Study participants were from the Guangxi Zhuang Birth Cohort (GZBC), which was conducted in county-level hospitals of six major counties of the Guangxi province in China from June 2015 (27). From July to September 2021, we successfully followed up 325 children born to the mothers of GZBC in Debao and Pingguo county. During the follow-up study phase, face-to-face interviews with parents were conducted to investigate the consumption status of seven ultraprocessed foods (i.e., chocolate, biscuits, candy, fast-food, ice cream, SSB, and sweet bakery products) for the children using a Food Frequency Questionnaire (FFQ). The Mandarin-language version of the Wechsler Preschool and Primary Scale of Intelligence (WPPSI, Fourth Edition) was applied to assess the cognitive function of children. Among them, a total of 97 children were excluded which included 4 children who did not complete the food consumption survey and 93 children who did not finish the cognitive function assessment. Finally, 228 children were included for the further analysis. Ethical approval was granted by Guangxi Medical University, China (No. 20140305-001), and maternal informed consent was provided.

Children’s cognitive function was assessed using the Mandarin-language version of the Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (28), which has been widely used in cognitive function assessments because of its high reliability and validity (29). The specific cognitive function domains included Verbal Comprehension Index (VCI, a measure of the ability to understand, learn, and retain verbal information, as well as to use language to solve novel problems), Visual Spatial Index (VSI, a measure of the ability to understand visual information and to solve novel abstract visual problems), Fluid Reasoning Index (FRI, the test measures fluid intelligence, non-verbal concept formation, analysis, and problem-solving, and integration ability, etc.), Working Memory Index (WMI, a measure of the ability to hold verbal information in short-term memory and to manipulate the information), and Processing Speed Index (PSI, a measure of mental speed that may also be affected by factors such as attention) (30). The Full-Scale Intelligence Quotient (FSIQ) is a composite score of the above five domains scores (VCI, VSI, FRI, WMI, and PSI), which represents the general intellectual ability of the subject. The FSIQ and single index was standardized to have a mean of 100 and a standard deviation of 15, ranging from 40 to 160 (31). Higher values of the FSIQ or the single index represent a better test performance, and therefore stronger cognitive abilities. In addition, an FSIQ of less than 80 scores were defined as having a cognitive deficit (8, 32, 33). Regarding assessments, all examiners were trained professionally and certified before the test. The assessments were administered by well-trained examiners one-on-one, without any guidance from teachers or guardians in a standard and quiet assessment room.

Through face-to-face interviews, children’s consumption of common ultraprocessed foods in the past 3 months was determined from mothers’ responses to the Food Frequency Questionnaire (FFQ). The ultraprocessed foods in the present study included chocolate, biscuits, candy, fast-food (French fries or hamburgers), ice cream, sugar-sweetened beverage (SSB), and sweet bakery products. The food consumption frequency was reported as “never,” “once a week,” and “more than twice a week.” Finally, we defined “never” and “once a week” as “infrequent” and “more than twice a week” as “frequent” (34).

Sociodemographic and family characteristics were collected via a face-to-face interview questionnaire, including child sex (male or female), child age (years), residence area (rural or urban), siblings (0. ≥ 1), parental education levels (junior high school or below, senior high school or above), annual household incomes (low = ≤ 5.99 w/year, medium = 6–14.99 w/year, high = ≥ 15 w/year), outdoor exercise time (h/day), parental accompaniment time (h/day), breastfeeding duration (months), daily sleeping time (h), and secondhand smoke (no = never, infrequent = 1–2 times/week, frequent = more than 3 times/week).

Categorical variables were described in frequencies and proportions, while continuous variables were described in terms of means and standardized deviations. Two variables with missing data in our analysis included daily sleeping time (18%) and secondhand smoke (7%). They were assumed to be missing at random and imputed with the multivariate imputation by chained equations (MICE) method for five times. Continuous variables were imputed using predictive mean matching, and categorical variables using logistic regression factor (2 levels) (35). The Chi-square test and Wilcoxon rank sum test was used to determine whether there were any statistically significant differences of the covariates between cognitive deficits and non-cognitive deficits. The multiple linear regression and logistic regression models were applied to analyze the relationship between ultraprocessed food consumption and cognitive function and the risk of cognitive deficit, respectively. Specifically, two models were constructed. Crude model did not adjust any factors. Adjusted model adjusted the potential confounders. A Directed Acyclic Graph¹ was used to inspect possible pathways for confounders (36). In addition, covariates were selected a priori based on the previous literature (37). Finally, the following variables were selected as confounders: child age, residence area, siblings, parental education levels, annual household incomes, outdoor exercise time, parental accompaniment time, daily sleeping time, breastfeeding duration, and secondhand smoke (Figure 1). All statistical analyses were performed using R 4.03 software. The p-value < 0.05 were considered statistically significant. Forest plots were produced using the GraphPad Prism 8.0.1 software.

A description of the participants (n = 228) characteristics is presented in Table 1. Of the 228 children, 50.88% were boys, of which, 30.70% were frequently exposed to secondhand smoke. Most of the children were from urban environments (76.75%), have siblings (73.25%), had mothers (61.40%), and fathers (62.72%) with a senior high school education or above, and less than 2 h of outdoor exercise time per day (70.18%). Among them, 45.61% had low annual household incomes. A total of 83 (36.40%) children had cognitive deficits. The distributions of residence area, siblings, parental education levels, and annual household incomes (p < 0.05) were significantly different between cognitive deficits and non-cognitive deficits. Especially, higher incidence of cognitive deficit was observed among children who were from urban environments, had siblings, had parents with junior high school or below education level, and had lower annual household incomes.

First, we applied a multiple linear regression model to assess the association between frequent ultraprocessed food consumption (independent variables) and FSIQ scores (dependent variables) (Table 2). In the crude model, we found children who frequently consumed candy had lower FSIQ scores compared with infrequent consumers (β = −2.49, 95% CI: −4.79∼−0.18; p = 0.035). After adjusting for potential confounders, we found that frequent consumption of candy (β = −3.34, 95% CI: −5.62∼−1.06; p = 0.004) and sweet bakery products (β = −2.77, 95% CI: −5.58∼0.04; p = 0.054) were significantly associated with lower FSIQ scores.

Second, we applied logistic regression models, including all children, to examine the association between frequent consumption of ultraprocessed foods (independent variables) and risk of cognitive deficit (dependent variables) (Table 3). In the crude model, children who frequently consumed candy showed an association with an increased risk of cognitive deficit (OR = 1.61, 95% CI: 0.94∼2.78; p = 0.085), whereas children who frequently consumed biscuits showed an association with a decreased risk of cognitive deficit (OR = 0.59, 95% CI: 0.34∼1.05; p = 0.072), both close to the level of significance. After adjusting for potential confounders, however, children who frequently consumed candy was associated with a significantly increased risk of cognitive deficit (OR = 2.05, 95% CI: 1.11∼3.79; p = 0.023).

The associations between frequent ultraprocessed food consumption and the specific cognitive domains, including VCI, VSI, FRI, WMI, and PSI scores, are also shown in Figure 2. In the crude model, children who frequently consumed candy showed a significant decrease in VCI scores (β = −3.20, 95% CI: −5.66∼−0.74; p = 0.011) while a not fully significant decrease in PSI scores (β = −2.63, 95% CI: −5.64∼−0.37; p = 0.085). However, children who frequently consumed biscuits showed an increase in FRI scores (β = 2.94, 95% CI: −0.14∼6.01; p = 0.061), which was close to the level of significance.

After adjusting for potential confounders, the associations between frequent candy consumption and lower VCI scores remained statistically significant (β = −3.85, 95% CI: −6.28∼−1.43; p = 0.002). In addition, a new significant association was found between lower VCI scores and frequent consumption of sweet bakery products (β = −3.48, 95% CI: −6.47∼−0.49; p = 0.023).

In order to analyze the combined effects of frequent consumption of the seven ultraprocessed foods on the cognitive function of children, we calculated the number (0–7) of frequent consumption of ultraprocessed foods and divided the children into frequent consumption of 0–1 and 2–7 ultraprocessed foods. Table 4 shows the multiple linear regression analysis of the associations between the numbers of ultraprocessed foods types of frequent consumption and specific cognitive function domains. In the crude models, we did not find any significant associations. However, in the adjusted models, we found that children who frequently consumed more than two kinds of ultraprocessed foods were significantly associated with decreased VCI scores (β = −2.66; 95% CI: −5.12∼−0.19; p = 0.035). Furthermore, we also analyzed the association between the numbers of ultraprocessed food types of frequent consumption and cognitive function and found there were no significant associations no matter FSIQ scores or risk of cognitive deficit (Table 5).