Association between healthy eating index-2015 and various cognitive domains in US adults aged 60 years or older: the National Health and Nutrition Examination Survey (NHANES) 2011–2014

The data currently used are publicly available without personally identifiable information from NHANES (https://​www.​cdc.​gov/​nchs/​nhanes/​index.​htm). A detailed description of the NHANES study design and methods is available elsewhere [18]. Briefly, NHANES is a program that administers ongoing 2-year-cycle cross-sectional surveys, which is conducted by the Centers for Disease Control and Prevention (CDC). Based on a complex, multistage probability sampling design, NHANES examines a nationally representative sample of ~ 5000 persons each year selected from 15 different locations that are chosen from a sampling frame of all US counties. A major objective of NHANES is to estimate the health and nutritional status of the non-institutional population in the United States. In the present analysis, we combined 2 NHANES cycles (NHANES 2011–2012 and 2013–2014), because these two cycles specifically carried out a series of cognitive function testings. In total, 19,931 individuals participated in NHANES during 2011–2014, and our analysis was limited to 3632 adults aged 60 years or older who were eligible to complete the cognitive function testings. Among them, we excluded individuals with incomplete 24-h dietary recall data (n = 564), extreme total energy intake of < 500 or > 5000 kcal/day for females, and < 500 or > 8000 kcal/day for males (n = 18), and incomplete cognitive function data (n = 350). Moreover, individuals who had missing data on body mass index (BMI), sedentary time, blood pressure (BP), glycohemoglobin (HbA1c), and total cholesterol (TC) were further excluded (n = 250). Finally, 2450 individuals were included in our study (Fig. 1). Compared to the original population, study population was more likely to be younger (69.3 years VS 70.1 years), Non-Hispanic white, drinker, to have higher educational level and ratio of family income to poverty. No significant differences were observed in terms of other characteristics (Additional file 1).

Dietary intake data was collected from NHANES 24-h recall using the USDA automated multiple-pass method [19]. The 24-h dietary recall data were collected for 2 days, which were conducted by a trained interviewer face-to-face in the Mobile Examination Center on one day and a follow-up interview 3–10 days later via phone. Based on the two 24-h dietary recall data, food groups were determined using USDA Food Patterns Equivalence Database, and energy or nutrient content for food and beverage recorded were computed using USDA Food and Nutrient Database for Dietary Studies.

HEI-2015 is a diet quality index developed in partnership by researchers from USDA and NCI to assess the adherence to 2015–2020 DGA and consists of 13 components (food groups or nutrients), including 9 adequacy components (total vegetables, greens and beans, total fruits, whole fruits, whole grains, dairy, total protein foods, seafood and plant proteins, and fatty acids) and 4 moderation components (sodium, refined grains, saturated fats, and added sugars) [20]. These components are scored based on an energy density of 1000 kcal, except for fatty acids which is a ratio of unsaturated to saturated fatty acids. The specific scoring standards for each component are shown in Additional file 2. Briefly, higher intakes result in higher scores for the adequacy components, while lower intakes result in higher scores for moderation components. The component scores are scored separately and summed to compute an overall score with a maximum of 100 (the higher the scores, the higher the compliance with 2015–2020 DGA).

A series of cognitive function testings were performed among adults aged 60 years or older in NHANES 2011–2014, including Digit Symbol Substitution Test (DSST), Animal Fluency Test (AFT), and the word learning subtest from the Consortium to Establish a Registry for Alzheimer’s disease (CERAD). Participants who needed a proxy, did not pass the practice pretest, refused audio record, had communication problems, or gave up will have a missing score for the formal testings. In brief, DSST is used to assess the ability of processing speed [21]. The test provides a paper form that has a key at the top containing 9 numbers paired with symbols. Participants were asked to match corresponding symbols in the 133 boxes that adjoin the numbers in 120 s. The score is the total number of correct matches. AFT is used to evaluate the categorical verbal fluency, a component of executive function, regardless of cultural context [22]. Participants were asked to name as many animals as possible in 60s. A point is given for each named animal, and the final score represents the total number of correct named animals. CERAD word learning subtest was performed to assess both immediate and delayed memory, which consists of three consecutive learning trials and a delayed recall trial [23]. In the learning trials, participants were asked to read aloud 10 irrelevant words. The order of these 10 words is different in each of the three learning trials. After finishing the former two cognitive function testings (DSST and AFT), the delayed recall trial requiring participants to recall 10 words used in the learning trials which were conducted. In addition, a composite-z score was created by summing the z scores [(individual test score - mean score)/SD] of these three individual tests (DSST, AFT, CERAD). For all the tests, higher scores represent better cognitive performance.

The NHANES release information about socio-demographic factors, health-related lifestyles, and health conditions through demographics, questionnaires, examination, and laboratory data. We included some of them as covariates that were thought to be related to cognitive function and/or diet quality based on the previous researches [24, 25]. Socio-demographic factors included age, gender (male and female), ethnicity (Hispanic, non-Hispanic White, non-Hispanic Black, and Asian or other race), and education (less than high school, high school, and more than high school). Additionally, the ratio of family income to poverty was divided into three categories (≤1.30, 1.31–1.85, > 1.85) to reflect house income. The higher ratio indicates better family financial situation. This index was calculated by dividing the family income by the Department of Health and Human Services’ poverty guidelines, specific to family size, survey year and state [26].

As for health-related lifestyles, total time spent sitting except time sleeping (sedentary time) in a day was used as a potential physical activity indicator [27]. Smoker was defined as smoked at least 100 cigarettes in life [28]. Participants who drunk at least 12 times in the previous year were considered drinkers [29].

Concerning health conditions, the depressive symptom was assessed using 9-item Patient Health Questionnaire (PHQ-9). The score of each item ranges from 0 (not at all) to 3 (nearly every day), incorporating a total score with the maximum possible score of 27. Individuals with a PHQ-9 total score of 10 or greater were categorized with depressive symptoms [30]. BMI was calculated as weight (kg) divided by height squared (m) which were measured in Mobile Examination Center. Hypertension was defined as a SBP ≥140 mmHg or a DBP ≥90 mmHg, and/or current use of antihypertensive medication [31]. Hypercholesterolaemia was defined as TC ≥ 240 mg/dl or current use of prescribed medicine for hypercholesterolaemia [32]. Participants were classified as having diabetes based on HbA1c ≥6.5% and/or current use with insulin or diabetic pills [33].

Characteristics of the study population are described as mean ± standard deviation (SD) for continuous variables and number (percentage) for categorical variables. HEI-2015 were categorized based on quartiles (quartile 1: < 25th percentile, quartile 2: ≥25 to 50th percentile, quartile 3: ≥50 to 75th percentile, quartile 4: ≥75th percentile). We tested differences in characteristics between groups with a one-way analysis of variance for continuous variables and with a chi-square test for categorical variables. Linear regression analyses were used to examine the associations between HEI-2015 scores, both as continuous variable and categorical variable (quartiles), and cognitive function. To test for trends across increasing quartiles of HEI-2015 scores, the median of HEI-2015 scores in each quartile was calculated and used as a continuous variable. Model 1 was adjusted for age and gender; Model 2 was additionally adjusted for energy intake, ethnicity, BMI, drinking status, smoking status, sedentary time, education, ratio of family income to poverty, depressive symptom, hypertension, hypercholesterolaemia, and diabetes. To explore the dose-response relationship between HEI-2015 and cognitive function, we applied a restricted cubic spline with knots at the 5th, 50th and 95th percentiles of the distribution of the exposure in a fully adjusted model (Model 2) using SAS macro “%RCS_Reg” constructed by Desquilbet et al. [34]. We also performed a sensitivity analysis by restricting participants without a depressive symptom. Additionally, stratified analyses were conducted to test whether these associations differ by age groups (60–69 and 70~), gender (male and female), or ethnicity (Hispanic, non-Hispanic white, non-Hispanic black, Asian, or other race). For the exploratory purpose, we further examined the associations between HEI-2015 components (dichotomous) and domain-specific cognitive function adjusting for all covariates and other HEI-2015 components to identify whether statistically significant associations might be attributable to specific components. Data management and dose-response curve were performed using SAS version 9.4 (SAS Institute, Cary, North Carolina, USA) and other statistical analyses were performed using SPSS 24.0 (IBM, Armonk, NY, USA). A two-tailed P value < 0.05 was considered statistically significant.

The characteristics of the study population according to HEI-2015 quartiles are shown in Table 1. A total of 2450 adults aged 60 years or older were included for analysis. The mean age was 69.3 ± 6.8 years, and 49.8% of the participants were male. Participants with higher adherence to HEI-2015 were more likely to be older, female, Asian or other race, to have lower BMI, higher educational levels, and higher ratio of family income to poverty; while less likely to be smoker, non-Hispanic black race, diabetics, and to have depressive symptom (P < 0.05). No significant differences across HEI-2015 quartiles were observed in terms of sedentary time, energy intake, hypertension, and hypercholesterolaemia.

Table 2 presents the associations between HEI-2015 scores and cognitive performance. In an age and gender-adjusted Model 1, the highest HEI-2015 quartile was significantly associated with higher DSST score (B: 5.26; 95%CI: 3.45, 7.06; P<0.001), AFT score (B: 1.31; 95%CI: 0.71, 1.91; P<0.001), CERAD score (B: 0.98; 95%CI: 0.30, 1.66; P<0.01), and composite z-score for cognitive function (B: 0.70; 95%CI: 0.45, 0.96; P<0.001), compared to the lowest HEI-2015 quartile. There was evidence of a linear trend in all of these relations. When treating HEI-2015 as a continuous measure, each 1-unit increase in HEI-2015 score was associated with higher DSST score (B: 0.16; 95%CI: 0.11, 0.21; P<0.001), AFT score (B: 0.04; 95%CI: 0.02, 0.05; P<0.001), CERAD score (B: 0.03; 95%CI: 0.01, 0.05; P<0.01), and composite z-score (B: 0.02; 95%CI: 0.01, 0.03; P<0.001). After fully adjusting in Model 2, the associations of HEI-2015 (highest versus lowest) with AFT (B: 0.65; 95%CI: 0.09, 1.21; P<0.05) and composite z-scores (B: 0.26; 95%CI: 0.05, 0.47; P<0.05) attenuated but remained significant, whereas the associations with DSST and CERAD scores were no longer significant. We observed evidence of a linear trend in these associations of HEI-2015 scores with DSST, AFT and composite z-scores except CERAD. In addition, each 1-unit increase in HEI-2015 was associated with DSST (B: 0.05; 95% CI: 0.01, 0.09; P<0.05), AFT (B: 0.02; 95% CI: 0.00, 0.03; P<0.01), and composite z-scores (B: 0.01; 95% CI: 0.00, 0.01; P<0.01), but not associated with CERAD score.

Similar results were observed in sensitivity analyses (Additional file 3). When excluding individuals who suffered a depressive symptom (n = 177), HEI-2015 score (as a continuous variable) was still positively associated with DSST (B: 0.05; 95% CI: 0.01, 0.09; P<0.01), AFT (B: 0.02; 95% CI: 0.00, 0.03; P<0.05) and composite z-scores (B: 0.01; 95% CI: 0.00, 0.01; P<0.05) in fully adjusted model.

The dose-response relationship between HEI-2015 and domain-specific or composite z-score cognitive function after fully adjusting are shown in Fig. 2. In restricted cubic spline models, HEI-2015 scores were positively associated with DSST score (P_nonlinearity = 0.3965; Fig. 2A), AFT score (P_nonlinearity = 0.6107; Fig. 2B), and composite z-score (P_nonlinearity = 0.7546; Fig. 2C) in a linear manner. However, no significant association was observed between HEI-2015 score and CERAD score in the fully adjusted model (Fig. 2D).

Potential interaction effects were identified between HEI-2015 (treated as continuous) and ethnicity in relation to DSST score (P_interaction <0.01; Fig. 3A) and AFT score (P_interaction <0.05; Fig. 3B). HEI-2015 scores were positively associated with DSST (B: 0.08; 95% CI: 0.03, 0.14; P<0.01) and AFT scores (B: 0.04; 95% CI: 0.02, 0.06; P<0.01), respectively, among the non-Hispanic white population in the fully adjusted model, but not among Hispanic, non-Hispanic blacks, and Asian or other race. In addition, no interaction effects were observed between age group or gender and HEI-2015 scores (treated as continuous) on any domain-specific cognitive function or composite-z score in the fully adjusted model (P_interaction > 0.05; Fig. 3A-3D).

Given the significant associations between higher HEI-2015 total scores and better performance on domain-specific cognitive function assessed by DSST and AFT, an exploratory analysis was performed to test the associations of each HEI-2015 component (dichotomous) with them, adjusting for all covariates included in Model 2 as well as the other components (Additional file 4). Among 13 HEI-2015 components, higher intakes of the whole fruits (B: 1.59; 95% CI: 0.21, 2.98; P<0.05), seafood and plant protein (B: 1.46; 95% CI: 0.34, 2.58; P<0.05), and fatty acids (B: 1.28; 95% CI: 0.08, 2.48; P<0.05) were associated with better performance on DSST. Meanwhile, those with the higher consumption of whole fruits (B: 0.58; 95% CI: 0.03, 1.12; P<0.05) and seafood and plant protein (B: 0.78; 95% CI: 0.34, 1.22; P<0.001) were more likely to have better performance on AFT.