Study population
The National Health and Nutrition Examination Survey (NHANES) is a large, complex, multistage, survey of noninstitutionalised US population conducted by both the Center for Disease Control (CDC) and the National Center for Health Statistics to provide nationally representative estimates on the health and nutritional status [24]. This cross-sectional analysis included participants ≥ 60 years old from 2011–12 and 2013–14 cycles. The NHANES protocol was approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board. All the NHANES participants provided with informed consent [25].
A total of 3632 participants ≥ 60 years old were assessed in the two NHANES cycles included in this analysis. Among them, 2934 individuals underwent cognitive assessment. Participants were further excluded due to missing data for dietary intake (221), leaving 2713 older people for analysis (Supplementary Fig. 1).
Dietary assessment and estimated UPF intake
Dietary intake was assessed using two, non-consecutive 24-h dietary recalls. Data were collected by trained interviewers using the United States Department of Agriculture (USDA) Automated Multiple-Pass Method. The first dietary interview was conducted in person, followed by a second interview administered via phone 3–10 days later [24]. This study included participants who have reliably completed at least one of the dietary recalls. Dietary intake was reported as the average intake from both 24-h recalls when 2 days of data were available, and as day 1 otherwise.
In the dietary interviews, participants provided information on the type and amount of food and beverages they consumed in the previous day. The items recorded in the 24-h recalls were classified according to NOVA, a food classification based on the extent and purpose of industrial food processing, into four mutually exclusive groups: (1) unprocessed or minimally processed foods, (2) processed culinary ingredients, (3) processed foods, and (4) UPFs. Foods were classified into each of the four groups based on the variables “Main Food Description”, “Additional Food Description”, and “SR Code Description” from the NHANES 24-h recall datasets. Classification could be modified according to the variables “Combination Food Type” and “Source of food”. As such, most foods described as “Frozen meals” or “Lunchables”, as well as some items described as consumed in “Restaurant fast food/pizza” or acquired at a “Vending machine” were classified as UPFs. If an item was considered to be a hand-made recipe, the NOVA classification was applied to each of the underlying ingredients (Standard Reference Codes) as previously described [26]. For this study we used Food Codes energy values as provided by NHANES. For hand-made recipes, we calculated the underlying ingredient (Standard Reference Codes) energy values using variables from both the Food and Nutrient Database for Dietary Studies 6.0 [27] and 2013–2014 [28] and USDA National Nutrient Database for Standard Reference, release 26 [29] and 28 [30]. Classification was revised independently by two researchers and discrepancies were resolved by consensus. In this study, the exposure measure was the mean dietary contribution of UPF to total energy intake.
Cognitive assessment
Trained interviewers administered the cognitive tests at the beginning of the face-to-face private interview. The NHANES uses the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) Word Learning test, the Animal Fluency test, and the Digit Symbol Substitution test (DSST) to assess different cognitive domains. The CERAD Word Learning test assesses immediate and delayed recall of new verbal information, a component of the memory domain; the Animal Fluency test evaluates categorical verbal fluency (executive function); the DSST assesses processing speed, sustained attention, and working memory. The two parts of the CERAD Word Learning test consist of (1) three consecutive learning trials, where the participant is requested to recall a list of ten unrelated words immediately after their presentation. Each word corresponds to one point, and the result is presented as a total score across the three trials (range 0–30); and (2) a delayed word recall test, performed after the two other cognitive tests. The result ranges from 0 to 10. For the Animal Fluency test, the participant is requested to name as many animals as possible within a 60-s time period. Each animal corresponds to 1 point and the result is presented as the total sum of points. For the DSST, the participant is presented a single sheet of paper where they are asked to match a list of nine symbols to numbers according to a key located on the top of the page. The task had 133 numbers and the participant had 2 min to complete it. The result is shown as the total number of correct matches. For all the tests, higher scores represent better cognitive function. All these tests have been validated in large epidemiological and clinical studies.
Covariates
Demographics (age, gender, ethnicity, education, and poverty–income ratio), lifestyle information, and health history were collected by trained interviewers using a Computer-Assisted Personal Interviewing (CAPI) system. Ethnicity was categorized as Mexican American, Non-Hispanic white, Non-Hispanic black, or other/multiracial. Education attainment was categorized as incomplete high school, high school graduate, incomplete college, or college graduate. Poverty–income ratio, a measure that considers the ratio of household income to the poverty threshold after accounting for inflation and family size, was categorized into < 1.30 (poorer) or ≥ 1.30 (richer). Smoking status was categorized as smoker or non-smoker at the time of testing. Physical activity was assessed with a specific questionnaire (Global Physical Activity Questionnaire). Estimates of daily moderate and vigorous physical activity were calculated by multiplying the frequency per week by the duration (min) of physical activity divided by seven. Z scores for moderate/vigorous physical activity were then calculated using the mean and standard deviation of the NHANES ≥ 60 years sample. History of CVD was identified if the participant self-reported being informed by a physician about having congestive heart failure, coronary heart disease, angina pectoris, heart attack, stroke, high blood pressure, or high cholesterol levels as previously described [31]. Diabetes status was identified if the participant presented with glycohaemoglobin (Hb1Ac) ≥ 6.5% (Hb1Ac was measured in blood by the Tosoh Automated Hb1Ac Analyzer HLC-723G8) or if the participant self-reported being previously diagnosed with diabetes by a physician. Depression was identified with the Patient Health Questionnaire, a nine-item screening instrument that assesses the frequency of depression symptoms over the past 2 weeks. Four response categories were available for each question: "not at all", "several days", "more than half the days", and "nearly every day". The answers were assigned a point ranging from 0 to 3, and the final result corresponded to the sum of the points (range 0–27). Depression was defined as ≥ 10 points [32]. Participants underwent measurement of height and weight by trained interviewers to calculate body mass index (BMI). BMI was calculated as weight (kg) divided by squared height (m2), and then rounded to one decimal place. Participants were categorized according to BMI as follows: underweight (< 18.5 kg/m2), normal weight (18.5–24.9 kg/m2), overweight (25–29.9 kg/m2), and obese (≥ 30 kg/m2) as per the CDC cut-offs.
Statistical analysis
Demographic and clinical characteristics were presented as mean with standard error (SE) for continuous variables, or % weighted (SE) for categorical variables. Data were compared across the tertiles of the dietary contribution of UPF (% of total energy intake; %kcal UPF) using linear regression models for continuous variables or Pearson’s chi-square for categorical variables.
We used the restricted cubic spline in the unadjusted regression models (Model 1) with five knots (5th, 27.5th, 50th, 72.5th, and 95th) as per Harrell’s recommendations [33] to examine the shape of the dose–response relationship curve between %kcal UPF (as a continuous variable) and cognitive test scores.
Cognitive test scores were compared across tertiles of the dietary contribution of UPF (%kcal UPF) using linear regression models. Further to unadjusted models (Model 1), three adjusted models were tested: (1) Model 2: adjusted for demographics (age, gender, ethnicity, education, and poverty–income ratio); (2) Model 3: demographics and lifestyle factors (physical activity, smoking status); (3) Model 4: demographics, lifestyle factors, BMI, and chronic diseases (history CVD, diabetes, and depression). Tests of linear trend were carried out by treating tertiles as a continuous ordinal variable. To examine potential differences in the association between UPF and cognitive scores by gender, age, BMI, diabetes, and history of CVD, Wald F tests were used to evaluate interaction terms in the fully adjusted model (Model 4). We tested the interaction with history of diabetes and CVD because individuals with chronic diseases are commonly recommended to improve their diet quality, which could result in reverse causality and bias results towards the null. Thus, we further performed the regression models among individuals with and without chronic conditions (history of CVD or diabetes) separately.
Sample weights provided by NHANES for the first day 24-h recall were used to account for the complex NHANES survey design including nonresponse and oversampling. Statistical analysis was performed with STATA/SE 16.0 for Windows (StataCorp LLC). Statistical hypotheses were tested using a two-tailed P ≤ 0.05 level of significance.