Participants
The Women’s Health Initiative (WHI) enrolled 161,808 postmenopausal women across 40 WHI clinical centers nationwide between October 1, 1993, and December 31, 1998. Participants in the WHI study ranged in age from 50 to 79 years and were either randomized into three clinical trials (Hormone Therapy (HT) Trial, the Diet Modification (DM) Trial, and the Calcium and Vitamin D (CaD) Trial) or enrolled into an observational study (OS) [17]. The current study included 6,484 postmenopausal women who self-identified as Hispanic. Participants were excluded for a history of coronary heart disease (CHD) or stroke at baseline (N = 61), assignment to the treatment arm in the DM Trial (N = 751), energy consumption < 600 kcal/day or > 5000 kcal/day (N = 415), use of nonsteroidal anti-inflammatory drugs (N = 992), and missing genetic data (N = 857). After applying the exclusion criteria, the sample size for the analysis was 3,469 postmenopausal women (2,162 from the observational study and 1,307 from the clinical trials).
Cardiovascular disease
The study outcomes included incident CHD and stroke. CHD was defined as hospitalized myocardial infarction, definite silent myocardial infarction, or death due to coronary disease. Stroke was defined as rapid onset of a persistent neurologic deficit attributed to an obstruction or rupture of the brain arterial system, lasting more than 24 h and without evidence for other cause. Only strokes requiring hospitalization were considered outcome events for WHI10. The CVD outcomes were adjudicated and ascertained by physician review of medical records, as previously described [18].
Dietary Inflammatory Index (DII®)
Diet was evaluated with a standardized and validated self-administered food frequency questionnaire (FFQ) that was mailed to all participants at baseline [19]. The FFQ was developed to estimate the average daily consumption of 122 food items over the previous 3-month period and included information about the use of vitamin and mineral supplements. The estimation of the nutrient consumption of each participant was calculated using the University of Minnesota’s Nutrition Coordinating Center nutrient database, which is based on the U.S. Department of Agriculture Standard Reference Releases and manufacturer information [19].
The procedure used to calculate the DII scores from the FFQ responses from all subjects has been described elsewhere [4]. Briefly, the DII was created after an extensive literature review that identified 45 food parameters that were linked, with sufficiently robust literature, to six inflammatory biomarkers. Each whole food or nutrient received a food parameter-specific overall inflammatory effect score that was calculated based on the pro-inflammatory, anti-inflammatory, or null effect of that dietary component as reviewed in the scientific literature. Scoring also took into consideration the total number of articles published and the study design [4, 7].
Using the participants’ intake data as reported on the FFQ, a Z-score was calculated for each one of the available food parameters for each individual in the study based on the world average and standard deviation from a global composite dataset created for this purpose [4]. These Z-scores were converted to a proportion and then centered by doubling and subtracting 1. After this step, the centered proportion value for each food parameter was multiplied by the respective ‘overall food parameter-specific inflammatory effect score’ to obtain the ‘food parameter-specific DII score’ and finally, all the ‘food parameter-specific DII scores’ were added together to create the ‘overall DII score’ for an individual.
In the WHI FFQ, 32 of the 45 original DII components were available for inclusion in the overall DII score (Supplementary Table 1). The components ginger, turmeric, garlic, oregano, pepper, rosemary, eugenol, saffron, flavan-3-ol, flavones, flavonols, flavonones, and anthocyanidins that were included in the original DII calculation were not included in the WHI FFQ [6]. The WHI FFQ also included questions about the consumption of 15 nutritional supplements that are part of the DII components (iron, magnesium, niacin, riboflavin, selenium, thiamine, β-carotene, zinc, folic acid, and vitamins A, C, D, E, B6, and B12) [6]. Nevertheless, in our analysis the DII was calculated without considering the use of nutritional supplements.
Covariates
The study covariates were determined using baseline data from both the observational study (OS) and clinical trial (CT) components of the WHI study. Demographic information included age, ethnicity, and preferred language. Neighborhood socioeconomic status (NSES) was evaluated using a standardized geocoding protocol, which linked individual WHI participant addresses to the year 2000 U.S. Census Federal Information Processing Standards (FIPS) codes and tract-level socioeconomic data. A summary measure of each participant's neighborhood socioeconomic environment was estimated from the tract-level data using six variables representing several dimensions of wealth and income [20]. Higher NSES scores represent a higher neighborhood socioeconomic status.
Risk factors for CVD were also included: physical activity, smoking status, acculturation, alcohol intake, body mass index (BMI), genetic admixture, and other chronic diseases. Physical activity was evaluated with a validated physical activity questionnaire, and it was included in the model as the total minutes of recreational physical activity per week, including walking, mild, moderate, and strenuous physical activity [21]. Smoking status was determined as never or past/current smoker. Language preference (English or Spanish) was used as a proxy measure for acculturation status. Alcohol intake in grams per day was estimated using a food frequency questionnaire [19]. For BMI, weight was measured to the nearest 0.1 kg on a balance beam scale. Height was measured to the nearest 0.1 cm using a wall-mounted Harpenden stadiometer. BMI was calculated as weight (kg) divided by the square of measured height (m2) [22]. Hypertension was defined as systolic pressure ≥ 130 mm Hg or diastolic ≥ 80 mm Hg or self-reported hypertension with the use of antihypertensive medication [23]. Diagnosis of diabetes at baseline was obtained from the medical history questionnaire in response to the question “Did a doctor ever say that you had sugar diabetes or high blood sugar when you were not pregnant?” [24]. Hypercholesterolemia was defined by self-report at baseline and then by use of lipid-modulating medication [23]. Obesity was defined as a BMI ≥ 30 km/m2.
Genetic ancestry is an important and often ignored factor that may affect the way that ethnic groups respond to dietary patterns and that can alter the relationship between diet-associated inflammation and risk of metabolic diseases [25,26,27,28,29]. This study includes Amerindian ancestry as a covariate to account for the genetic diversity of Hispanic women when evaluating the relationship between dietary patterns and CVD. Genetic admixture was calculated using a marker set of 92 ancestry informative markers that demonstrated large differences in allele frequency between ancestral populations from Europe, sub-Saharan Africa, and the Americas (> 45%) [30, 31]. Genotyping was performed using the TaqMan OpenArrays system (Life Technologies/Applied Biosystems, Foster City, CA, USA) [30]. Admixture proportions were determined using the Bayesian clustering algorithms implemented in the program STRUCTURE v2.1 [31, 32].
Statistical analysis
Two-sample t-tests were applied to compare the mean differences of each continuous variable between participants with and without CVD. Chi-square tests were used to compare the two groups on categorical variables. Separate Cox regression models were fit to examine the association between DII and CHD/stroke. The analyses were restricted only to follow-up events and time to CHD/stroke occurrence were the outcomes of interest. CHD and stroke status were used as dichotomous traits (0 = no, 1 = yes) as the indicator variable for failure/censorship. The survival time for participants who did not develop the CVD outcome of interest was defined as the days from enrollment to the end of follow-up (the follow-up time for CVD events in this analysis includes data until September 2018). Hazard ratios (HRs) and 95% confidence intervals (CI’s) are presented for each model. Cox regression models were conducted with and without adjusting for age at entry, lifestyle-related risk factors (smoking, alcohol intake, physical activity, and acculturation), known CVD risk factors (diabetes, hypertension, hypercholesterolemia, BMI, and Amerindian ancestry), and neighborhood socioeconomic status (NSES). These covariates were serially added to the model and the final model included all covariates. Multiple imputation with the fully conditional specification method was used to estimate missing values of the variable NSES (N = 273 which represents 7.9% of the observations) and physical activity (N = 167 which represents 4.8% of the observations) assuming that data were missing at random. A sensitivity analysis excluding women in the control arm of the dietary modification trial was performed. Finally, the interaction between BMI and the DII scores was evaluated by adding interaction terms between BMI and the DII scores in the final models. Stratified models for women in the different BMI categories (normal weight BMI 18.5–24.5 kg/m2, overweight BMI 25.0–29.9 kg/m2 and obesity BMI > 30 kg/m2) were fitted if needed. Analyses were performed using SAS software version 9.4 (SAS Institute, Cary, NC USA). All statistical tests were two-sided and p ≤ 0.05 was considered statistically significant.