Dietary quality is associated with reduced risk of diabetes among adults in Northern China: a cross-sectional study

Abstract In this study, we analysed the prevalence of diabetes in Inner Mongolia and explored the relationship between dietary patterns and diabetes using the Chinese Dietary Balance Index-16 (DBI-16). This study was a surveillance survey of Chronic Disease and Nutrition Monitoring among Chinese Adults in Inner Mongolia in 2015. Dietary data were collected using the 24-h dietary recall and weighing method over three consecutive days. Dietary quality was evaluated via the DBI-16. A generalised linear model was used to examine the associations between the DBI-16 and dietary patterns. The relationship between dietary patterns and diabetes was analysed using logistic regression. In Inner Mongolia, the diabetes prevalence was 8·5 % and the estimated standardised prevalence was 6·0 %. Four major dietary patterns were identified: ‘meat/dairy products’, ‘traditional northern’, ‘high cereal/tuber’ and ‘high-salt/alcohol’. Generalised linear models showed that the ‘meat/dairy product’ pattern was relatively balanced (β LBS = –1·993, β HBS = –0·206, β DQD = –2·199; all P < 0·05) and was associated with a lower diabetes risk (OR 0·565; 95 % CI 0·338, 0·945; P < 0·05) after adjusting for potential confounders. The other three dietary patterns (i.e. ‘traditional northern’, ‘high cereal/tuber’ and ‘high-salt/alcohol’) exhibited relatively unbalanced dietary quality and were unassociated with diabetes risk. Diabetes prevalence in Inner Mongolia was moderate. The dietary quality of the ‘meat/dairy product’ pattern was relatively balanced and was correlated with a decreased risk of diabetes prevalence, suggesting that dietary quality may help decrease diabetes prevalence and provide a suggestion for local dietary guidelines.

Dietary factors are closely related to the development of diabetes. Among them, dietary intake is a major modifier in diabetes management (5) . Increased consumption of a healthy diet reduces the risk of diabetes (6) ; therefore, the association between dietary patterns and diseases is gaining attention. Prospective observational studies and clinical trials have evidenced the importance of a healthy diet in preventing and managing diabetes (7) . Examples include the Mediterranean-style diet, which emphasises synergy among various nutrients and foods (8) , and the Dietary Approaches to Stop Hypertension diet, which plays an important role in glycaemic control in long-term interventions (9) . One study found that adherence to the American Dietary Guidelines may reduce the risk of major chronic diseases (10) . Scholars often use the Alternate Healthy Eating Index to evaluate the dietary habits of local populations (11,12) . The Alternate Healthy Eating Index enables assessing the relationship between diet and chronic diseases. Additionally, a meta-analysis revealed that the Mediterranean diet, Dietary Approaches to Stop Hypertension and Alternate Healthy Eating Index are closely correlated with a reduced risk of diabetes, even with their different specific components (13) .
However, although the effects of a healthy diet on diabetes prevention are known, most studies are conducted in western populations in developed countries (14) . Therefore, the results of those studies may not be generalisable to populations with different dietary cultures in different regions, especially to certain Chinese populations who consume fewer whole grains and vegetables and more refined grains, red meat, unhealthy fats and fried foods (15) such as in northern China.
Inner Mongolia is a frontier in northern China, located southeast of the Mongolian plateau, extending 2400 km east to west and 1700 km north to south. Inner Mongolia accounts for nearly one-eighth of the territory in China, which differs geographically and culturally from other regions in China. Because the typical gathering areas are multiethnic and dominated by Han Chinese or Mongolians, Inner Mongolian residents exhibit unique dietary habits. However, large-scale, representative nutritional diabetes-related epidemiological studies are lacking in Inner Mongolia, in particular, research that explored the relationship between dietary patterns and diabetes. Therefore, we conducted this study to understand the diabetes prevalence in Inner Mongolia and explore the dietary quality in association with the risk of diabetes.

Study design
This study was a surveillance survey of Chronic Disease and Nutrition Monitoring among Chinese Adults in Inner Mongolia in 2015. This was a cross-sectional, population-based survey that characterised the dietary and nondietary factors (i.e. general demographics, lifestyle, diabetes prevalence rate, dietary behavioural habits and daily food intake) among residents aged ≥18 years in Inner Mongolia. The survey used a multistage-stratified cluster random sampling design covering eight monitoring sites including the urban, rural and pastoral areas in the eastern, central and western regions of Inner Mongolia.
The sample size estimation formula in this study according to the stratified random sampling was estimated based on the diabetes prevalence of the Health Service Survey in Inner Mongolia in 2012 (P = 0·0692). We considered a 0·015 permissible δ error value and a 10 % non-response rate, and we estimated the sample size to be 1209. In total, 1861 participants were included. According to the Health Service Survey in Inner Mongolia in 2012, the total number of participants surveyed was 21 500 (P = 0·0692) and our study was 1861 (P = 0·085). The statistical power (1 -β) was 71 %.
The Ethical Committee of the National Institute for Nutrition and Food Safety, Chinese Center for Disease Control and Prevention approved the survey (201519-A). All participants provided written informed consent before the start of the investigation.

Dietary data collection
Dietary data were collected from a 24-h dietary recall and weighing method over three consecutive days. The dietary survey was in accordance with the Chinese Dietary Guidelines for chronic disease and nutrition surveillance in Chinese adults and aimed to determine residents' nutrient and food intake. All household members (aged ≥2 years) were investigated. In the 24-h dietary recalls, participants recalled and described the types and quantities of food consumed over three consecutive days. Information on daily consumption of major seasonings, such as cooking oil, salt and monosodium glutamate, was collected using a weighing method. Condiments purchased and wasted were also recorded.
Drinking frequency, type (liquor with high alcohol content, liquor with low alcohol content, beer, yellow rice wine, rice wine, wine) and average drinking amount were measured. The average daily alcohol consumption was calculated according to the Manual of Chinese Chronic Disease and Nutrition Surveillance Survey (16) . Participants' height, weight and waist circumference were measured directly by trained investigators. Blood and urine samples were also collected. The laboratory director organised the sample quality control assessment at a field laboratory.
The DBI-16 contains three dietary quality indicators: the higher-bound score (HBS), the lower-bound score (LBS) and diet quality distance (DQD) (17) . The HBS is calculated by adding all positive scores as an indicator of excessive food intake. The LBS is calculated by adding the absolute values of all negative scores as an indicator of insufficient food intake. The DQD is an indicator of unbalanced food intake, calculated by adding the absolute values of both the positive and negative scores. The possible ranges for the HBS, LBS and DQD scores are 0-44, 0-72 and 0-96, respectively. Each indicator is divided into five levels to reflect dietary quality: (i) 'excellent' (a score of 0); (ii) 'good' (<20 % of the total score); (iii) 'low level' (20-40 % of the total score); (iv) 'moderate level' (40-60 % of the total score) and (v) 'high level' (>60 % of the total score) (17,19) . The total score of each DBI-16 component is divided by the total LBS, HBS and DQD to assess their contribution rate and how much of each food subgroup affects dietary quality.

Definition of diabetes
The main outcome indicator was type 2 diabetes. Meeting one of the following conditions was considered to indicate diabetes. The first condition was self-reported diabetes, in which patients were diagnosed with diabetes, are currently receiving diabetes treatment and/or are taking hypoglycaemic drugs. The second condition was that as per the International Diabetes Federation (2) , diabetes was diagnosed if one or more of the following criteria were met: (1) fasting plasma glucose ≥ 7·0 mmol/l; (2) glycated HbA1c ≥6·5 %; (3) 2-h plasma glucose ≥ 11·1 mmol/l following a 75 g oral glucose load and (4) random plasma glucose ≥ 11·1 mmol/l.

Definition of other variables
Age was categorised as 18-, 30-or ≥50 years. Participant household registration location was categorised as urban or rural. Based on regional characteristics, ethnic groups were categorised as Han, Mongolian or other minorities (i.e. all minorities living in Inner Mongolia except Han and Mongolian). Educational level was categorised as low (primary school or lower), medium (junior high school) or high (senior high school and above). Marital status was categorised as married, unmarried or widowed/divorced.
Abdominal obesity was diagnosed as per the Working Group on Obesity of China (20) : waist circumference ≥ 90 cm for men and waist circumference ≥ 85 cm for women.
Alcohol consumption was determined for the past year. Participants were categorised according to their average daily alcohol consumption: no drinking (0 g/d), moderate drinking (<25 g/d) or excessive drinking (>25 g/d) (21) .
Smoking status was categorised as non-smoker (never having smoked previously), ex-smoker (previously smoked but quit) or current smoker (has smoked at least 1 cigarette/d for more than 1 year and currently smokes). In the present study, metabolic exercise equivalent was calculated by the International Physical Activity Questionnaire (22) and physical activity was divided into low, medium and high levels.

Statistical analysis
Normally distributed continuous variables are presented as means and standard deviations; abnormally distributed continuous variables are presented as medians and interquartile ranges. Categorical variables are presented as numbers and percentages. Student's t test and the Mann-Whitney-Wilcoxon U test were used to assess two independent groups of continuous variables; the χ 2 test was used to assess categorical variables.
Principal component analysis was used to derive dietary patterns based on twenty-nine food groups. The varimax rotation was used to extract factor loadings. Factors were selected based on their eigenvalues (>1·00). The number of dietary patterns was determined based on scree plots, reasonability of food combination and variance contribution rate. Dietary patterns were named by combining the food composition characteristics of the dietary pattern with the main food types included. Factor scores for each pattern were calculated by adding the coefficient of the factor loading and the standardised daily intake amounts of every food type related to each pattern. Scores for each dietary pattern were categorised into quartiles Q 1 , Q 2 , Q 3 and Q 4 in ascending order of factor scores. Higher factor scores indicated more compliance with the dietary pattern; lower scores indicated more deviation.
Dietary pattern quality was evaluated using generalised linear models, with LBS, HBS and DQD as the dependent variables and dietary patterns as the independent variable after adjusting for other confounders. Multivariate logistic regression models were used to examine the association between dietary patterns and diabetes. The 'Forward: LR' method was used to select independent variables. With α = 0·05 as the significance level, P < 0·05 was considered statistically significant. All statistical analyses were conducted using IBM SPSS Statistics version 19.0 (IBM Corp.).

Dietary patterns
Principal component analysis identified four distinct dietary patterns. The most prominent dietary pattern (i.e. the pattern that explained the greatest amount of the total variance) was 'meat/dairy products' (eigenvalue: 2·615; 9·017 % of the total variance), characterised by milk tea and tea, fried wheat products, beef and mutton, milk and dairy products and had the highest observed positive loading factors. The second most prominent dietary pattern was 'traditional northern' (eigenvalue: 2·003; 6·908 % of the total variance), representing a typical traditional diet: high intake of starchy roots and products, pork, pickled vegetables/dried vegetables, maize and desserts. The third most prominent dietary pattern (eigenvalue: 1·875; 6·456 % of the total variance) was 'high cereal/tuber', characterised by intake of various vegetables, beans and bean products, wheat flour and products, fresh fruits, various aquatic products, eggs, vegetable oils, sugar, various beverages and nuts. The fourth (least prominent) dietary pattern was 'high-salt/alcohol' (eigenvalue: 1·503; 5·184 % of the total variance), characterised by high intake of salt, animal oils, condiments and alcoholic beverages. Together, these four dietary patterns accounted for 27·754 % of the variability in the original dietary variables. The Kaiser-Meyer-Olkin measure of sampling adequacy was 0·615, and Bartlett's test of sphericity was significant (P < 0·01), determining the suitability of the data set for the principal component analysis (shown as Table 2). Table 3 shows the food group characteristics of the four dietary patterns. For 'meat/dairy products', the dietary factor scores of the different foods for cereals, meat, milk and dairy products, eggs, vegetables, fish and shrimp, tea, fried food and salt intake were statistically significant between Q 1 and Q 4 (P < 0·05). Among them, Q 4 was higher than Q 1 for cereals, meat, milk and dairy products, vegetables, tea, fried food and salt. Egg, fish and shrimp intakes were higher in Q 1 than in Q 4 .

Dietary pattern characteristics
For 'traditional northern', the cereal, meat, vegetable and cooking oil intakes were higher in Q 4 than in Q 1 (P < 0·05). The milk and dairy product, fish and shrimp, tea and fried food intakes were lower in Q 4 than in Q 1 (P < 0·05). The soyabean, egg, fruit and salt intakes were not statistically different (P > 0·05).
For 'high cereal/tuber', the milk and dairy product, tea, fried food and salt intakes were lower in Q 4 than in Q 1 (P < 0·05). Intake of other food was higher in Q 4 than in Q 1 (P < 0·05).
For 'high-salt/alcohol', the dietary factor scores for the intake of foods other than fried foods were statistically significant between Q 1 and Q 4 (P < 0·05). Cereal, meat, soyabean, vegetable, fish and shrimp, cooking oil and salt intakes were higher in Q 4 than in Q 1 (P < 0·05). Milk and dairy product, egg, fruit and tea intakes were lower in Q 4 than in Q 1 (P < 0·05). Table 4 compares nutrient and energy intakes of the four dietary patterns.

Nutrient and energy intakes by dietary pattern
For 'meat/dairy products', the carbohydrate, fat, protein, Ca, Mg, K, Na, dietary fibre and energy intakes from protein were higher in Q 4 than in Q 1 (P < 0·05). Energy intake from carbohydrates was lower in Q 4 than in Q 1 (P < 0·05). For 'traditional northern', the carbohydrate, fat, protein, Ca, Mg, K, dietary fibre and energy intakes from fat were higher in Q 4 than in Q 1 (P < 0·05). For 'high cereal/tuber', energy intake from fat was lower in Q 4 than in Q 1 (P < 0·05) and the remaining nutrient and energy intakes were higher in Q 4 than in Q 1 (P < 0·05). For 'high-salt/alcohol', Ca intake and energy intake from carbohydrates and protein were lower Q 4 than in Q 1 (P < 0·05), while other nutrient and energy intakes were higher in Q 4 than in Q 1 (P < 0·05).

Discussion
Diabetes prevalence is related to economic level and tends to increase with societal and economic development (23) . In this study, Inner Mongolia had an 8·5 % diabetes prevalence, lower than that in high-income areas, such as Beijing (24) and Shanghai (25) , but higher than that in Jiangxi (26) and Guilin (27) . According to the age distribution of the 2010 population census in China, the estimated standardised diabetes prevalence was 6·0 % in Inner Mongolia, which was lower than the national levels in 2013 (28) and 2019 (2) . Therefore, diabetes is moderately prevalent in Inner Mongolia. In Inner Mongolia, four dietary patterns were identified: 'meat/dairy products', 'traditional northern', 'high cereal/tuber' and 'high-salt/alcohol'. These dietary patterns accounted for 27·754 % of the variability in the original dietary variables. And in other studies, the explicable variability for diet by dietary pattern ranged from 18·6 to 31·7 % (29)(30)(31)(32)(33) . Our result was consistent with other studies and was even higher than some studies.
Except for the 'meat/dairy product' pattern, our results showed no association between the other three dietary patterns and diabetes. Among them, the 'traditional northern' pattern was characterised by high potato, pork, pickled/dried vegetable, maize and dessert intake, which is similar to the typical traditional diet in northern China (50) . Characteristics of the 'high cereal/tuber' pattern were more vegetables, fresh fruits and aquatic products; thus, this dietary pattern is similar to the Dietary Approaches to Stop Hypertension diet but involved more cereals and meat, and the carbohydrate and Na intakes were higher than recommended.  Excessive alcohol and salt intake characterised the 'high-salt/alcohol' pattern, in which oil and salt intake was higher than recommended, and the vegetable, egg and dairy product intakes were lower than recommended. Therefore, we found no evidence that these dietary patterns help reduce the risk of diabetes. The relationship between dietary patterns and diabetes risk varies among populations with different dietary cultures in different regions and racial/ethnic groups (51,52) . Inner Mongolia across the eastern, central and western regions of north China has a multiethnic culture, thus formed a unique dietary culture. Although some a priori methods can assess a population's compliance with a specific dietary pattern, such as the Mediterranean and Dietary Approaches to Stop Hypertension and so on, but these high-quality diets included special foods and components such as olive oil and seafood. For geographical and climatic reasons, these types of foods are relatively lacking in Inner Mongolia. Thus, our study used the DBI-16 to assess the dietary quality in participants. The results showed that participants more adhered to the 'meat/dairy product' pattern, the dietary quality was relatively balanced. In our research, the relatively balanced dietary pattern was associated with a lower risk of diabetes, which was consistent with other study results (53,54) .
Our study had some limitations. Using a cross-sectional study prevents drawing conclusions about causality between dietary patterns and diabetes prevalence. More cohort and randomised controlled trials are needed. Although we simultaneously adjusted for major sociodemographic characteristics and lifestyle factors, residual unknown or unmeasured confounding factors may have been present.

Conclusions
In summary, Inner Mongolia exhibited a moderate diabetes prevalence and four major dietary patterns were identified. Among them, the dietary quality of the 'meat/dairy product' pattern as evaluated via the DBI-16 was relatively balanced and was correlated with a decreased risk of diabetes prevalence. However, more prospective studies are needed to determine the relationship between dietary quality and diabetes in Inner Mongolia.