Exploration of dietary patterns in southern China: application-based comparison of factor analysis and latent class analysis

Abstract

Background

In recent years, compared with traditional dietary analysis (simply focused on individual nutrients or foods), the analysis of dietary patterns has emerged as a comprehensive approach. This study aims to explore the dietary patterns of residents in a certain region of southern China through factor and latent class analysis (LCA), and compare the advantages and disadvantages of the two methods, providing data support for future research.

Methods

We conducted a cross-sectional study using random stratified cluster sampling in the Gaozhou County, Maoming, Guangdong Province, China. Overall, 12,212 participants were recruited for the study, and data were collected using a general questionnaire consisting of two parts focusing on sociodemographic characteristics and residents’ dietary behaviors. Factor and latent class analysis (LCA) were then performed to identify patterns of dietary behaviors, and logistic regression was used to explore the associations between sociodemographic characteristics and dietary behavior classes.

Results

Both factor analysis and LCA were useful when assessing the classification of residents’ dietary patterns. However, unlike prior models, the LCA identified emergent dietary behavior, highlighting previously unrecognized variations. Five latent classes (the balanced diet: 10.75%, tending-to-be-balanced diet: 8.03%, meat-loving diet: 22.19%, traditional diet: 45.00%, and unbalanced diet: 14.03%) were identified. The results showed that sex, age, marital status, education level, monthly income, and chronic disease status (all P < 0.05) were the main factors influencing dietary patterns.

Conclusions

This study reveals previously uncharacterized dietary patterns in Southern China, offering novel insights for future research in this field.

Introduction

According to Maslow’s Hierarchy of Needs Theory, the lowest or most basic individual needs are physiological, one of which is diet [1]. Similarly, an old Chinese proverb states that food is a paramount necessity [2]. Over the past half-century, with the rapid economic development and urbanization of the population, societies have entered different stages of what has been called the nutrition transition [3]. Particularly in the past decade, this rapid dietary transition—from traditional diets to energy-dense, nutrient-poor foods—has repositioned NCDs as the dominant nutrition-related health burden in developing nations, surpassing malnutrition in disease attribution [4]. The Centers for Disease Control and Prevention recently reports that over one-third of U.S. adults (39.8%) and approximately 18.5% of youths aged 2–19 years are considered obese [5]. This phenomenon also exists in China, where it is even more serious. According to data from the “Report on the Nutrition and Chronic Diseases of Chinese Residents (2020),” more than half of the Chinese adult population was classified as overweight or obese, and the proportion among youths aged 6–17 years was close to 20.0%. This is largely related to uncontrolled dietary behaviors such as emotional and binge eating, which are significant factors leading to excess weight [6, 7]. Importantly, obesity is not only a chronic disease but also an antecedent or risk factor for most diseases, such as cardiovascular disease and metabolic disorders [8]. Therefore, paying attention to dietary behavior is important for reducing the occurrence of preventable diet-related chronic diseases.

Existing research on dietary behavior has mostly focused on the most common types of eating behaviors, including uncontrolled diet (UE), emotional diet (EE), and cognitive restraint (CR) [9–11], and the relationship between dietary behavior and body mass index (BMI) [12, 13]. Some studies have also summarized the factors influencing dietary behaviors, including sex, age, and social, economic, and lifestyle factors [14–17]. Additionally, research into dietary patterns has recently gained momentum. Unlike dietary behavior, dietary patterns focus more on the entire daily diet consumed by individuals and populations for months or years [8]. Therefore, the examination of dietary patterns more accurately reflects real-world dietary behaviors.

In 2015, the Dietary Guidelines Advisory Committee defined dietary patterns as the quantities, proportions, variety, or combinations of different foods, drinks, and nutrients (when available) in diets and the frequency with which they are consumed [18]. Notably, prevailing dietary guidelines emphasize the roles of dietary patterns in preventing cardiovascular disease [19]. For example, fruits and vegetables are often identified as the most important parts of the diet for preventing age-related diseases [20, 21]. Similarly, previous studies have found that the intake of vegetables, fruits, fiber, folate, and whole grains (as food patterns) may be independently associated with a reduced risk of coronary heart disease (CHD) [22]. Thus, a shift toward healthy dietary patterns has the potential to curtail the current unsustainably high levels of obesity, cardiovascular disease, diabetes mellitus, and cancer. However, dietary patterns of individuals or groups are often not directly measurable and must be determined using the appropriate statistical methods.

Previous studies have conventionally analyzed dietary patterns through three methodological approaches: factor analysis (for dimension reduction), cluster analysis (for population segmentation), and dietary indices (for predefined quality assessment) [23–27]. Factor analysis is one of the most commonly used methods for determining dietary patterns. For example, using factor analysis, Hu et al.. summarized two dietary patterns of men aged 40–75: the “prudent pattern” and the “western pattern” [28]. Additionally, using data from the China Kadoorie Biobank baseline survey, Yu et al.. analyzed the dietary behavior of the Chinese population (aged 30–79 years) and found three dietary patterns: the traditional southern dietary pattern, the traditional northern dietary pattern, and the Western/new affluence dietary pattern [29]. However, owing to changes in food preferences and availability, the types of dietary patterns can change over time. Furthermore, dietary patterns of residents in different countries and regions show significant differences. Therefore, existing dietary patterns applicable to Chinese residents require further investigation and discussion.

Besides the methods mentioned above, some scholars have indicated that latent class analysis (LCA) is suitable for discrete and binary classification and is a more efficient way to explore healthy behaviors [30, 31]. LCA is a data-driven analysis method that divides heterogeneous populations into homogeneous groups based on categorical indicator variables [32]. This method can address the complexity of dietary behavior patterning and capture meaningful key behavior patterns in the underlying population. At present, LCA has been widely applied in many research fields—such as sociology, biomedicine, and psychology—and some scholars have used it to explore residents’ dietary patterns [33–35]. However, different statistical methods have their advantages and disadvantages when investigating dietary patterns. Some experts have suggested that multiple statistical methods could be used to mutually verify and supplement dietary patterns to better understand potential patterns. For example, the variable-centered factor analysis primarily investigates population-level characteristics of the study subjects, whereas the person-centered latent class analysis first accounts for potential heterogeneity among participants by classifying them into distinct subgroups before examining their shared features. However, domestic research in this field remains limited.

Therefore, this study aims to explore dietary behavior patterns through both factor analysis and latent class analysis, compare the similarities and differences between these two methodological approaches, identify more appropriate dietary pattern classifications, and analyze the associations between these patterns and sociodemographic characteristics.

Methods

Study design and participants

This questionnaire-based cross-sectional study was conducted using stratified cluster random sampling in Gaozhou County, Maoming, Guangdong Province from July to September 2021. Generally speaking, the sample size is calculated using the following formula N=, and the sample size is determined to be 385 (Z=1.96, P=0.5, e=0.05). Under the ‘Healthy Gaozhou 2030’ initiative, this citywide health survey adopts a full-coverage sampling strategy (385*28=10780) across all 28 administrative units (5 subdistricts, 23 towns). An additional 15% was added to this sample estimate in anticipation that the final sample would include individuals who would not consent to participate in the survey. Thus, the final sample size was estimated to be 12,397 (10780*115%) at least. To build our sample, 10% of the villager committees (resident committees) from each town government (street office) in Gaozhou were randomly selected using the equal probability method, after which, a cluster sampling method was used to extract respondents from these committees, with the same probability of each respondent being selected. Owing to the impact of the COVID-19 pandemic, the survey was conducted through an online survey platform—Survey-Star (Changsha Ranxing Science and Technology, Shanghai, China) the most popular survey software in China, by investigators who issued one-on-one questionnaires to residents. This survey was conducted with support from the Gaozhou Municipal Health Commission, where staff from community health service centers served as volunteers to assist participants, especially providing face-to-face assistance to elderly people (≥ 55 years old) who are not familiar with smartphones. The questionnaire was anonymous and can only be submitted after all questions are completed to ensure the confidentiality and reliability of data.

The inclusion criteria were as follows: (1) aged 18–65 years, (2) residence record in Maoming that can be detected according to official records, (3) have not moved away from the residing address, (4) have clear awareness and barrier-free communication skills, and (5) willing to cooperate and able to complete the survey. Those who had difficulty communicating or were unwilling to cooperate were excluded. Meanwhile, due to the unique dietary habits of pregnant women, they were not included in the survey. A total of 12,750 questionnaires were collected in this survey. After data screening, the final number of valid questionnaires was 12,212, with a valid return rate of 95.78%. Ethical approval for this study was obtained from Southern Medical University.

Measures

The questionnaire consisted of two parts focusing on sociodemographic characteristics and the current status of the residents’ dietary behaviors. The first part included questions regarding sex, age, marital status, educational level, monthly income, BMI, and chronic diseases. Educational level was categorized as “primary school or below,” “secondary school,” “high school,” or “post-secondary school or above.” Moreover, monthly income was categorized as “no income,” “less than 1,000 RMB,” “1,000–3,000 RMB,” “3,000–5,000 RMB,” and “more than 5,000 RMB.” To calculate the BMI, the height and weight of each participant were collected, and the following internationally accepted formula was used: BMI = weight (kg)/height (m)². For a clear representation of BMI, BMI status was classified as “underweight” (BMI < 18.5 kg/m²), “normal weight” (18.5 kg/m² ≤ BMI < 24.0 kg/m²), “overweight” (24.0 kg/m² ≤ BMI < 28.0 kg/m²), and “obesity” (BMI ≥ 28.0 kg/m²) according to the BMI classification criteria set by the Working Group of Obesity in China (WGOC) [36].

The second part of the questionnaire included 10 items related to dietary behaviors. The participants were asked whether they ate vegetables and fruits, cereals and tubers, beans and eggs, milk or other dairy products, and livestock, poultry, and fish every day. The possible answers were “none,” “1–3 times per week,” and “4–7 times per week”. The Cronbach’s alpha of the scale was 0.833.

Statistical analysis

The data analysis comprised three parts. First, factor analysis and LCA were used to identify the dietary patterns, and the differences between them were compared. Subsequently, the chi-square test was performed to compare dietary patterns according to different sociodemographic variables. Finally, multinomial logistic regression analysis and the adjusted odds ratios (OR) and 95% confidence intervals (CIs) were calculated to examine the potential factors influencing dietary patterns. Prior to this, Harman’s single-factor test was conducted to detect common method bias, which has been widely used in common method bias testing. After principal component analysis, seven eigenvalues greater than 1 were extracted. The first factor explained 27.1% variance, which is far below the critical value of 40%. Accordingly, we concluded that common method bias is not a serious problem in the current study. All analyses were performed using SPSS 23.0 and Mplus 8.3 statistical software, and P values < 0.05 were considered significant.

Before factor analysis, the Kaiser–Meyer–Olkin (KMO) test and Bartlett’s test of sphericity were used to determine whether the data were applicable. After verification, dietary patterns from the 10 dietary behaviors were constructed using factor analysis. First, principal component analysis was used to identify the major common food factors based on the above dietary behaviors. Subsequently, an orthogonal (varimax) rotation was performed to achieve a structure with independent factors and greater interpretability [37]. The number of factors was determined according to the eigenvalue (> 1), scree plot, factor interpretability, and variance explained by each factor. Factor loadings represent the correlation between identified components and each variable, used to indicate the resulting dietary patterns [38]. The greater the factor load, the stronger the correlation between the food group and the corresponding diet pattern.

Indicators chosen for the latent class models included the 10 dietary behaviors previously described. To select the appropriate number of classes, a two-class model was first fitted to the data, and the number of classes was successively increased by one, fitting a new LCA model to the data at each step until we identified the simplest model that provided an adequate fit. The following six model fit indices were adopted to evaluate the optimal model: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted Bayesian Information Criterion (aBIC), Lo–Mendell–Rubin (LMR), Bootstrapped Likelihood Ratio Test (BLRT), and Entropy [39]. Among these model fit indices, AIC and BIC are the most widely used indices for LCA selection [40]. Furthermore, Lin and Dayton indicated that the BIC is more reliable when the sample size exceeds 1,000; otherwise, the AIC is better [41]. We used LMR and BLRT to compare the estimated model and a model with k-1 class(es), with k equal to the number of classes [42]. If the significance of the LMR or BLRT was < 0.05, the k-class solution was significantly better than the k-1 class solution. Furthermore, we used entropy as an important criterion to choose the best model, which was measured in the range of 0–1, with higher values preferred.

Results

Sociodemographic characteristics of the participants

Overall, 12,212 participants were recruited for the study, of which 7,062 were male (57.83%) and 5,150 were female (42.17%). The mean age of the sample was 39.76 ± 11.75 years, with 2,964 participants (24.27%) aged 18–29 years, 4,628 (37.90%) aged 30–44 years, 4,124 (33.77%) aged 45–59 years, and 496 (4.06%) aged 60–65 years. Notably, most participants were married (78.67%), and 5,806 participants (47.55%) had a college degree or above. Moreover, 60.53% of the residents had a normal weight, 27.30% were overweight, and 6.23% were obese.

Regarding dietary patterns, the majority of participants had a high intake (4–7 times per week) of vegetables and cereals (90.76% and 80.36%, respectively), while less than half (44.65%) had a high intake of fruits. Most residents had a limited intake (1–3 times per week) of beans (72.60%), eggs (67.24%), and fish (74.74%). In addition, 29.66% of participants never ate milk or its products. The specific sociodemographic information and statistical descriptions of the participants are shown in Table 1.

Table 1.

Sociodemographic and dietary behavior characteristic

Variable	Category	N/%
Sex	Male	7062 (57.83%)
	Female	5150 (42.17%)
Age (years)	18–29	2964 (24.27%)
	30–44	4628 (37.9%)
	45–59	4124 (33.77%)
	60–65	496 (4.06%)
Marital status	Married	9607 (78.67%)
	Unmarried	2605 (21.33%)
Education level	Primary school and below	738 (6.04%)
	Secondary school	2935 (24.03%)
	High school	2733 (22.38%)
	Post-secondary school and above	5806 (47.55%)
Monthly income (RMB)	No income	779 (6.38%)
	< 1,000	845 (6.92%)
	1,000–3,000	3511 (28.75%)
	3,000–5,000	4314 (35.3%)
	> 5,000	2763 (22.63%)
BMI	Underweight	725 (5.94%)
	Normal weight	7392 (60.53%)
	Overweight	3334 (27.3%)
	Obese	761 (6.23%)
Chronic disease	Yes	3534 (28.94%)
	No	8678 (71.06%)
Vegetable intake frequency	None	82 (0.76%)
	1–3 times per week	1046 (8.57)
	4–7 times per week	11,084 (90.67%)
Fruit intake frequency	None	707 (5.79%)
	1–3 times per week	6052 (49.56%)
	4–7 times per week	5453 (44.65%)
Cereal intake frequency	None	302 (2.47%)
	1–3 times per week	2096 (17.16%)
	4–7 times per week	9814 (80.36%)
Tuber intake frequency	None	1728 (14.15%)
	1–3 times per week	8266 (67.69%)
	4–7 times per week	2218 (18.16%)
Bean intake frequency	None	1258 (10.30%)
	1–3 times per week	8866 (72.60%)
	4–7 times per week	2088 (17.10)
Egg intake frequency	None	414 (3.39%)
	1–3 times per week	8211 (67.24%)
	4–7 times per week	3587 (29.37%)
Milk intake frequency	None	3622 (29.66%)
	1–3 times per week	6251 (51.19%)
	4–7 times per week	2339 (19.15%)
Livestock intake frequency	None	185 (1.51%)
	1–3 times per week	4938 (40.44%)
	4–7 times per week	7089 (58.05%)
Poultry intake frequency	None	198 (1.62)
	1–3 times per week	8076 (66.13%)
	4–7 times per week	3938 (32.25%)
Fish intake frequency	None	677 (5.54%)
	1–3 times per week	9127 (74.74%)
	4–7 times per week	2408 (19.72)

Factor analysis

The statistical results showed that the KMO statistical value was 0.849, and Bartlett’s test result was P < 0.001, indicating a strong correlation between the original variables, which was suitable for factor analysis. Three factors can be extracted based on eigenvalues greater than 1 (Fig. 2). The eigenvalues are 3.786, 1.308, and 1.014, respectively. The pattern loadings showed that the first factor, termed the “modern health model,” had a high loading on various health foods such as tubers, beans, milk, fruits, eggs, and fish. The second factor had high loadings for livestock and poultry and was named the “meat model.” Furthermore, the third factor, named the “staple vegetable model,” was characterized by a high loading of vegetables and cereals. Overall, the variance contribution rate of the three factors was 61.08%.

Fig. 2.

Scree plot of the factor analysis

Fig. 1.

Factor-loading coefficients of the three major factors after varimax rotation

Latent class analysis

Model fit and selection

Table 2 presents the LCA fit statistics for the one-to-seven-class models. The AIC, BIC, and aBIC were calculated for each potential category model, and based on the statistical values, an optimal model with five potential categories was selected. Compared to the other classes (one to four), the five-class model showed the smallest AIC, BIC, and aBIC. Furthermore, entropy was reported to only demonstrate the precision with which the cases were classified in the profiles (on a scale of 0–1). Although the BIC value was the smallest when the number of categories was six, the entropy was the most ideal when the number of categories was five. Therefore, the five-class model was selected as the best-fit model.

Table 2.

Model fit information for the latent class models (Classes 1–7)

Model	AIC	BIC	aBIC	Entropy	LMR	BLRT
Class 1	178428.866	178577.070	178513.512
Class 2	157235.010	157538.827	157408.534	0.903	< 0.001	< 0.001
Class 3	150709.315	151168.746	150971.717	0.849	< 0.001	< 0.001
Class 4	147041.822	147656.866	147393.101	0.787	< 0.001	< 0.001
Class 5	145752.532	146523.190	146192.689	0.786	< 0.001	< 0.001
Class 6	144530.410.	145456.682	145059.445	0.775	< 0.001	< 0.001
Class 7	143419.166	144501.052	144037.080	0.779	0.0202	< 0.001

AIC: Akaike information criterion; BIC: Bayesian information criterion; aBIC: adjusted Bayesian information criterion; LMR, Lo–Mendell–Rubin; BLRT, Bootstrapped Likelihood Ratio Test. The latent class model chosen is highlighted in bold

Class description

The five classes identified in this study, defined by their clustered dietary behaviors, were named “balanced diet,” “tend to be balanced,” “meat-loving,” “traditional diet,” and “unbalanced diet.” Each latent class corresponds to an underlying subgroup of individuals characterized by a particular behavioral pattern. For example, in Class 1 (balanced diet), the participants had relatively balanced eating habits, including a high-frequency (4–7 times per week) weekly intake of all foods listed (vegetables, fruits, cereals, eggs, etc.), accounting for 10.75% of the sample. Class 2 (tend to be balanced) was characterized by a moderately high probability of eating vegetables, fruits, and cereals 4–7 times per week, as well as milk, poultry, and fish 1–3 times per week. This group had a variety of diets—with the intake of various foods differing slightly—and represented 8.03% of the total sample. In contrast, Class 3 (meat-loving) comprised 22.19% of the sample and had limited food types and a high-frequency intake of only some foods. This class had a high probability of eating livestock and poultry food 4–7 times per week, with an insufficient weekly intake of other foods such as tubers, beans, and milk. Class 4 (traditional diet) was the most prevalent among the five dietary patterns, accounting for 45.00% of the sample. The dietary behaviors of individuals in this subgroup were similar to the traditional dietary habits in China [43, 44], mainly manifested in high-frequency (4–7 times per week) intake of vegetables and cereals, with a higher probability of eating other foods 1–3 times per week. Finally, Class 5 (unbalanced diet) represented 14.03% of the sample and was characterized by a high probability of almost never eating tubers, beans, and milk, as well as a high probability of eating fruits, eggs, poultry, and fish 1–3 times per week. The item-response probabilities for each class are shown in Fig. 3.

Fig. 3.

Graphical display of the item-response probabilities for each class

Comparison of factor analysis and latent class analysis

This study found that both factor analysis and LCA can classify residents’ dietary patterns through “dimension reduction,” although the results of the two methods differed. However, the three categories in the factor analysis corresponded to some categories in the latent class analysis (Fig. 4). For example, the “meat model” was dominated by livestock and poultry in both classifications. Moreover, the “staple vegetable model” in the factor analysis corresponded to an “unbalanced diet” in the LCA. The dietary characteristics of this group were mainly cereals and vegetables, whereas the intake of other foods was limited. Additionally, the “modern health model” in the factor analysis was dominated by tubers, beans, milk, fruits, eggs, and fish, similar to the characteristics of a “balanced diet.” This result shows that when the sample size is large, LCA can classify the population in more detail than factor analysis. Therefore, we selected the results of the classification of five dietary patterns according to the LCA.

Fig. 4.

Comparison between factor analysis and latent class analysis

Differences between sociodemographic characteristics and latent class membership

The differences in the distribution of sociodemographic characteristics such as sex, age, education level, monthly income, and chronic disease across different potential classes were significant (P < 0.05). For example, participants in the “balanced diet” classes were mainly male, aged 60–65 years, and married, with a high school education, normal BMI, and no chronic diseases. In contrast, a significantly higher proportion of residents in the “meat-loving” group were women aged 18–29, obese, unmarried, with a college degree or above, and a monthly income of more than 5,000. Furthermore, the “unbalanced diet” group was mainly male, aged 60–65 years, unmarried, suffering from chronic disease, without income, and had an education level of primary school and below. Table 3 presents the differences in dietary patterns according to the different sociodemographic characteristics.

Table 3.

Differences in dietary patterns according to the different sociodemographic characteristics

Variables	Dietary Patterns (N = 12,212, %)					χ2
	Class 1	Class 2	Class 3	Class 4	Class 5
Sex						47.839**
Male	11.60	7.41	20.73	45.23	15.04
Female	9.59	8.89	24.19	44.68	12.64
Age (years)						225.631**
18–29	12.21	8.44	27.46	39.00	12.89
30–44	10.28	7.61	24.59	45.31	12.21
45–59	9.72	8.10	17.29	48.42	16.47
60–65	14.92	9.07	9.07	49.40	17.54
Marital status						69.235**
Married	10.78	8.08	20.92	46.68	13.53
Unmarried	10.63	7.87	26.87	38.77	15.85
Education level						490.523**
Primary school and below	9.62	9.76	8.27	49.73	22.63
Secondary school	11.55	8.18	13.26	50.02	16.80
High school	14.09	8.85	19.21	44.71	13.14
Post-secondary school and above	8.92	7.35	29.78	41.99	11.95
Monthly income (RMB)						213.182**
No income	9.37	9.50	17.71	41.59	21.82
< 1,000	15.98	8.76	11.36	44.38	19.53
1,000–3,000	11.73	8.43	20.28	45.40	14.16
3,000–5,000	10.31	7.90	20.28	45.40	14.15
> 5,000	8.98	7.90	22.88	46.08	12.82
BMI						24.645*
Underweight	8.83	7.59	24.00	44.14	15.45
Normal weight	11.38	8.39	21.43	45.05	13.76
Overweight	10.11	7.77	22.71	45.53	13.89
Obese	9.33	6.18	25.62	42.97	15.90
Chronic disease						256.867**
Yes	4.73	5.86	24.65	47.40	17.37
No	13.21	8.92	21.19	44.02	12.66

BMI: body mass index; *P < 0.05, **P < 0.001

Analysis of the influencing factors of different dietary patterns

Multinomial logistic regression was used to explore the relationship between sociodemographic factors and potential classes. Five dietary patterns were used as dependent variables, and the “balanced diet” class was used as the reference group. All factors were used as independent variables for inclusion in the logistic regression analysis. The results showed that sex, age, marital status, educational level, monthly income, and chronic disease status were the main factors influencing dietary patterns. Notably, all variables had different effects on dietary patterns, of which, sex, monthly income, and chronic disease status were the most common influencing factors (Table 4).

Table 4.

Analysis of the factors influencing dietary patterns

Variables	Class 2 (Ref = Class 1)	Class 3 (Ref = Class 1)	Class 4 (Ref = Class1)	Class 5 (Ref = Class1)
	OR (95% CI)	OR (95% CI)	OR (95% CI)	OR (95% CI)
Sex
Male	1	1	1	-
Female	1.469(1.236–1.745)***	1.456(1.265–1.674)***	1.248(1.098–1.419)**	-
Age (years)
60–65		1		1
18–29		1.857(1.201–2.869)**		0.986(0.659–1.474)
30–44		2.057(1.364–3.103)**		1.395(0.967–2.010)
45–59		1.698(1.133–2.544)*		1.601(1.127–2.274)**
Marital status
Unmarried		1		1
Married		0.807(0.656–0.992)*		0.554(0.440–0.696)***
Education
Post-secondary school and above		1	1	1
Primary school and below		0.482(0.324–0.716)***	1.50(1.097–2.050)*	2.183(1.53–63.103)***
Secondary school		0.534(0.438–0.652)***	1.164(0.978–1.386)	1.38(1.121–1.699)**
High school		0.514(0.432–0.610)***	0.777(0.667–0.909)**	0.805(0.663–0.977)*
Monthly income (RMB)
No income	1	1	1	1
< 1,000	0.513(0.330–0.798)**	0.533(0.358–0.795)**	0.552(0.396–0.770)***	0.469(0.324–0.681)***
1,000–3,000	0.749(0.520–1.078)	0.971(0.707–1.334)	0.865(0.651–1.149)	0.552(0.403–0.755)***
3,000–5,000	0.819(0.568–1.180)	1.081(0.787–1.484)	1.030(0.775–1.369)	0.601(0.438–0.824)**
> 5,000	0.803(0.540–1.193)	1.266(0.904–1.772)	0.990(0.730–1.344)	0.542(0.384–0.764)***
Chronic disease
No	1	1	1	1
Yes	1.796(1.426–2.261)***	2.989(2.477–3.608)***	2.908(2.436–3.473)***	3.938(3.234–4.796)***

OR: adjusted odds ratio; CI: confidence interval

*P < 0.05, **P < 0.01, ***P < 0.001

Discussion

Healthy eating is related to multiple social health issues, such as obesity, diabetes, and other medical conditions [45]. In 2022, the Chinese Nutrition Society released the dietary guidelines for Chinese residents, summarizing reasonable dietary guidelines to guide residents in cultivating healthy eating habits. This guide also proposes the “Oriental healthy dietary pattern”, representing the dietary pattern in the southeast coastal area of China. The two fundamental principles of this dietary pattern are reduced salt intake and dietary diversity. Specifically, (1) strict sodium restriction (< 5 g/day), (2) consumption of over 12 different food types weekly, and (3) predominant use of plant-based ingredients while maintaining balanced animal protein intake. Unlike the Mediterranean diet [46, 47], the Dietary Approaches to Stop Hypertension diet [48, 49], and the traditional Japanese dietary pattern [50], the Oriental health dietary pattern aligns more closely with Chinese culinary traditions—prioritizing grains, diverse plant-based foods, and moderate animal protein, while maintaining low salt and oil usage. However, further investigation regarding other eating patterns among Chinese residents is required.

Dietary patterns

Previous studies have assessed dietary patterns among children and adults in different regions of China; however, the results have been inconsistent owing to different subjects and survey times. For example, Xu et al.. (2015) analyzed dietary patterns of older adults using data from the 2009 China Health and Nutrition Survey and found that they had two different dietary patterns: a traditional dietary pattern and a modern dietary pattern [51]. In this study, we identified the following five dietary patterns: the balanced diet, tending-to-be-balanced diet, meat-loving diet, traditional diet, and unbalanced diet. These patterns account for different dietary behaviors, with the classification reflecting, to a certain extent, the transformation of the dietary structure with the continuous development of the economy. With an increase in income and health awareness, residents’ dietary patterns have gradually changed from an unbalanced diet to a balanced diet. Notably, the most common dietary pattern among the participants in this study was the traditional diet (high intake of vegetables, cereals, and livestock), which was similar to the results of previous research [52, 53]. The second-most common pattern was the meat-loving pattern, in which the residents’ diet mainly consisted of high-fat food such as animal foods. This corresponds with the results of Du et al.. (2004), who indicated that the structure of the Chinese diet is shifting away from high-carbohydrate foods and toward high-fat, high-energy-density foods [54]. This was confirmed by the dietary patterns in our study.

We also found that the proportion of residents with a relatively balanced diet was 18.78%. The Chinese Nutrition Society recommends that the daily diet include cereals, vegetables, fruits, livestock, poultry, fish, eggs, and beans, maintaining food diversity. In this study, residents in the balanced diet and tend-to-be-balanced diet classes exhibited diversity in diet types. However, residents in the unbalanced diet class can only maintain a high-frequency intake of vegetables and cereals while rarely consuming beans and milk. Further analysis revealed that this group mainly comprised older adults with a low socioeconomic status.

Correlation between dietary patterns and sociodemographic characteristics

Besides the nutritional status, this study showed a correlation between the demographic characteristics of the participants and their identified latent classes. The results of this study indicated that sex, age, marital status, education level, and monthly income had complex effects on different dietary patterns. Using the balanced diet group as a reference, we found that women (OR = 1.469, 95% confidence interval [CI]: 1.236–1.745) and patients with chronic diseases (OR = 1.796, 95% CI: 1.426–2.261) were more likely to maintain a dietary pattern that tends to be balanced. In contrast, sex, age, marital status, education level, monthly income, and chronic diseases were the main influencing factors for the meat-loving group. Previously, some studies found that females usually have higher body weight concerns and are more prone to limit food consumption to control their weight [55]; however, this was not observed in the present study. In fact, in this study, women were 1.456 times more likely to maintain a meat-loving diet than men. Further analysis found that more than half of the women with this dietary pattern had a normal BMI. Therefore, although these women followed a meat-loving diet, they tended to pay attention to maintaining a normal weight. This may be related to the recent popularity of the National Fitness Plan in recent years [56].

Notably, compared with the participants aged 60–65 years, residents in other age groups were more likely to follow a meat-loving diet. Red meat such as pork, beef, and mutton has been found to be an energy-dense food that contains a large amount of fat, and the long-term consumption of red meat may be a risk factor for hyperlipidemia, hypertension, and diabetes [57–59]. Therefore, older adults may choose to control their meat intake for their own health. Notably, we found that being married (OR = 0.807, 95% CI: 0.656–0.992) seems to be a protective factor for a meat-loving diet. In contrast, a lower education level and monthly income were associated with a lower likelihood of choosing a meat-loving diet. This is largely related to the residents’ socioeconomic status (SES) because income is an important factor in dietary choices. Drewnowski and Popkin indicated that diets high in fat, especially meat and milk products, are tied to a high level of income, whether at the national or individual level [3]. Therefore, residents with a low SES have limited opportunities to choose meat-based foods. Contrary to the results of previous studies, we found that residents with chronic diseases were more likely to choose a meat-loving diet. However, this was a reasonable phenomenon because previous studies have also shown that a high-fat diet is significantly associated with the occurrence of chronic diseases. Similarly, for patients with chronic diseases who maintain a tending-to-be-balanced diet, healthy dietary behaviors are largely due to gradual formation after illness to alleviate chronic diseases.

This study also found that residents aged 45–59 years who were unmarried and had an education level of primary school or below, no income, and chronic diseases had a higher risk of an unbalanced diet. In general, this group had a low SES and carried the heavy economic burden of disease. Additionally, due to limited education, this group cannot realize the importance of a healthy diet or how to maintain a balanced diet.

Although dietary patterns of residents with different BMIs were significantly different upon univariate analysis, the relationship between BMI and dietary patterns was not significant in the regression analysis. This finding is consistent with the results of previous studies. For example, Wang et al.. found that both the fruit-egg and nut-wine dietary patterns were not significantly correlated with overweight or obesity [60]. This phenomenon may occur because variables showing significance in univariate analysis actually contribute minimally to the outcome in practice, with their effects being overshadowed by more influential variables in the regression model. Therefore, the relationship between BMI and dietary patterns requires longitudinal data to further determine a causal relationship.

Strengths and limitations

This study combined factor analysis and LCA to identify the dietary patterns of Chinese residents. Factor analysis is widely used to reduce data and extract a small number of factors depending on the correlation matrix, whereas LCA is performed to identify subtypes of related behaviors (latent classes) from multivariate categorical data. Nonetheless, similarities were observed across the methods, suggesting some of the basic qualities of a healthy diet. Ultimately, we identified five dietary patterns. The findings also showed that unmarried participants, women, and residents with a higher SES seemed to prefer a meat-loving diet, which deserves attention.

However, this study has some limitations. First, the intake of both healthy and unhealthy food should be focused on when analyzing dietary behavior and patterns. In this regard, this study only included 10 major food categories and lacked information on unhealthy dietary behaviors such as fried food, pickled food, snacks, and beverages. Therefore, future research should investigate this topic with more focus on both healthy and unhealthy food to evaluate dietary patterns more comprehensively. Second, due to the use of questionnaire survey methods, participants may provide responses that the surveyor considers “expected,” which is difficult to avoid. Third, given the cross-sectional design, causal inferences could not be made in this study, especially regarding the relationship between BMI and dietary patterns. Therefore, future longitudinal studies are required to elucidate the mechanisms underlying these associations. Nevertheless, these limitations do not affect the significance of this study.

In conclusion, this study identified dietary patterns through both factor analysis and LCA, with LCA demonstrating superior ability to capture population-level dietary heterogeneity. Five dietary patterns have been identified: balanced diet, tending-to-be-balanced diet, meat-loving diet, traditional diet, and unbalanced diet. It is noteworthy that the majority of residents (45.0%) adhered to traditional dietary patterns, with only 18.78% exhibiting balanced or tend-to-be-balanced dietary behaviors, indicating persistent gaps in optimal dietary diversity. The low prevalence of balanced diets (18.78%) underscores urgent needs for: targeted nutrition education programs; and policy interventions to promote dietary diversity. Moreover, the results showed that sex, age, marital status, education level, monthly income, and chronic disease status were the main factors influencing dietary patterns. This suggests that government agencies, such as community health service centers and disease control/prevention centers, could utilize National Nutrition Week (annually held during the third week of May) to enhance dissemination of evidence-based nutrition knowledge. Key messaging should promote healthy dietary principles such as reducing oil and salt intake while increasing consumption of legumes and dairy products, with particular focus on socioeconomically vulnerable populations.

Author contributions

Conceptualization, C.Z.; methodology, K.T. and Y.X.; validation, K.T.; formal analysis, Y.X., H.C. and Z.T.; writing—original draft preparation, Y.X., H.C. and Y.Z.; writing—review and editing, Y.Z. and Z.Q.; supervision, Z.Q. and C.Z.; funding acquisition, C.Z. and Z.T. All authors have read and agreed to the published version of the manuscript.

Funding

This study was funded by the National Nature Science Foundation of China [Grant number: 72274091], Guangdong Basic and Applied Basic Research Foundation [Grant number: 2022A1515011591], Planning Project of Guangdong Philosophy and Social Science [Grant number: GD23CGL06], and The Second People’s Hospital of Changzhi City Hospital-level Research Projects [CE202401].

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study was conducted according to the guidelines laid down in the Declaration of Helsinki. It obtained ethics approval from Southern Medical University. All participants gave informed consent to participate. The authors followed all measures to ensure the privacy and confidentiality of the participants, including excluding personal identifiers during data collection.

Consent for publication

Verbal consent was witnessed and formally recorded.

Competing interests

The authors declare no competing interests.