Over recent decades, the worldwide prevalence of diabetes and impaired glucose tolerance in adults has increased drastically, especially in developing countries(Reference Shaw, Sicree and Zimmet1–Reference Guariguata, Whiting and Hambleton4). It has been estimated that 693 million or more people might suffer from diabetes worldwide by 2045, while 374 million people would become impaired glucose tolerance patients(Reference Cho, Shaw and Karuranga5). The prevalence of diabetes in China has more than quadrupled in past decades; an estimated 110 million adults have suffered from diabetes in 2010, and 490 million adults were diagnosed as pre-diabetes(Reference Xu, Wang and He6). Moreover, the prevalence of diabetes in Chinese rural areas was 4·1 %(Reference Bragg, Holmes and Iona7), sugesting that diabetes issues in rural China cannot be ignored.
Genetic variations play an important role in the development of the disease, while epidemic data indicate that lifestyle changes, as well as environmental exposures, contribute to an increased risk of type 2 diabetes mellitus (T2DM)(Reference Studart, Arruda and Sampaio8). It is well known that diet is one of the main affecting factors of a variety of diseases(Reference Frank, Kroger and Schulze9). Recent studies have shown that with the onset of diabetes, food choice and eating habits played a crucial role in controlling disease progression(Reference Colles, Singh and Kohli10).
There are many studies that assessed the relationship of dietary patterns and diabetes in China; however, there is a gap and controversies in this area(Reference Cai, Zhang and Nuli11–Reference Shi, Zhen and Zimmet14). The research of Cai et al. showed that the ‘meats’ and ‘grains’ dietary patterns were associated with an increased risk of T2DM, whereas the ‘dairy product’ dietary pattern showed no association with the risk of T2DM(Reference Cai, Zhang and Nuli11). The study of Shu et al. indicated that western dietary pattern was associated with an elevated risk, whereas the grains–vegetables dietary pattern was associated with a reduced risk of T2DM(Reference Shu, Shen and Li13).
In traditional epidemiology, the main approach is to establish the relationship between nutrients or food with chronic diseases, such the case of diabetes. However, human diet is a complicated issue, as food consumed in China is varied and there are synergistic or inhibitory effects among nutrients, making it difficult to verify these possible associations. Nevertheless, the use of dietary patterns may overcome these limitations via analysing the effects of multiple dietary factors on the individual’s health(Reference Frank, Kroger and Schulze9,Reference de Castro, Sichieri and Brito15) . Many diabetes-related studies have found that early lifestyle modifications, particularly healthy diets and exercises, were effective in delaying the onset of diabetes(Reference Tuso16).
Thus, it is important to know the eating patterns of diabetic individuals and their association with clinical aspects of the disease. In rural China, no studies have yet analysed the dietary patterns in both T2DM and pre-diabetes, nor their relationship with glycaemic control. In this study, we aimed to identify the main dietary patterns of patients with T2DM or pre-diabetes.
Data from the baseline survey of the Henan Rural Cohort were used to analyse the association between dietary patterns and T2DM. The Henan Rural Cohort was an ongoing cohort including 39 259 individuals aged 18–79 years in five rural regions of Henan province in China (Suiping county, Yuzhou county, Xinxiang county, Tongxu county and Yima county) at baseline (2015–2017). The cohort detail information is described elsewhere(Reference Liu, Mao and Li17). For the present analysis, we excluded participants with type 1 diabetes (n 4), gestational diabetes (n 64), malignant tumours (n 332) and kidney failure (n 18). We additionally excluded participants with missing information on fasting glucose (n 63). This resulted in a final sample size of 38 779 participants for analyses. And, all participants completed the informed consent.
Data were collected from a standard questionnaire including information on general demographic characteristics, lifestyles, personal and family history of diseases that was completed by well-trained investigators by face-to-face interview. Venous blood samples were collected for routine blood measurements, biochemical indexes after at least 8 h of overnight fasting. Fasting plasma glucose (FPG), TAG, total cholesterol and HDL-cholesterol were assayed by an automatic chemistry analyzer (Cobas C501, Roche Diagnostics GmbH). Glycosylated HbA1C was determined enzymatically on an automatic analyser. Insulin was measured using RIA. Anthropometric measurements included height, weight, waist circumference, resting blood pressure, heart rate and lung function. Subjects were in light clothing and without shoes when measuring anthropometric variables. Two measurements of height and weight were performed to the nearest 0·1 cm and 0·1 kg separately, on the basis of a standard protocol. Blood pressure measurements were performed after 5-min rest in seated position, with 30-s intervals between measurements, using an electronic sphygmomanometer (HEM-770AFuzzy). The average of the three measurements from the right arm of every person was applied to the analyses. All the measurements and tests were collected using standard protocol by trained staff. The detailed data collection protocol has been described elsewhere(Reference Liu, Yu and Mao18,Reference Liu, Li and Mao19) . BMI was calculated as weight divided by the square of height.
Dietary data were collected via a self-reported FFQ, which was evaluated for reliability and validity(Reference Xue, Yang and Wang20). A food consumption questionnaire contains eleven-item food groups including staple food such as rice, noodles and steamed bread; red meat such as pork, lamb and beef; white meat such as chicken and duck; fish including freshwater and marine fish; eggs such as chicken, duck and goose eggs; milk and dairy products including milk, goat’s milk, yogurt, cheese and other dairy products; fruits; vegetables; legumes and products like soya beans, mung beans, red beans, black beans, broad beans, tofu, soya milk, dried tofu, vegetarian chicken, etc.; nuts and peanuts; grains like corn, sweet potatoes and sorghum. Consumption questions covering intake of food groups during the previous 12 months were asked by a trained investigator. Each item had a choice of five frequency categories (never, daily, weekly, monthly, yearly) and the amount (g/ml) of food consumed at that frequency. For each question, the respondents were shown a card with samples of a standard portion of each food group relevant for typical diets.
Definition of type 2 diabetes mellitus and pre-diabetes
The definition of T2DM and pre-diabetes was based on the American Diabetes Association diagnostic criteria (2018)(21). After excluding type 1 diabetes, gestational diabetes and other special type of diabetes at the same time, pre-diabetes was defined as 5·6 mmol/l ⩽ FPG < 7·0 mmol/l, or 5·7 % < HbA1C < 6·5 %. T2DM was confirmed as FPG ⩾ p;7·0 mmol/l, or HbA1C > 6·5 %, or having a self-reported previous diagnosis of diabetes by a physician, or taking antiglycaemic agents during the previous 2 weeks.
Age was used on a continuous scale. Education was grouped into five categories (Illiteracy, Primary school, Middle school, High school, College and above) by asking the question ‘What is your highest level of education?’ Marital status was categorised as married, widowed, divorced and single. Smoking and drinking status were allocated to three categories as ‘never, ever, current’. Per capita monthly income was calculated by total household income and family size and categorised into five groups and expressed in Ren Min Bi (RMB: <500 RMB, 500–999 RMB, 1000–1999 RMB, 2000–2999 RMB, ≥3000 RMB. Physical activity level was estimated based on self-reported activities and grouped into three categories (low, medium and high).
Dietary patterns were identified by factor analysis using the standard principal component analysis method. The varimax method was used for rotation in the analysis. Patterns were determined by scree plot, eigenvalue (>1) and factor interpretability by each factor. Labelling of the patterns was primarily descriptive and based on our interpretation of the pattern structures. Factor loadings are equivalent to simple correlations between the food items and the patterns. Factor loadings were presented in tabular form. Participants were assigned pattern-specific factor scores, which were calculated as the sum of the products of the factor loading coefficients and standardised daily intake of each food group is associated with that pattern.
Continuous data are presented as mean and SD or median and quartile (median (p25, p75)). Categorical data are presented as number and percentage. The categorical data of each group were compared by χ 2 test. The Kruskal–Wallis method was used to assess the differences in non-normally continuous data. Normally continuous data using one-way ANOVA assessed the P value. The circus diagram and Spearman’s rank correlation coefficient were used to show the relationship between dietary patterns and variables. The association between diabetes/pre-diabetes and dietary patterns scores by using Multinomial logistic regression analysis and was adjusted for age, region, gender, education level, marital status and per capita monthly income in ‘Model 1’. Health behaviours (BMI, smoking, alcohol drinking, physical activity and family history of diabetes) were further adjusted in ‘Model 2’. ‘Model 3’ was the final adjusted model by additional energy. For the quintiles of each dietary pattern, the first quintile (Q1) was considered as the reference group. Restricted cubic spline(Reference Desquilbet and Mariotti22) was used to explore the dose–response relationship between continuous factor scores with T2DM and pre-diabetes, with five knots located at 17th, 33rd, 50th, 67th and 83rd percentiles of factor scores. All the analyses were conducted using SPSS21.0 (IBM Corp.) and R software (version 3.5.3). Statistical significance was considered when P < 0·05 with two-sided tested.
Characteristics of study participants
A total of 38 779 remaining participants (15 368 males and 23 411 females) were included in the final analysis, and the mean age was 55·58 years. Table 1 presents the general characteristics of the participants according to case and control groups. The prevalence of pre-diabetes and T2DM was 6·8 % and 9·4 %, respectively. Statistically significant differences in gender, age, education level, marital status, region, per capita monthly income, physical activity, family history of diabetes, smoking and alcohol drinking were observed among different groups (P < 0·05). Furthermore, height, weight, BMI, waist circumference, as well as expressions of total cholesterol, TAG, insulin, fasting glucose and HDL-cholesterol also differed significantly among different groups (Table 2). Table 3 displays the differences in food consumptions among three groups. According to our study, participants with T2DM were more likely to consume less red meat, white meat, eggs and fruits. Furthermore, participants with pre-diabetes tended to consume fewer staples, red meat, fish, milk, vegetables and nuts in daily life.
T2DM, type 2 diabetes mellitus; RMB, Ren Min Bi.
* Continuous data are presented as mean and sd using one-way ANOVA assessed the P value.
† Categorical data are presented as numbers and percentages, and using χ 2 test.
WC, waist circumference; TC, total cholesterol.
* P value was assessed using the Kruskal–Wallis method.
* P value was assessed using the Kruskal–Wallis method.
Definition of dietary patterns
The three-cluster dietary patterns were obtained according to scree plot, eigenvalue (Fig. 1, see online Supplemental Fig. S1) and factor interpretability (see online Supplemental Table S1), with an accumulating contribution rate of 43·56 %. Factor 1, the ‘meat’ pattern, characterised by a high intake of red and white meat. Factor 2, the ‘grains-nuts-egg’ dietary pattern, stands for a high intake of grains, nuts, milk and eggs. Factor 3, the ‘vegetables-staple-fruits’ pattern, is featured by a high consumption of vegetables, staple food and fruits.
Figure 1 exhibits the proportion of dietary patterns components’ number of each group. The normal subjects conformed to ‘meat’ pattern, ‘grains-nuts-egg’ pattern and ‘vegetables-staple-fruits’ pattern were 31·6 %, 32·1 % and 36·5 %, respectively. Among people with pre-diabetes, the proportions of each pattern were 28·1 %, 40·0 % and 31·9 %. 30·9 % of T2DM patients were subjected to ‘meat’ pattern, 36·3 % with ‘grains-nuts-egg’ dietary pattern and 32·8 % with ‘vegetables-staple-fruits’ pattern.
Figure 2 and Table S2 show the relationship among dietary patterns with insulin, fasting glucose and food groups. The width of the ribbon represents the intensity of the correlation (in positive correlation). The top three correlation coefficients for the ‘vegetables-staple-fruits’ pattern were vegetables, staple and fruits. Egg, milk, bean and nut were closely related to the ‘grains-nuts-egg’ dietary pattern. For ‘meat’ pattern, red meat, white meat and fish were on the top of the menu. Furthermore, patients with ‘vegetables-staple-fruits’ pattern showed lower levels of fasting glucose and insulin, this pattern was negatively correlated with FPG (r = −0·106, P < 0·001) and positively correlated with insulin (r = 0·174, P < 0·001), while ‘meat’ pattern was also negatively related to the fasting glucose level.
Table S3 presents the food consumption of each group based on the quintiles of different pattern factor scores. Under certain dietary patterns, the consumption of characteristic foods grows as the factor score increases. Take ‘meat’ pattern for example, the intake of fish, red and white meat advanced as the scores grew.
Relationship between dietary patterns and pre-diabetes
Table 4 and Fig. 3 show the associations between dietary patterns and pre-diabetes. In the crude model, the highest quintile (Q5) of the meat pattern was significantly associated with a reduced risk of pre-diabetes (OR: 0·678, 95 % CI (0·596, 0·771), P trend < 0·001) as compared with the lowest quintile (Q1). After adjusting for covariates (models 1–3), the ‘meat’ pattern was not significantly correlated with pre-diabetes. There was no significant correlation between ‘grains-nuts-egg’ dietary pattern and pre-diabetes in both crude and adjusted models. In all models, the highest quintiles of ‘vegetables-staple-fruits’ pattern were significantly associated with reduced risks of pre-diabetes (OR crude: 0·616, 95 % CI (0·540, 0·703), P trend < 0·001); (OR model 1: 0·690, 95 % CI (0·602, 0·791), P trend < 0·001); (OR model: 2 0·708, 95 % CI (0·616, 0·813), P trend < 0·001); (OR model 3: 0·683, 95 % CI (0·569, 0·820), P trend < 0·001, respectively). There was a linear relationship for ‘vegetables-staple-fruits’ pattern in pre-diabetes according to the restrictive cubic spline.
T2DM, type 2 diabetes mellitus.
* Crude: unadjusted.
† Model 1: adjusted for age, region, gender, education level, marital status and per capita monthly income.
‡ Model 2: further adjusted for BMI, smoking, alcohol drinking, physical activity and family history of diabetes based on model 1.
§ Model 3: further adjusted for energy based on model 2.
‖ Factor scores were divided into quartiles.
¶ For the quintiles of each dietary pattern, the first quintile (Q1) was considered as the reference group.
** Factor 1, ‘meat’ pattern.
†† Factor 2, ‘grains-nuts-egg’ dietary pattern.
‡‡ Factor 3, ‘vegetables-staple-fruits’ pattern.
The relationship between dietary patterns and type 2 diabetes mellitus
The OR of T2DM across quintiles of dietary patterns are shown in Table 4. Before adjustment (crude model), the highest quintile (Q5) of the ‘meat’ pattern was significantly associated with a reduced risk of T2DM (OR: 0·720, 95 % CI (0·645, 0·804), P trend < 0·001) when compared with the lowest quintile (Q1). After adjusting for covariates (models 1–3), the ‘meat’ pattern was not significantly correlated with T2DM. Nevertheless, the restrictive cubic spline suggested a risk trend between ‘meat’ pattern and T2DM after fully adjusted.
The highest quintile of the ‘grains-nuts-egg’ dietary pattern was not significantly associated with a decreased risk of T2DM (OR: 0·942, 95 % CI (0·846, 1·049), P trend = 0·343); however, there was a significant association in the fully adjusted model (OR: 0·845, 95 % CI (0·747, 0·956), P trend = 0·125), and this dose–response relationship was shown to be U-shaped by restrictive cubic spline (Fig. 3 ).
The risk of T2DM decreased gradually as the intake of ‘vegetables-staple-fruits’ pattern increased. In both unadjusted and adjusted models, the highest quintiles of ‘vegetables-staple-fruits’ pattern were significantly associated with reduced risks of T2DM (OR unadjusted: 0·634, 95 % CI (0·567, 0·708), P trend < 0·001); (OR adjusted: 0·582, 95 % CI (0·497, 0·682), P trend < 0·001, respectively). A linear relationship between ‘vegetables-staple-fruits’ pattern and T2DM was shown by restrictive cubic spline (Fig. 3).
Our study suggested that ‘vegetables-staple-fruits’ pattern is beneficial for reducing risk of pre-diabetes or T2DM, which is consistent with the results of previous studies(Reference Medina-Remon, Kirwan and Lamuela-Raventos23–Reference Papamichou, Panagiotakos and Itsiopoulos26). Plant-based dietary patterns, which are rich in fruits, vegetables and whole grains, play a pivotal role in preventing various chronic diseases. A research reported that non-vegetarians were 74 % more likely to develop diabetes than vegetarians over 17 years of follow-up (via a 17-year period-study)(Reference Vang, Singh and Lee27). Plant-based diet may help prevent and treat diabetes, possibly by improving insulin sensitivity and ameliorating insulin resistance(Reference McEvoy, Temple and Woodside28). Reports show that fruits, vegetables and whole grains may improve glycaemic control and reduce carbohydrate production and the risk of diabetes due to their low energy density, low glycaemic load, high fibre and antioxidant content(Reference Tonstad, Stewart and Oda25,Reference Psaltopoulou, Panagiotakos and Pitsavos29) . Similarly, vegetarian and vegan diets have also been shown to improve plasma lipid concentrations and reverse atherosclerosis progression(Reference Barnard, Katcher and Jenkins24). What is more? Vegetarian diet seems to be more effective in controlling glycaemia and plasma lipid concentrations(Reference Barnard, Cohen and Jenkins30). In fact, the dietary pattern of high fruits and vegetables is indeed beneficial for the prevention of T2DM(Reference Shu, Shen and Li31).
On the contrary, evidences have linked red meat intake to an increased risk of CAD, cancer and T2DM(Reference Pounis, Tyrovolas and Antonopoulou32). which is in consistent with our results. After adjusting for covariates, the ‘meat’ pattern was not significantly correlated with pre-diabetes, whereas there was a risk trend between ‘meat’ pattern and T2DM. High consumption of meat, especially for red meat which contains high amounts of saturated fat and cholesterol, was associated with an increased risk of obesity, an important risk factor for T2DM(Reference Shu, Zheng and Zhang33). Meanwhile, higher intakes of red meat may result in more absorption of haem iron(Reference Maghsoudi, Ghiasvand and Salehi-Abargouei34). Previous studies have also demonstrated that body iron overload may induce insulin resistance and increase the risk of T2DM(Reference Jiang, Ma and Ascherio35). Worsely, advanced glycation of high-fat products and meats can enhance oxidative stress and inflammation, which in turn accelerates insulin resistance and increases the risk of T2DM(Reference Palli and Consortium36).
The ‘grains-nuts-egg’ dietary pattern, characterised by high consumption of grains, nuts, egg, beans and milk, reduced the risk of T2DM. However, no such correlation was seen with pre-diabetes. A recent report by Song et al. has shown that higher consumption of dairy products(Reference Gijsbers, Ding and Malik37,Reference Tong, Dong and Wu38) and nuts(Reference Afshin, Micha and Khatibzadeh39) was associated with a reduced risk of T2DM(Reference Song and Lee40). In addition, a dose–response meta-analysis of prospective cohort studies revealed an inverse correlation between dairy foods, particularly yogurt, with T2DM(Reference Gijsbers, Ding and Malik37). Some investigators(Reference Abiemo, Alonso and Nettleton41) stressed that whole-fat dairy could be detrimental, for that an inverse association between diary food and T2DM was seen most frequently in higher yogurt and low-fat milk intake rather than high-fat milk or other dairy products such as cheeses(Reference Gijsbers, Ding and Malik37,Reference Liu, Choi and Ford42) . Our study failed to establish a similar inverse correlation with pre-diabetes since our dairy food consumption category referred to all kinds of dairy foods.
The present study holds several strengths. First, to our knowledge, this is the first study to focus on the relationship between dietary patterns and both risks of pre-diabetes and T2DM among the rural adults in China. Our findings provided valuable information for the primary prevention of T2DM, as well as pre-diabetes, through dietary modifications to the rural Chinese population. Second, this survey is a large-scaled cross-sectional study of its kind in rural China. Third, in our analyses, we have also adjusted for some known and proposed potential confounders in the statistical model. However, some potential limitations still warrant a mention. First, based on the cross-sectional design, we cannot evaluate the causal associations between dietary patterns and the risk of pre-diabetes and T2DM. However, the cross-sectional research reflects the individual exposure and outcome at the time of the investigation, which helps to indicate the risk factors of the disease. Second, the definition of T2DM based on FPG and HbA1C was flawed. Thus, additional diagnostic tests were needed in future research. Third, factor analysis, which was used for data reduction, has limitations. The definition of the dietary patterns, including determination on number of factors, the type of rotation, as well as the interpretation and naming of the factors, involves subjective decisions (may not be so objective)(Reference Shu, Shen and Li31). We used rotation in the analysis to make the extracted factors more explanatory. Fourth, we had no further calculation of nutrients, since FFQ was collected according to food group. For further researches, we will evaluate the reliability and validity of the amounts of nutrients obtained through FFQ by calculating the nutrient weighting coefficients for each food group based on data from the 3-d 24 h-dietary review of a small sample. Fifth, Differences between the group with normal glucose metabolism and the group with diabetes may reflect changes in diet as a result of the diabetes diagnosis. We lack the investigation of newly diagnosed diabetes; nevertheless, we investigated the diet of patients with pre-diabetes, which provides a reference for the prevention of diabetes on dietary precautions.
The ‘vegetables-staple-fruits’ pattern is beneficial for reducing risk of pre-diabetes or T2DM. People with pre-diabetes should consume more fruits and vegetables instead of meat or eggs/milks. Although a higher consumption of ‘grains-nuts-egg’ dietary patterns may be associated with a reduced risk of T2DM, excessive intakes should be avoided. This study may provide a reference for the prevention of diabetes on dietary precautions.
Acknowledgements: We are grateful to the villagers who took part in this study, the research team and hospital staff for their cooperation and assistance. Financial support: This work was supported by the National Key Research and Development Programme Precision Medicine Initiative of China, grant number: 2016YFC0900803; National Natural Science Foundation of China, grant numbers: 81872626 and 82003454; Chinese Nutrition Society – Bright Moon Seaweed Group Nutrition and Health Research Fund, grant number: CNS-BMSG2020A63; Chinese Nutrition Society – Zhendong National Physical Fitness and Health Research Fund, grant number: CNS-ZD2019066. Conflict of interest: The authors declare no conflicts of interest. Authorship: Conceptualisation, W.-J.L. and C.-J.W.; methodology, Y.X.; investigation, C.L., B.-Y.W., D.-D.Z. and Y.W.; data curation, Z.-X.M. and S.-C.Y.; writing-original draft preparation, Y.X.; writing-review & editing, X.L. Ethics of human subject participation: This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human subjects/patients were approved by the Ethics Committee of Zhengzhou University. Ethic approval code: (2015) MEC (S128). Written informed consent was obtained from all subjects/patients.
For supplementary material accompanying this paper visit https://doi.org/10.1017/S1368980021000227