The study site was Gongcheng Yao Autonomous County, Guilin, Guangxi, China. According to the cross-sectional sampling survey, the sample size calculation formula is as follows:

where N is the sample size, Z is the statistic, α = 0.05, Z = 1.96, p is the expected prevalence, q = 1-p, and d is the allowable error (d = 0.1p). The prevalence of cognitive dysfunction in rural Chinese older adults was 42.9% (25). The estimated sample size for this study was 589, with an additional 10–15% of the sample. The number of people surveyed in this cross section was N = 1,246. Each participant completed the MMSE and FFQ questionnaire and underwent a physical examination and laboratory tests.

Inclusion criteria were as follows: older adults who (a) resided in the study area; (b) were over 60 years of age; (c) had blood sample, height, weight, and questionnaire information; and (d) had a reasonable energy intake of 500–6000 kcal (26). Finally, the participants were propensity score matched (PSM) (1:3) by age and years of education. A total of 884 adults aged 60 years and older were included in this study (2018–2019). The study included men (n = 370) and women (n = 514).

This research was conducted in accordance with the guidelines set forth in the Declaration of Helsinki, and all procedures involving human participants were approved by the Ethics Committee of Guilin Medical College. Written informed consent was obtained from each participant (No. 20180702-3).

The MMSE questionnaire has high reliability and validity for the assessment of cognitive function. The questionnaire has the advantage of being easily applied and implemented in a short period. To assess the relationship between dietary patterns and the risk of cognitive dysfunction, MMSE scores have been classified according to the level of literacy: illiterate group ≤17, primary group ≤20, secondary and higher group ≤24 as the cutoff point, regarded as the cognitive impairment group (CI), compared to the non-CI group, and the rest are classified as the non-cognitive impairment (non-CI) group (27, 28).

The main food intake during the previous year was determined by reference to the food frequency questionnaire (FFQ), which included the frequency and quantity of food consumed (29). The intake of each group (g/day) was then calculated by adding the intake of the member's food. Daily nutrient intake was calculated by multiplying the frequency of consumption of each component of each food by the nutrient content of that particular component. In this study, foods were combined into different food groups according to their nutritional content and use. To extract dietary patterns, we divided 109 foods into 14 food groups with reference to similar nutrients: cereals and potatoes, vegetables, fruits, soybeans and nuts, poultry, gut, aquaculture, egg, milk, mushroom, alcohol, oil, salt, and oil tea. Dietary patterns were generated by utilizing the factor analysis method (FA) for 14 predefined food groups. Factor analysis is a nutritional epidemiological method. By distinguishing one or more dietary factors, we found that certain types of food tend (or not) to be ingested in the same sitting.

Feature values of >1 and scree plots were used to determine the number of dietary patterns. Furthermore, by using orthogonal transformations of rotation factors, the results were easier to interpret. Food groups with absolute factor loadings of >0.3 were significant in the dietary pattern because the food items in these food groups appeared to be strongly associated with the identified factors. Factor scores were calculated by weighting each participant's intake of the 14 food groups by factor loadings. We named dietary patterns based on food groups with factor loadings of >0.6.

Diabetes was defined as fasting blood glucose (FBG) of ≥7.0 mmol/L or a regimen of glucose-lowering medications (30). Hypertension was defined as systolic blood pressure (SBP) of ≥140 mmHg and/or diastolic blood pressure (DBP) of ≥90 mmHg or a regimen of antihypertensive medication (31) (for this study, hypertension includes having had hypertension and not knowing that you have hypertension but having hypertension on your test.). Dyslipidemia was defined as total cholesterol (TC) of ≥6.22 mmol/L, triacylglycerol (TG) of ≥ 2.26 mmol/L, low-density lipoprotein cholesterol (LDL-C) of ≥4.14 mmol/L as high LDL-hemorrhage, and high-density lipoprotein cholesterol (HDL-C) of <1.04 mmol/L as hypo-HDL-C (32). The following demographic information was collected through questionnaires: age (60–69, 70–79, and ≥80 years), sex (male or female), ethnicity (Han, Yao, or other), marital status (unmarried, married, or other), education (illiterate [uneducated], primary education [educated for 0–6 years], and higher education [educated for more than 6 years]), agricultural activities [involved in agriculture (plowing, planting, and weeding)], and BMI (body mass index). Participants who were retired, homemakers, unemployed for more than 1 year, or physically disabled were considered nonworking individuals, unlike those involved in physical activities in agriculture. Smoking was defined as currently smoking at least one cigarette per day. Alcohol consumption was defined as drinking at least 50 g of alcohol or more once a month (33).

The chi-square test and Student's t-test were applied for continuous and categorical variables, respectively, and the Wilcoxon rank-sum test was used for comparisons between groups of nonnormally distributed information. The energy and nutrient contents of food were obtained from the Chinese Food Composition Table (2009 version) (34). Nutrients were corrected using the residual method. Variance inflation factors (VIF < 10) (35) were employed to avoid multicollinearity between nutrients. The ranges of the included nutrients are as follows: SFAs, 229.324; MUFAs, 1159.459; PUFAs, 220.699; energy, 336.380; protein, 13.121; fat, 3052.548; carbohydrates, 81.488; and dietary fiber, 2.182. Case matching and controls by age and educational years were performed using the propensity score matching method. Four dietary patterns were extracted using the factor analysis (FA) method with a factor loading matrix criterion of 0.3. Food groups that exceeded the criterion of 0.6 showed extreme correlation with their corresponding dietary patterns and were then named according to those patterns. Unconditional regression of binary logistic regression was used to fit a continuous intake of 14 food groups to the cognitive function model, and linear trend analysis was implemented. Simultaneous restricted cubic spline (RCS) plots were used to represent the relationship between continuous dietary pattern scores and cognitive function. The lowest quartile (first quartile) was defined as the reference group in each model. The results were expressed as weighted OR (95% CI). After adjusting for potential confounders, unconditional binary logistic regression was applied to analyze the relationship between dietary patterns and cognitive function and to perform between-group trend analysis stratified by ethnicity. All analyses were performed in R (4.1.3), and two-tailed p-values of <0.05 were considered statistically significant. The authors can be contacted for the source code should there be any questions about the analysis.

The propensity score matching (PSM) results were as follows: non-CI (non-cognitive impairment) (N = 663) and CI (cognitive impairment group, compared to the non-CI group.) (N = 221). The jitter scatter plot and the histogram of the propensity value distribution show good results (Supplementary Figure 1).

Statistically significant differences occurred between the sex (P-values = 0.001), hypertension (P-values = 0.049), MMSE scores, orientation, memory, attention calculation, language skills, and visuospatial skills (P-values < 0.001) of the non-CI and CI groups, as shown in Table 1. Different ethnic groups had statistically significant variations in sociodemographic characteristics, including marital status (P-values = 0.009) and agricultural activities (P-values = 0.009), as shown in Table 2.

Significant differences were observed in the dietary pattern scores of SFAs (P-values = 0.040), protein (P-values = 0.010), carbohydrate (P-values = 0.027), dietary fiber (P-values = 0.025), fruit (P-values = 0.008), soybeans and nuts (P-values = 0.037), poultry (P-values = 0.030), aquaculture (P-values = 0.005), eggs (P-values < 0.001), mushrooms (P-values < 0.001), and vegetables and mushrooms (P-values = 0.001) of the non-CI and CI groups, as shown in Table 3. For the non-CI and CI groups, dietary pattern scores for PUFAs (P-values = 0.008), soybeans and nuts (P P-values = 0.002), oil tea (P P-values = 0.020), oil and salt (P-values = 0.042), and seafood and alcohol (P-values = 0.049) significantly varied among different ethnic groups, as shown in Table 4.

Kaiser–Meyer–Olkin (KMO) factor adequacy is equal to 0.81, Bartlett's test of sphericity is less than 0.001, and the cumulative variance explained is 44.64%, all of which meet the conditions for the use of factor analysis. The left side of the Scree plots shows the characteristic roots, and the slope for 4–5 factors becomes flatter, with the first four factors already covering most of the information, as shown in Supplementary Figure 2. Thus, four can be chosen as the number of factors. The dietary patterns were named according to the factor loadings of the 14 food groups in the dietary patterns (vegetables and mushrooms, oil, and salt, seafood and alcohol, and oil tea), as shown in Table 5. The first factor, which was defined by a high intake of vegetables, mushrooms, fruits, cereal potatoes, soybean nuts, gut, egg, and milk, was labeled “vegetables and mushroom dietary pattern.” The second factor, which was defined by a high intake of oil, salt, poultry, and soybean nuts, was labeled the “oil and salt dietary pattern.” The third factor, which was defined by a high intake of aquaculture, alcohol, poultry, and fruits, was labeled “seafood and alcohol dietary pattern.” The fourth factor, which was defined by a high intake of oil tea, milk, and gut, was labeled the “oil tea dietary pattern.”

There was a nonlinear relationship between consumption and cognitive function values in the cereal and potato, fruit, and poultry food groups, with P-values of p = 0.009, p = 0.01, and p = 0.024, respectively (Supplementary Figure 3-1). There was also a non-linear relationship between dietary intake and cognitive function in the aquaculture, egg, milk, and mushroom food groups, with P-values of p = 0.001, p < 0.001, p = 0.037, and p = 0.045, respectively (Supplementary Figure 3-2).

We used restricted cubic spline (RCS) to flexibly model and visualize the relationship between dietary pattern scores and cognitive function. The seafood and alcohol dietary pattern showed a non-linear relationship with cognitive function, with a trend consisting of an initial decline, a brief increase, and a final flattening off, P-values = 0.003 (Supplementary Figure 4).

In Model I, after adjusting for sex and ethnicity, the score of the vegetable and mushroom dietary pattern factor was in the fourth quartile (0.399, 7.056); the vegetable and mushroom dietary pattern may be a protective factor for cognitive function, with an OR (95% CI) of 0.484 (0.302, 0.769). Other dietary patterns were not associated with cognitive function. In Model II, after adjusting for sex, ethnicity, marital status, agricultural activities, smoking, alcohol consumption, hypertension, diabetes, dyslipidemia, and BMI, when the score of the vegetable and mushroom dietary pattern factor was in the fourth quartile (0.399 and 7.056), it was found that the vegetable and mushroom dietary pattern may be a protective factor of cognitive function, with an OR (95% CI) of 0.517 (0.320, 0.830). Other dietary patterns were not associated with cognitive function. In Model III, after adjusting for sex, ethnicity, marital status, agricultural activities, smoking, alcohol consumption, hypertension, diabetes, dyslipidemia, BMI, and dietary fiber, when the score of the vegetable and mushroom dietary pattern factor was in the fourth quartile (0.399, 7.056), it was found that the vegetable and mushroom dietary pattern may be a protective factor of cognitive function, with an OR (95% CI) of 0.578 (0.348, 0.951). Other dietary patterns were not associated with cognitive function. A linear trend was noted between the quartiles of the dietary scores of the vegetable and mushroom (P-values = 0.017) and oil tea (P-values = 0.042) dietary patterns, as shown in Table 6.

This study also has some limitations that must be carefully considered. First, this study is a cross-sectional study, consisting mainly of older Chinese adults, and therefore, it is not possible to establish a clear chronological order or to exclude the possibility of reverse causality and the existence of Neyman bias. Second, the dietary intake assessment was not performed before the cognitive function measures, and we cannot infer the existence of the resulting causality. Third, the dependence of the FFQ method on memory or cognitive performance in this study did not allow for a precise estimation of nutrient intake. Fourth, we controlled for the effects of nutrient dietary intake in the model using the residual method and VIF. Fifth, we were unable to adjust for genetically related lipoprotein E status, but we controlled for a wide range of health-related factors as potential confounders. Sixth, because we conducted a large cross-sectional epidemiological survey, the MMSE questionnaire is not clinically adequate to accurately classify cognitive impairment, and our data do not adequately support studies with a detailed classification of patients with cognitive impairment. The MMSE questionnaire can only be used to initially assess the status of cognitive function. Therefore, to obtain more precise findings, longer follow-up and larger prospective studies are necessary to explore the relationship between dietary patterns and cognitive function.