Study population
This cross-sectional study recruited participants from NHANES 2011–2012 and 2013–2014 (https://www.cdc.gov/nchs/nhanes/index.htm). NHANES is a public health survey program conducted by National Center for Health Statistics (NCHS) in America. The NHANES project recruits participants using a complex, multistage probabilistic sampling design in two-year cycles. NHANES collects information from questionnaires at home, physical and laboratory examinations in mobile examination center (MEC) and telephone interviews. In this study, samples from NHANES 2011–2012 and 2013–2014 were obtained and combined because NHANES project provided the outcome of several cognitive test in these two cycles specifically. There were totally 19,931 participants in these two cycles. Our study implemented a three-stage exclusion protocol: (1) Primary exclusion of 16,530 participants based on age criterion (< 60 years), (2) subsequent removal of 478 cases with incomplete cognitive assessments or missing subjective cognitive questionnaires, followed by (3) elimination of 1,947 individuals lacking essential metrics for LE8 calculation. Finally, we get 976 eligible participants. (Supplement Fig. 5)
Measurement of LE8 score
LE8 scoring system is comprised of 8 metrics including 4 behavioral metrics (diet, physical activity frequency and duration, nicotine exposure and sleeping) and 4 biological metrics (blood lipids, blood glucose, blood pressure and BMI score) (Supplement Table 1) [7]. Total LE8 score is the average score of above 8 metrics which range from 0 to 100 with higher score indicate healthier cardiovascular condition. In our study, LE8 score was further classified by quartile into 4 groups named Q1, Q2, Q3 and Q4 with Q1 as reference category.
Of the 8 metrics, diet score was assessed according to Healthy Eating Index 2015 (HEI-2015). NHANES collects dietary data with two 24-hour recalls interviews, one is conducted in person in MEC while the other is on telephone several days later. Researchers are able to calculate the dietary intake of participants by combining 24-hour food intake files from NHANES and food patterns equivalents data from United States Department of Agriculture (USDA) (https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/docs/fped-data-tables/). HEI score includes 13 components of 2 classifications: 9 adequacy components and 4 moderation components. Scoring of dietary components is based on energy density which represents the amount of food components per 1000kcal (Supplement Table 2). Self-report questionnaires from NHANES provides information about physical activity intensity, cigarette smoking behaviors, sleeping duration, diabetes history and medication usage. As for the data of BMI, blood pressure, blood lipid and hemoglobin A1c, blood samples are obtained in MEC and then processed, stored, and transported to laboratories for test. The height and weight used to calculate BMI were measured in the MCE. The full test includes three measurements of both systolic and diastolic blood pressure. When data from all three measurements were available, the average systolic and diastolic blood pressure were calculated. If the full set of three measurements was not available, the average was computed using as many measurements as were available.
Measurement of cognitive test score and subjective cognitive performance
NHANES conducted three widely utilized [11,12,13] cognitive tests for participants aged > 60 years in the cycle of 2011–2012 and 2013–2014. First, the word learning and recall modules from the Consortium to Establish a Registry for Alzheimer’s disease (CERAD test) was conducted to assess immediate and delayed learning ability for new verbal information. The test comprises three sequential learning trials followed by delayed recall. During each trial, participants verbally articulate 10 randomly ordered unrelated words presented sequentially, followed by immediate free recall, with word sequence randomization repeated across trials to minimize order effects. A maximum score of 10 per trial is attainable based on correct item retrieval. Second, the Animal Fluency test (AFT) was performed for assessing the executive function where participants are asked to name as many animals as possible in one minute and one point is given for each named animal. Third, Digit Symbol Substitution test (DSST) was conducted which depend on the rapid processing of information, maintaining attention, and retrieving working memory. Participants are provided with a paper which contains 9 numbers with paired symbols, then they are asked to fulfill 133 boxes nearby the numbers with corresponding symbols as much as possible in 2 min. One point is given for one correct match. In this study, we calculated z-score [(individual test score – mean)/standard deviation] of immediate CERAD test, recall CERAD test, AFT test and DSST test respectively. The total cognitive test score refers to the average of above four z-scores [14]. Higher cognitive test score indicates better cognitive function. Participants who reported “being limited in any way because of difficulty remembering or because experience periods of confusion” were defined as having subjective cognitive performance. Participants were defined as having subjective cognitive performance if the participant answered “yes” to the question: “being limited in any way because of difficulty remembering or because experience periods of confusion”.
Assessment of other covariables
We included variables of demographic characteristics and health behaviors that are possibly associated with cognitive function [15], including age (continuous variables), sex (male, female), race (non-Hispanic White, non-Hispanic Black, other Hispanic, other race), education level (college graduate or above, some college or associate degree, high school/GED or less ), the ratio of family income to poverty guideline (< 1.3, 1.3–3.5, > 3.5 ) [16] and alcohol consumption (drinker, non-drinker). Alcohol drinker was defined as those who had at least 12 alcohol drinks a year [17].
Statistical analysis
NHANES selected participants with a complex multistage probabilistic sampling design thus all analysis in this study were weighted with provided weight variables: WTDRD1, SDMVPSU and SDMVSTRA. Since we combined the 2011–2012 and 2013–2014 cycle, the weights of the combined study population were calculated as 1/2* WTDRD1 (https://wwwn.cdc.gov/nchs/nhanes/tutorials/module3.aspx).
Initially, we inspected the characteristic of the study population across four quartiles of LE8 score. After examined the normality of variables’ distribution and the homogeneity of variance, continuous variables were described with mean and standard deviation [mean (SD)] and compare with Wilcoxon rank-sum test while categorized variables were described as the case amount and its percentage [n (%)] and compared with chi-squared test. Afterwards univariable regression (linear regression for cognitive test score, logistic regression for subjective cognitive performance) was conducted to select covariables that were significantly associated with cognitive performance.
Cognitive performance models
We constructed three regression models to explore the association between LE8 score and cognitive performance, both as continuous variables and categorized variables. Model 1 wasn’t adjusted. Model 2 was adjusted for age, gender, and race. Model 3 was adjusted for age, gender, race, family income, education, and alcohol consumption. We also applied restricted cubic spline (RCS) models to explore the dose-response relationship in the above three models. The 5th, 35th, 65th, and 95th percentiles of the total LE8 score distribution were chosen as the knots of the RCS curves [18]. The R² (coefficient of determination) metrics were derived from our multivariable linear regression Model3 (Supplement Fig. 6). The P– non-linear value of RCS was obtained using a likelihood ratio test (LRT) to assess whether the higher-order coefficients of RCS curve are significantly different from zero. This test was conducted using the rms package in R.
Subjective cognitive performance models
We explored the effect of individual LE8 metrics on cognitive performance in a full-adjusted multivariable regression model. In order to validate the reliability of regression model, Receiver Operating Characteristic (ROC) curves of LE8 (both categorized and continuous) were plotted (Supplement Fig. 6). To assess the predictive capacity of LE8 for subjective cognitive performance, we compared the ROC curves of LE8 in relation to cognitive test scores and subjective cognitive performance. For the ROC analysis, cognitive test score was converted into a binary variable according to the lower quartile.
Subgroup models
To identify the subpopulation that benefits most from elevating LE8 score, the study population was stratified by all the variables in Model 3. We then calculated the regression coefficients of LE8 score in different subgroups. P values for interaction were calculated using likelihood ratio tests.
Cross-validation analysis
We made stratified ten-fold cross-validation with ten independent repetitions (10 × 10 CV) on both the objective cognitive testing and subjective cognitive assessment datasets. Model performance was quantified using R², providing robust measurement of predictive consistency and variance explicability (Supplement Table 6).
All statistical tests were two-tailed and conducted with R v. 4.2.1 statistical analysis software. Adobe Illustrator v2023 was used for figure preparation. P < 0.05 was considered statistically significant.