [19] used to calculate the GRS are listed in Additional file 2: Table S2, along with the odds ratios (OR).

Missing genotypes were not replaced by substitute values, meaning no genotype data imputation was performed. Only subjects with all the major risk variants genotyped (CFH rs570618, CFH rs10922109, C2/CFB/SKIV2L rs429608, ARMS2/HTRA1 rs3750846 and C3 rs2230199) [19] were considered for the GRS computation. In the absence of at least one major risk variant, GRS was considered null, and the subject was excluded from this analysis. A high GRS was considered equal or superior to the median GRS of the population.

Statistical analysis

General (demographic, environmental, lifestyle, diet and genetic) characteristics between cases and controls were compared using Mann-Whitney U test and Pearson’s chi-squared test (or Fisher’s exact test) for continuous and categorical variables, respectively. Categorical variables were presented with absolute frequency and percentage, while continuous variables were presented with mean and standard deviation.

The associations of AMD with the mediSCORE and the GRS were estimated with individual eye as the unit of analysis, using a logistic regression model with generalized estimating equations accounting for inter-eye correlation and adjusted for other risk factors, such as age (60–70 years, 70–75 years, > 75 years), sex, smoking (non-smokers, smokers/ex-smokers) and physical exercise. An exchangeable correlation structure was used.

Next, we investigated the interaction between the mediSCORE and the GRS on AMD, to find if the effect of one of the risk factors is different across strata of the other factor. We evaluated the interactions between these two risk factors on a multiplicative and an additive scale. It is recommended to address both multiplicative and additive interactions, although evidence of additive interaction is considered more clinically relevant in epidemiological studies than multiplicative.

We used multiplicative interactions to find if the GRS may multiply the magnitude of the risk to AMD associated with the adherence to the Mediterranean diet. In other words, to see if the effect of the combination of the two risk factors is greater than the multiplication of their individual effects. To do so, a multiplicative term of the two risk factors was added into the logistic regression model described above, adjusted for covariates, and considering the correlation between eyes. Deviations from the multiplicative interaction were tested based on the Wald test statistics.

The other strategy was to compute additive measures to determine whether the combined effect of genetic risk and adherence to Mediterranean diet was more than the sum of the risk of each individually. We computed the relative excess risk due to interaction (RERI), and also two more additive measures for completeness and consistency: the attributable proportion due to interaction (AP) and synergy index (SI). RERI is the part of the total effect that is due to interaction and can be calculated as the difference between the risk when both factors are present and the sum of the risks for each of the two factors (RERI = OR11 − OR01 − OR10 + 1), where OR11 refers to the exposure to both factors and OR01 and OR10, to the exposure to only one of them. AP can be interpreted as the proportion of effect that is due to interaction among subjects exposed to both factors: AP = RERI/OR11 and SI is the ratio between combined and individual effect (OR11 − 1)/[(OR01 − 1)+(OR10 − 1)].

Following Knol and Vanderweele recommendations for reporting analyses of effect modification/interaction [31, 32], an improvement on earlier proposals as Strengthening the Reporting of Observational Study (STOBE) was made, and besides the additive and multiplicative measures, we also presented the ORs for the adherence to the Mediterranean diet within strata of the GRS. Additive measures were developed for risk factors rather than preventive factors, so we recoded the mediSCORE protective factor so that the stratum with the lowest risk became the single reference category, when both factors are considered jointly [31, 32]. Therefore, the joint reference category reflects the subgroup of subjects with high mediSCORE and low GRS.

The additive measures RERI, AP and SI were calculated in terms of ORs, with 95% confidence intervals (CI) and P values, by two methods, the delta-method and bootstrapping. When there is no interaction, i.e., the two effects are purely additive, the RERI and AP values equals 0 and SI values equals 1.

Significance level was set to 0.05. All statistical analyses were performed using Stata (16.1, StataCorp LLC, College Station, TX, USA) and R Statistical Software (v4.0.2; R Core Team 2020).