Details of the study design and methodology used have previously been published.16 In brief, the Aragon Workers Health Study is a prospective cohort aimed at investigating the determinants of the development and progression of subclinical atherosclerosis in a middle-aged population. The study population consisted of a random sample of 5690 employees of the General Motors Spain automobile-assembly plant located in Zaragoza (Spain) who were free of cardiovascular disease at baseline.16 Each year, a random one-third of the study participants aged 40 to 55 years are selected for subclinical atherosclerosis imaging, clinical and physical examination, and diet, behavior, and lifestyle assessment. The present cross-sectional study was carried out in a subsample of 1593 participants with complete dietary data at baseline. Of these, 104 participants with extreme values for total energy intake (< 800 or > 4200 Kcal, and < 500 or > 3500 for men and women, respectively),17 and 199 participants with missing data were excluded. The final sample available for analysis consisted of 1290 participants. The study was approved by the central Institutional Review Board of Aragón CEICA (Comité Ético de Investigación de Aragón), and all study participants provided written informed consent.16

Habitual food intakes over the past 12 months were collected through a validated 136-item food-frequency questionnaire, administered by trained dietician.18,19 The frequency of consumption varied from “never or almost never” to “more than 6 times per day”. Individuals’ total energy and nutrient intakes were derived through a standardized nutrient database (ENDB).20 Using this data, factor analysis was used to determine the main dietary patterns prevalent in our population. Furthermore, to validate the results of factor analysis, previously reported diet quality indices (AHEI [Alternate Healthy Eating Index],21 aMED [alternate MD Index],22 MEDAS [MD Adherence Screener],23 and the recently developed MEDLIFE [MEDiterranean LIFEstyle Index]24) were computed. The details of the indices’ development and scoring systems are described elsewhere.21–24

At baseline, participants provided a clinical history, including the occurrence of any clinical events and hospitalizations over the past year, indicating the presence of a personal or family history of early cardiovascular disease, current medication use, and, if diagnosed, hypertension, diabetes, or dyslipidemia. Seated resting blood pressure was measured by using an OMRON M10-IT (OMRON Healthcare Co Ltd; Japan) automatic oscillometric sphygmomanometer. Three measurements were taken and the average of the measurements was used for the analysis. Blood and urine samples were collected at baseline and were processed and stored for further analysis and biobanking. Fasting serum glucose, TG, total cholesterol, and HDL-C concentrations were measured by spectrophotometry (Chemical Analyzer ILAB 650, Instrumentation Laboratory). Low-density lipoprotein cholesterol was calculated using the Friedewald formula. Levels of HDL-C of ≥ 40mg/dL and ≥ 50mg/dL for men and women, respectively, were considered optimal.25 Serum ApoA1, B100, and lipoprotein (a) were measured by kinetic nephelometry (Immunochemistry Analyzer IMMAGE 800, Beckman Coulter), and fasting serum insulin by immunoenzymatic chemiluminiscence (Access Immunoassay System, Beckman Coulter). Whole blood glycated hemoglobin was measured by reverse-phase cationic exchange chromatography and quantification by double wave-length colorimetry quantification (Analyzer ADAMS A1c HA-810, Arkray Factory). The HOMA (Homeostatic Model Assessment) index was used to assess insulin resistance using glucose and insulin data.26

Leisure time physical activity was assessed using the Spanish-validated version27 of the Nurses’ Health Study and Health Professionals’ Follow-up study physical activity questionnaires.28,29 Participants were asked about the average weekly time spent on 17 different types of physical activity, which was multiplied by its typical energy expenditure, expressed in metabolic equivalent transfer units,30 and summed over all activities, to estimate the total level of physical activity spent per week.

Anthropometric measurements of body weight, height, and waist circumference were performed at baseline following standardized procedures.31 Data were also collected on baseline sociodemographics, education, smoking history, and employment.

The main dietary patterns were determined by factor analysis by deriving factor loading for predefined food groups using Varimax rotation option. Combinations of eigen values, the scree plot, and interpretability were used to determine the number of factors retained. Each factor had an eigen value > 0.3. Factor scores were computed for each participant for each dietary pattern by summing intakes of food groups weighted by their factor loadings. Based on the score, participants were then divided into quintiles of adherence into the specific dietary pattern. To describe baseline characteristics, categorical variables are presented as count and percentage, and continuous variables as mean (standard deviation). The P-trend was tested using the factor adherence as a continuous term in the regression model. The consistency of the factor analysis-derived patterns was tested by comparing factor loadings with a priori-defined scores, namely the AHEI, the aMED, the MEDAS, and the recently developed MEDLIFE by studying the strength of the association across quintiles of factor score. Linear regression analyses were conducted between plasma lipid concentrations and dietary pattern scores after controlling for the following possible confounders: age, sex, education level, dietary energy, physical activity level, plasma lipid-lowering medication, and body mass index. The odds ratio for decreased HDL-C concentration was assessed through logistic regression (adjusted for the same possible confounders as in linear regression) analysis across quintiles of dietary patterns (the first quintile was set as a reference). STATA 12.0 (StataCorp LP; College Station, Texas, United States) was used for all statistical analyses.

Food items from the food frequency questionnaire were classified into 17 main food groups (Table 1 supplementary material). Based on the factor loading of the food groups, 2 main dietary patterns were identified (Table 1). The first dietary pattern was characterized by higher intakes of vegetables, fresh fruits, nuts, fish, olive oil and, to a lesser extent, regular consumption of nonfat dairy products and white meat and was named the “Mediterranean dietary pattern” (MDP). The second dietary pattern was characterized by higher intakes of cereals, red meat, full-fat dairy products, fast food, desserts and sweets, and, to a lesser extent, by regular consumption of vegetable oils, soda, coffee, tea and wine/beer and was named the “Western dietary pattern” (WDP). The food group of legume pulses contributed to both patterns and was not considered determinant. These 2 patterns accounted for 22% of the variance of total food intake.

The analysis of the association between the 2 dietary patterns identified in our population and previously reported dietary indices capturing a healthy diet/lifestyle and the Mediterranean diet is shown in Table 2. All indices were positively associated with the MDP, indicating that those participants with higher adherence to the MDP also scored highly in distinct a priori indices. In contrast, all indices were inversely associated with the WDP, suggesting that this pattern is indeed associated with lower quality diets. This association between a priori and a posteriori-derived dietary patterns provides additional reliability to the results.

Baseline characteristics according to quintiles of the 2 major dietary patterns (MDP and WDP) among the 1290 participants are shown in Table 3. As energy intake was not considered during factor analysis, adherence to both factors increased with energy intake due to the wider variety of foods consumed by those with higher caloric intake. On average, those participants in the highest quintile of adherence to the MDP were slightly older (P = .01), were either less likely to currently smoke (P < .001) or were more likely to be a former smokers (P < .001), were more physically active (P < .001), and had higher energy intake (P < .001) compared with those in the lowest quintile. Adherence to none of the patterns was associated with cardiovascular risk factors or biochemistry indicators, either across the quintiles of distribution (P-trend > .05) or when we compared the samples in extreme quintiles (P > .05). In contrast, participants with the highest adherence to the WDP were slightly younger men, belonged to families of 3 or more members, and had a lower education level (P < .05). No differences were found for cardiovascular risk factors except for a lower percentage of medication use (P < .05 for all medication) and a higher prevalence of smoking (P < .001) among those adhering more closely to the WDP.

Plasma lipid concentrations across quintiles of adherence to the MDP and WDP are shown in Table 4. Fully adjusted linear regression models were used for comparison. On average, participants adhering more closely to the MDP had higher HDL-C (quintile 1 = 51.5 mg/dL; quintile 5 = 54.8mg/dL; P < .001) and a lower TG/HDL-C ratio (quintile 1 = 3.38; quintile 5 = 2.95; P = .043). In contrast, participants who scored high on the WDP had significantly (P < .001) lower HDL-C (quintile 1 = 54.5mg/dL; quintile 5 = 49.9mg/dL), and serum ApoA1 (quintile 1 = 147mg/dL; quintile 5 = 139mg/dL; P = .005). Figure shows the risk of having HDL-C lower than 40mg/dL for men and lower than 50mg/dL for women across quintiles of adherence to the WDP and MDP. The risk of having lower HDL-C increased with greater adherence to the WDP (quintile 5 vs quintile 1, odds ratio = 3.19; 95% confidence interval, 1.36-7.50; P-trend = .03). Regarding the MDP, the trend across quintiles was not significant (quintile 5 vs quintile 1, odds ratio = 0.603; 95% confidence interval, 0.329-1.100; P-trend = .203).

Figure.

Odds ratio and 95% confidence interval for low high-density lipoprotein cholesterol levels across quintiles of adherence to the main dietary patterns. MDP, Mediterranean dietary pattern; WDP, Western dietary pattern.

(0.08MB).

An advantage of our study is the validation of an a posteriori-defined dietary pattern through comparison with a priori defined indices of diet quality. Based on our results, it can be concluded that factor analysis-defined dietary patterns are valid tools to assess the relationship between overall diet quality and biomarkers of diseases. In relation to the associations found, and due to the cross-sectional design of our study, a causal relationship cannot be established. In addition, this sample may not be representative of the general population, because these participants were mostly men and active workers and therefore healthy or at least without disabling diseases. As in most diet-related studies, reporting biases due to social desirability of overreporting healthier food items and underreporting less favorable foods also cannot be ruled out.