Food-group and nutrient-density intakes by Hispanic and Latino backgrounds in the Hispanic Community Health Study/Study of Latinos¹²³

Sample population

The Hispanic Community Health Study/Study of Latinos (HCHS/SOL) is a population-based cohort study designed to identify risk factors and disease-prevalence rates of Hispanic and Latino populations residing within 4 communities in the United States (Miami, Bronx, Chicago, and San Diego) and representing individuals with origins from Cuba, the Dominican Republic, Mexico, Puerto Rico, and Central and South America. HCHS/SOL participants were selected by using a probability sampling design within these areas to provide a representative sample of the target population [ie, all noninstitutionalized Hispanic and Latino adults aged 18–74 y and residing in the defined geographical areas (census block groups) across the 4 participating field centers]. The sample design and cohort selection has been previously described (10). Briefly, a stratified 2-stage area probability sample of household addresses was selected in each of the 4 field centers. The first sampling stage randomly selected census block groups with stratification on the basis of Hispanic and Latino concentrations and proportions of high and low socioeconomic status. The second sampling stage randomly selected households with stratification formed by whether the occupant had a Hispanic surname from US Postal Service registries that covered the randomly selected census block groups. Both stages oversampled certain strata to increase the likelihood that a selected address yielded a household with at least one self-identified Hispanic or Latino adult. Disproportionate stratified sampling was implemented in both stages of the sample selection to increase the likelihood that a selected address would yield an eligible Hispanic or Latino household. Moreover, oversampling of the 45–74-y age group (n = 9714, 59.2%) was done to facilitate an examination of cardiovascular outcomes in older adults (10). Overrepresentation of this age group required subsampling households or persons within households during screening. Originally, households were kept intact, but 6 mo into recruitment, adults were subsampled within a household to screen fewer households. Hence, the 45–74-y-old subclusters were selected with certainty (probability = 1), whereas the 18–44-y-old subclusters were selected with a probability <1. To recruit individuals, a letter was sent to selected households, individuals within the household were contacted and screened for eligibility (living in the household, age from 18 to 74 y, ability to attend a clinic visit, and not planning to move within 6 mo) (11). Eligible individuals visited the field-center clinic where informed consent was obtained, and study assessments were completed. The study enrolled 16,415 Hispanics and Latinos between 8 March 2008 and 30 June 2011. The baseline data collection included a medical history and physical examination as well as assessments of acculturation, health behaviors (including diet and physical activity), and health care access (11).

Ethics

HCHS/SOL protocols were approved by the institutional review boards at each field center, the University of North Carolina (Coordinating Center), and the National Heart, Lung and Blood Institute.

Dietary data collection

Diet was assessed by using two 24-h recalls and a food propensity questionnaire (FPQ). Both the 24-h recall and FPQ have similar limitations because of the self-reported nature of data that result in a substantial systematic bias in the direction of underreporting (12, 13). Methods developed by Subar et al (14) combine data from the 2 instruments to estimate a population's distribution of usual food intake. Frequencies from a FPQ are used to enhance the predictive power of the 24-h recall data (15). The FPQ is similar to a food-frequency questionnaire but without portion sizes. The instrument makes it possible to distinguish individuals who are nonconsumers from consumers of specific foods and, thus, allows for the calculation of the probability of consumption. For example, it is likely that a food that is not eaten very frequently may be not consumed and reported during the two 24-h periods covered by dietary recalls. The FPQ allows for the identification of these less-frequently eaten foods to be included in estimations of usual intakes.

One 24-h food recall was conducted in person at the time of the baseline interview and the other recall was conducted via telephone ≤30 d of the baseline interview. Interviews were conducted in Spanish (80% of participants) or English (20% of participants) depending on the participant's preference with the Nutrition Data System for Research (NDS-R) software (version 11) developed by the Nutrition Coordinating Center at the University of Minnesota, which uses the multiple-pass method (16). A food-amounts booklet was provided to each participant for use in estimating portion sizes at the subsequent telephone interview. The NDS-R version 2011 contains over 18,000 foods, 8000 brand-name products, and many Hispanic and Latino foods. The software provides values for 139 nutrients, nutrient ratios, food-group serving counts, and other food components. Virtually all participants (99%) provided at least one recall.

The FPQ, which was administered at the 1-y follow-up call, asked participants to report frequencies of foods eaten in the previous year (17). We adapted the FPQ developed by the National Cancer Institute (NCI) for this population following a similar methodology used to create the tool (18), and the FPQ was shortened to reduce a participant burden (time). This reduction entailed the elimination of foods and food groups that were poorly correlated between 24-h recalls and the FPQ (r < 0.5 and P-monotone trend > 0.05) by using data from the first 1200 recalls, questions related to seasonality, and food items that could be encompassed with an overall food-group question (ie, other types of vegetables) or details of the information that would be available from the 24-h recall data (ie, type of milk). After all exclusions were applied, the FPQ was reduced from 139 food items and 228 individual questions to 115 food items and 137 individual questions that represented a 40% reduction in the number of items. This increase in efficiency was important for a study with so many other instruments competing for the limited time available during the follow-up call.

As previously mentioned, the HCHS/SOL sample size is 16,415 individuals. For this analysis, we excluded individuals with an incomplete or no follow-up interview (n = 1907), missing Hispanic or Latino background (n = 87), Hispanics and Latinos from other or mixed backgrounds (n = 503), missing dietary recalls (n = 37), recalls considered unreliable or with extreme energy intake (n = 148), or a missing FPQ (n = 448). All of these exclusions resulted in a final sample size of 13,285 individuals for analyses.

Statistics

Estimated usual intake distributions were modeled by using the NCI method (19) with SAS software macros (version 9.3; SAS Institute) developed at the NCI (20). This method allowed for the estimation of within- and between-person variance components and correction for the high intraindividual variation intrinsic to 24-h recalls because individuals do not eat the same foods every day. Recalls with energy below the sequence-sex-specific first percentile or >99th percentile or that were unreliable according to the interviewer were excluded. Models were adjusted for sex, age, Hispanic or Latino background, field center, weekend (including Friday), self-reported intake amount (more, same, or less than usual amount), and sequence (first recall conducted in person or second recall majority conducted by phone). A one-part nonlinear mixed model was used for all nutrients (except alcohol) and food groups consumed daily by almost everyone (19) and operationalized by both recalls at <5% (unweighted) with zero intake, and zeros were replaced with one-half the minimum observed nonzero value. The Box-Cox exponential power variable and covariate effects were estimated at the same time during the model fitting so that the best transformation was chosen after adjustment for these covariates. A 2-part model was used for all food groups, except for vegetables (all and other), grains (refined and all), meat, milk, diet beverages, sugar (all), and fat (from oils and all), which were estimated by using an amount-only model because these foods were almost universally consumed by everyone. The first part of the model estimated the probability of consumption by using logistic regression with a person-specific random effect, and the second part specified the consumption-day amount by using a linear regression on a transformed scale, also with a person-specific effect. Person-specific random effects were allowed to be correlated across the 2 parts because the probability of consumption is often related to the amount consumed (21). The same covariates were specified in both parts of the model, and the corresponding food group from the FPQ was included to improve estimates for episodically consumed foods. Although repeated 24-h recalls capture the natural day-to-day (intra-individual) variation in dietary intake, the FPQ captures the consumption of episodically consumed foods; hence, the FPQ can substantially improve the power to detect associations between dietary intakes as predictor variables and health outcomes, with a relatively greater impact for those foods with a large proportion of zero intakes from 24-h recalls (21).

Because the HCHS/SOL uses a complex, multistage, probability design, a variance estimation was carried out via the balanced repeated replication (BRR) (22) technique with Fay's adjustment (23). Primary sampling units were first combined randomly into 2 pseudo–primary sampling units per strata. Twenty-four replications for BRR were used. For each replicate, Fay's method used a Fay coefficient 0.5 to impose a perturbation of the original weights in the full sample that was gentler than the use of only one-half samples as in the traditional BRR method. Furthermore, distributions of usual intakes were adjusted to the overall mean age (42.4 y) and percentage of women (53.5%), which accounted for differences in age and sex distribution in Hispanic and Latino backgrounds. We tested the null hypothesis that usual mean intake was equal in backgrounds, and when significant, we performed pairwise comparisons by adjusting for multiple comparisons by using the Tukey-Kramer method (unplanned and unbalanced design) with α = 0.05. All analyses accounted for the complex survey design and sampling weights by using SAS (version 9.3) corresponding procedures or SAS-callable SUDAAN (version 10) software package (RTI International).

Total energy and macronutrient intakes

Age- and sex-adjusted energy intake distributions are shown in Figure 1. Usual mean intake was significantly different in Hispanic and Latino backgrounds (overall test P < 0.001). Cubans had higher mean total energy intakes than those of other backgrounds (2235 ± 14.2 kcal), whereas Dominicans reported consuming the lowest intake (1708 ± 18.7 kcal). Because of these differences in energy intake, we present selected nutrients standardized as a ratio per 1000 kcal in Figure 2 (see Supplemental Table 1 under “Supplemental data” in the online issue) to illustrate differences in nutrient intakes that did not reflect total energy intake. Total fat intake was higher in Cubans followed by Puerto Ricans and Mexicans; lowest intakes were reported by Dominicans. Cubans had higher intakes of monounsaturated and polyunsaturated fats, whereas Puerto Ricans had higher saturated fat intakes. Carbohydrate intake was higher in Dominicans followed by Central Americans, South Americans, and Mexicans; lowest intakes were reported by Cubans. However, daily protein intake varied little by Hispanic or Latino background, ranging from 41 to 43 g/1000 kcal.

The macronutrient composition of the diet expressed as a percentage of energy is shown in Figure 3. The percentage of energy from fat ranged from a high of 32% in Cubans to a low of 28.3% in Dominicans. The percentage of energy from protein varied little across backgrounds (16.6–17.6%), whereas the percentage of energy from carbohydrates was ∼52% in Dominicans and Central and South Americans and lower, closer to 50%, in Cubans and Puerto Ricans.

Micronutrients

Mexicans had higher intakes of vitamins A and C, folate, calcium, and iron whereas Cubans had lowest intakes except for of vitamin C, for which Puerto Ricans had the lowest intake (Figure 2; see Supplemental Table 1 under “Supplemental data” in the online issue).

Sodium, potassium, and fiber

The mean sodium intake was 1705 mg/1000 kcal (see Supplemental Table 1 under “Supplemental data” in the online issue). Cubans tended to have higher intakes, whereas Mexicans had lower intakes, than those of other backgrounds. Potassium intake was higher in Dominicans and Mexicans, whereas Puerto Ricans had lowest intakes of backgrounds (Figure 2; see Supplemental Table 1 under “Supplemental data” in the online issue). Fiber intake was highest in Mexicans (11 g/1000 kcal) followed by Central Americans and Dominicans, whereas Puerto Ricans and Cubans had lowest intakes (∼7.8 g/1000 kcal).

Food groups

The previously discussed nutrient intakes were reflective of food-group intakes by Hispanic backgrounds shown in Table 2 (see table footnotes for serving-size definitions). Dominicans and South Americans had higher total fruit intake (∼2 one-half servings/d) of which citrus made up from one-quarter to one-third of fruit intakes, whereas Puerto Ricans had the lowest total fruit intake (1.4 servings/d of total fruit) with one-third of intake from citrus fruit. Total vegetable intakes ranged from two 0.5 to ∼4 servings/d, with Cubans having the highest and Dominicans and Puerto Ricans having the lowest intakes. The proportion of vegetables that came from nutrient-rich sources such as dark-green or orange vegetables was low and accounted for only 3–10% of the total intake each, whereas tomatoes made up ∼10–19% of total vegetable intake. On average, the proportion of vegetable intake from tomatoes (17%) was greater than that from white potatoes (8%) or starchy vegetables (5%).

Grain intakes ranged from 5 (Dominicans) to 7.3 (Mexicans) servings/d. Refined grains made up 77%, on average, of total grain intake, and this proportion did not vary much across backgrounds (mostly 88–95%), with the exception of Mexicans who had much higher intake of whole grains (close to 44%) and less refined-grain intake.

Red meat made up approximately one-third of total meat intake, and intakes were higher in Cubans reaching almost 3 servings (1 serving = 1 oz or 28.3 g)/d, whereas Mexicans, Central Americans, and South Americans had, on average, 2 servings/d, and Dominicans had ∼1 serving/d. Poultry consumption was ∼2 servings/d for all groups, with the exception of Mexicans who consumed ∼1.5 servings/d. Fish intake was between 0.5 and 1 serving/d across all Hispanic and Latino backgrounds, whereas egg intake was close to one-half of a serving, and nut and seed intake was ∼0.25 serving/d with slighter higher intakes in Mexicans.

On average, the daily consumption of dairy products was ∼1.6 servings/d, with Cubans having higher intakes and Dominicans lower intakes than those of other backgrounds. Milk as opposed to cheese, yogurt, or a dessert contributed most of the dairy-product intake.

Mean servings of fats (eg, margarine, butter, oils, creams, and salad dressings) ranged from a high of 6.7 servings/d in Cubans to a low of 3.9 servings/d in Dominicans.

Intakes of sugar-sweetened beverages ranged from 1.1 servings/d in Dominicans to a high of 1.8 servings/d in Puerto Ricans. In contrast, diet beverages were not consumed much by Cubans (0.3 servings/d) but averaged ∼1 serving/d for all other Hispanic and Latino backgrounds.

Mean alcohol consumption was 0.3 servings/d, which translated to 4 fl oz (0.12 L) beer or 1.5 fl oz (0.04 L) wine/d. Cuban and Mexicans had intakes at the higher end (∼0.4 servings/d), whereas intakes for Dominicans and Central Americans were almost one-half of the overall mean daily serving.