There were two sets of data in the analysis. The primary data were from a cross-sectional survey designed and conducted by our team, the Chinese Urban Adults Diet and Health Study (CUADHS), conducted from March to July 2016, in which a multi-stage sampling method was utilised to recruit adult subjects. Firstly, eight cities were selected based on geographical location and economic status, including two first-tier cities with higher economic performance. Secondly, two communities from each first-tier city and one community from each non-first-tier city were chosen by convenience sampling. In the last step, subjects were recruited by age groups, with at least 60 people from the age group of 18–44 years, 60 from 45 to 64 years, and 50 from over 65 years in each community. The study recruited 1806 subjects, and those with a physical disability, mental illness, or memory problems and women who were pregnant were excluded from analysis. In the end, a total of 1520 subjects were considered eligible for this study. We obtained a formal written agreement from each participant.

The verification data were from the China Health and Nutrition Survey (CHNS) of 2009. The CHNS is an international collaborative project between the National Institute for Nutrition and Food Safety of the Chinese Centre for Disease Control and Prevention, and the University of North Carolina at Chapel Hill. As a longitudinal, household-based survey, it was conducted in 1989, 1991, 1993, 1997, 2000, 2004, 2006, 2009, and 2011 in sequence. Only the survey conducted in 2009 contained information about blood biochemical examination. Details are provided elsewhere [30]. Questionnaires and blood biochemical tests were used to collect information of adults (except pregnant women) living in urban areas (i.e., residence in city and town or county capital city). In this study, we analysed 2398 individuals from the CHNS who met our inclusion criteria described in the data collection section.

In the CUADHS, the interviewer-administered questionnaires contained questions about socio-demographic characteristics, lifestyles, disease history, health literacy, dietary intakes, and physical examination. Anthropometry measurement and blood tests were also conducted. The food frequency questionnaire (FFQ) and 24-h dietary recall for 1 day were used in this study to collect dietary intake information. The FFQ was a semi-quantitative questionnaire, in which 41 groups of food were asked about regarding the frequency and amount of consumption within the past month. Intakes of different kinds of alcohol and beverages were included. One-time 24-h dietary recall was used to obtain the data on food intakes, and then nutrient intakes were calculated based on it. About 1/8 (205 adults) of the subjects were invited to complete the 24-h dietary recall for 3 days to evaluate the representativeness of one-time 24-h dietary recall. Energy and nutrient intakes were calculated and analysed based on the Chinese Food Composition Tables (CFCT) of 2004 and 2009 (CFCT, National Institute of Nutrition and Food Safety, China CDC), the Standard Tables of Food Composition in Japan (2010), and the nutrient composition table on the food packaging. Training of the interviewers and a pilot investigation were completed prior to data collection.

Questionnaire information and blood biochemical indices of the CHNS 2009 were downloaded from the official website of the CHNS, and information was gathered about socio-demographic characteristics, lifestyles, disease history, 24-h dietary recalls for 3 days, consumption of household food inventory, energy and macronutrient intakes, anthropometry information, and blood biochemical indices labelled by specimen collection and processing by the China–Japan Friendship Hospital (CJFH).

For the CUADHS, we calculated energy and nutrient intakes based on CFCT, Standard Tables of Food Composition in Japan [31], and ingredient lists of common supplements. Then we computed the absorbed amount of special micronutrients by bioavailability adjustment, the detailed processing procedure of which is described elsewhere [32]. We referred to the FAO protocol [33] to calculate the probabilities of nutrient adequacy (PA) of some nutrients based on Chinese Dietary Reference Intakes (DRIs) 2013 edition, which included definitions and specified values for estimated average requirements (EAR) and recommended nutrient intakes (RNI). The equation was PA = Probnorm [(estimated intakes-EAR)/CV] for most nutrients except iron; the CV was set to 10% of EAR for all nutrients except vitamin A (20%), niacin (15%), and zinc (25%).

For iron, since its distribution was not normal, we used the eq. PA = estimated participant’s intake/RNI. PA was defined as the ratio of a certain nutrient intake to its recommended allowance, and when intakes of a certain nutrient exceeded requirement, the value should be capped at 1, indicating 100% adequacy. We calculated the PA of protein, carbohydrates, vitamin A, niacin, vitamin B6, vitamin B1, vitamin B2, vitamin B12, vitamin C, folate, calcium, iron, and zinc. We then calculated the mean probability of adequacy (MPA) of micronutrients by applying equal weight to every individual micronutrient.

We evaluated the representativeness of one-time 24-h dietary recall in the CUADHS, and the results are showed in Additional file 1: Table S1. There were no significant differences in energy and macronutrient intakes between one-time 24-h dietary recall and three-day 24-h dietary recall. We decided that one-time 24-h dietary recall was able to represent the results of 24-h dietary recall for 3 days, which was similar to other studies [34‐36].

In the CHNS the food intakes of every participant were the summation of two parts. Firstly, daily average food intake was calculated from three-day 24-h dietary recalls. Secondly, for food consumed at home, the daily amount of each ingredient from the household food inventory consumed by each individual was estimated based on his/her respective proportion of energy intake in the whole family.

For the two studies, all of the food items were divided into 15 groups according to the Dietary Guideline and Balance Diet Pagoda for Chinese Residents (as showed in Table 1), and, subsequently, the consumption of each food group was calculated, respectively. Though there were only 12 food groups in the Balance Diet Pagoda, in the detailed Dietary Guidelines there were precise recommended daily/weekly intakes of 15 food groups, in which cereals, tubers, and beans were divided into three groups: whole grains and legumes, tubers, and refined grains. Soya and nuts were divided into two groups: soya products and nuts and seeds [37]. According to similar research [28, 38], weight based on the recommended proportions of each food group at the 2000-kcal level in the Dietary Guideline and Balance Diet Pagoda for Chinese Residents was selected to build the Chinese HFD index (Table 1). There was no precise definition for dark green and yellow or orange vegetables in the recommendation, so we redefined them as vegetables rich in vitamin A (content of vitamin A ≥ 150 RAE/100 g; RAE = retinol equivalent) or vitamin C (content of vitamin C ≥ 50 mg/100 g). The Chinese HFD index values were calculated using the following algorithm:

Chinese HFD index=\( \left(1-\sum {S}_i^2\right)\times \mathrm{hv} \),

in which \( {S}_i=\frac{\ Amount\ of\ recommended\ food\ group\kern0.5em i\ by\ weight\ }{\ Amount\ of\ 15\ food\ groups\ by\ weight} \),

hv =  ∑ hf_i × S_i, hf_i = Broad food share × share of food group,

where S_i is the quantitative share of a single food group; hv is the health value; and hf_i is health factors.

Theoretically, the range of the Chinese HFD index is between 0 (i.e., a diet with a single food group) and nearly 1 (i.e., a more balanced diet). In fact, the maximum hv that can be achieved according to the recommendation at the 2000-kcal level is 0.187, and the Chinese HFD index was calibrated by dividing hv by its maximum. Then, the Chinese HFD index values and energy intakes were equally divided into four groups (Q1, Q2, Q3, and Q4).

We categorised age groups as 18~ 44.9 years old, 45~ 54.9 years old, 55~ 64.9 years old, and ≥ 65 years old. Smoking status was classified as never smoked, former smoker, and current smoker. Drinking behaviours were defined as not drinking and drinking now. Physical activity (PA) levels calculated from a PA questionnaire were trisected from the smallest to the highest and marked as T1, T2, and T3, which were expressed as metabolic equivalents (MET)/d; subjects who only reported sedentary behaviours were classified as a separate group. Educational background was classified as illiterate, middle school or lower, high school or professional training, undergraduate, and postgraduate student or higher. Body mass index (BMI) was defined as weight (in kilograms) divided by the squared height (in metres). Consistent with the WHO criteria, obesity, overweight, normal, and underweight were defined as BMI ≥28 kg/m², BMI ≥24 kg/m² and < 28 kg/m², BMI ≥18.5 kg/m² and < 24 kg/m2, and BMI<18.5 kg/m², respectively [39].

The definition of MetS was consistent with the most recent Joint Interim Statement (JIS) by adopting the Asian criteria for WC recommended by the International Diabetes Federation. Indicators of MetS are shown in Table 2, and subjects were diagnosed as patients if they had at least three of the indicators [5]. The methods of measuring anthropometry indicators and blood biochemical indices are presented in Additional file 2: Table S2.

Normality was examined before the related analysis. Values were presented as mean ± standard deviation (\( \overline{x}\pm SD \)) for continuous variables or as number (percentage) for categorical variables. Student t tests or one-way ANOVA were performed to compare means between different groups. Chi-square tests or trend Chi-square tests were performed to compare the distribution of categorical variables. Kendall’s tau-b correlation was tested between PA/MPA and the Chinese HFD index, with energy intake as a covariate. Binominal unconditional logistic regression models were used to estimate the effects of factors on MetS and the regression coefficients (ORs), and their 95% confidence interval (CI) was obtained. All statistical analyses were performed using the Statistic Package for Social Science (SPSS) version 20.0 (SPSS Inc., Chicago, IL, USA). P values lower than 0.05 were considered statistically significant.

There were 1520 adults analysed in the CUADHS, of which 527 were male. Those who were 18~ 44.9 years old and older than 65 years accounted for about 1/3 of the total, respectively. The educational attainment of about 70% participants was high school or lower. About half of the participants never smoked and were non-drinking now. The prevalence of overweight and obesity was 35.4% and 11.0%, respectively. The detailed distribution of demographic characteristics is showed in Table 3.

Table 3 also shows that in the CUADHS the prevalence of MetS was 36.4%, which was 29.0% after standardisation of age and gender by the 2010 Chinese national census. Significant differences in the prevalence of MetS were found between different groups of gender, age, education, smoking behaviour, BMI, and HFD index. The prevalence of MetS and its five components was significantly higher in participants who are male, older, poorly educated, a former or current smoker, and overweight or obese.

In the verification data of the CHNS, 2398 participants were analysed, and the distribution of gender (47.9% male) and age groups (30.9%, 25.4%, 20.7%, and 23.0%, respectively) was more balanced than that of the CUADHS. For the educational attainment, more than 85% of the participants were high school or lower. About 70% of the participants never smoked and were non-drinking now. The prevalence of overweight and obesity was 32.7% and 11.1%, respectively. Detailed information is shown in Table 4.

The crude prevalence of MetS in the CHNS was 27.3%, and the standardised prevalence was 19.5%. The prevalence of MetS was significantly higher in participants who are older, poorly educated, overweight or obese, and lack PA, while no similar trend was found in the five components of MetS.

In the CUADHS, the Chinese HFD index ranged from 0.04 to 0.63. The values of the index were significantly lower in participants who are male, young, poorly educated, drinking or smoking now, and with high energy intakes (Table 5).

To validate the Chinese HFD index, correlations between the Chinese HFD index and PA of nutrients were analysed, and the results are showed in Table 6. The Chinese HFD index was positively correlated with PA of carbohydrates, vitamin B2, niacin, vitamin B6, folate, vitamin A, vitamin C, and MPA of micronutrients after adjusting for energy.

In the verification study, the Chinese HFD index ranged from 0.02 to 0.62, as showed in Table 5. Participants who were male, young, poorly educated, drinking or smoking now, obese, and with the highest energy intakes had a significantly lower Chinese HFD index.

As shown in Table 7, in the CUADHS the Chinese HFD index was not significantly associated with MetS in covariate-adjusted models or its components. Only in an unadjusted model was the Chinese HFD index positively correlated with elevated fasting glucose and reduced HDL.

In the verification data of the CHNS, the results were different, as shown in Table 8. The Chinese HFD index was significantly negatively correlated with MetS and its components of elevated fasting glucose and elevated WC in covariate-adjusted models. It indicated that a higher Chinese HFD index decreased the risk of MetS.