Dietary patterns and the effect of long-term PM2.5 exposure on metabolic syndrome among Chinese adults: a cross-sectional study

Limited evidence exists regarding the causal effects of air pollution and metabolic syndrome (MetS), as well as the potential moderating effect of adherence to healthy dietary patterns. We recruited participants with accessible clinical characteristics, dietary patterns, and blood biomarkers data from the 2009 China Health and Nutrition Survey. Multi-biomarkers combined with physical examinations were used to determine the status of MetS. The fine particulate matter (PM2.5) concentration was calculated by the near-real-time historical data at the resolution of 1 km. The control function (CF) combined with probit model (CF-Probit) was used to determine the effect of long-term PM2.5 exposure on MetS risk, with wind speed serving as the instrument. We assessed the dietary patterns of each participant using the dietary balance index (DBI). The modification analyses were conducted to investigate the potential moderating role of dietary patterns. The study included 4,277 adult participants participated with a mean age of 50.18 years and an incidence of MetS of 22.38% (957/4277 cases). The mean score for the DBI was 40.23. The mean long-term PM2.5 level was 65.79 μg m−3. The CF-Probit marginal effects analysis showed significant causal effects of chronic PM2.5 exposure on MetS incidence, with a marginal effect of 0.013 (95% confidence interval (CI): 0.003–0.022), suggesting that the average partial effect of long-term PM2.5 level on the risk of MetS in adults is 1.3 percentage points. The modification analysis indicated that the average partial effect of PM2.5 level on the risk of MetS is higher for male compared to female (4.22 pencetage points, 95% CI: 2.12 percentage points, 6.35 percentage points) and greater associated with unhealthy dietary patterns (1 percentage point, 95% CI: 0.17 percentage points, 1.86 percentage points). This study found that long-term exposure to PM2.5 increases the risk of MetS, while a healthy dietary pattern can modulate this effect. The findings can provide scientific basis for health protection guidelines for air pollution and provide dietary recommendations for populations.


Introduction
Metabolic syndrome (MetS), characterized by a cluster of metabolic risk factors, is recognized as a significant global public health concern due to its increasing high prevalence (affecting approximately 25% of the population) [1].MetS may heighten the short-term and long-term risk of systemic metabolic diseases, including but not limited to coronary heart disease [2], type-2 diabetes mellitus [3], fatty liver disease [4], and subsequent increased risk of all-cause mortality [2].Previously, unbalanced lifestyles and inherent genetic susceptibility were regarded as key factors in mediating the state of MetS [5].However, these findings were insufficient to fully elucidate the underlying reasons for the high prevalence and escalating trend of MetS.Identifying additional potential risk factors for MetS development could significantly contribute to reducing its incidence and further enhancing life expectancy, particularly in developing and resource-limited countries [1,6].
Importantly, emerging evidence indicates that environmental exposures such as ambient fine particulate matter (PM 2.5 ) pollution were adversely associated with metabolic dysfunction in multiple organs [7,8].In most areas of China, PM 2.5 has been and continues to be a major environmental concern [9].Accumulating clinical studies and basic experiments have identified the positive associations between PM 2.5 and oxidative stress, systemic proinflammation modification as well as insulin resistance, which may consequently increase the risk of various chronic cardiometabolic diseases [10].Furthermore, experimental research has highlighted that PM 2.5 could induce cell death via different molecular patterns and multiple signaling pathways [11].Therefore, studies for evaluating the association between long-term PM 2.5 exposure and the risk of MetS draw much attention in recent years [12,13].However, existing findings from observational studies and meta-analyses have not shown consistent results [14,15].Moreover, there is a shortage of well-designed studies that assess the causal effects of between long-term PM 2.5 exposure and MetS.This is particularly important, as reverse causality, omitted variables, and other factors often introduce biases in assessing the association between long-term air pollution and health outcomes [16][17][18].
Recently, dietary patterns have been noted to have a potential modifying role in reducing adverse outcomes such as cardiovascular disease risk, and brain aging after long-term exposure to air pollution [19].Furthermore, compelling evidence has shown that the dietary pattern had a strong correlation with MetS, as it could exacerbate or alleviate the MetS state depending on different dietary strategies [20].Thus, the modifying role of dietary patterns in MetS development and progression inspired us to further investigate whether dietary patterns might influence the underlying association between PM 2.5 exposure and MetS risk [21].
To fill the research gaps in this field, we aimed to investigate the effects of long-term PM 2.5 exposure and the MetS risk in the setting of a nationwide cohort from China and to further explore the potential modifying effect of dietary patterns on PM 2.5 exposure and MetS.

Study population
The study population in the present study was derived from the China Health and Nutrition Survey (CHNS), which is a population-based prospective study recruiting participants from 12 provinces and autonomous cities in China [22].The detailed description can be reviewed in previous studies [22].Specifically, the CHNS, although not a nationally representative survey, covered four regions of China, namely Northeast, East Coast, Central, and Western, and thus captured different levels of socioeconomic development.To comprehensively assess the clinical characteristics of the participants, we conducted a cross-sectional study from the wave 2009 CHNS that included the biomarkers of the participants.The selection process and the distribution of PM 2.5 concentrations and districts/counties visited in the study were shown in figure 1.
The CHNS was approved by institutional review boards at the University of North Carolina (Chapel Hill, NC, the US) and the National Institute of Nutrition and Food Safety (Chinese Center for Disease Control and Prevention).Informed consent was given to each participant before enrollment.

Covariate controlling
The different levels of demographic characteristics of the study population were controlled for, mainly at the community and individual levels, based on the questionnaires obtained from the participants.Sixteen covariates were identified in our study.These included age at interview, gender, race, educational level, employment status, body mass index (BMI), 3 day average energy, hypertension, diabetes mellitus, medical history, drinking habits, smoking habits, type of cooking fuel, use of air conditioning, income, and place of residence.The detailed variables' definitions and classifications were summarized in table S1.We also obtained daily gridded PM 10 data at the resolution of 1 km from the China high air pollutants database [23].We calculated the long-term PM 10 exposure levels for each adult over the twelve months before the survey.

PM 2.5 and wind speed measurement
We obtained daily PM 2.5 data for China covering the study time from the tracking air pollution in China (TAP) database at the resolution of 1 km, which was analyzed by using a machine learning algorithm combined with a generalized summation model [24].The TAP generated the long-term historical near-realtime high resolution air pollution data and the stability has been previously validated [25].PM 2.5 concentrations were calculated in µg m −3 .We calculated for respondents at the district and county level the average of the PM 2.5 over the twelve months before the survey.
Wind speed was selected as an instrumental variable for this study because of its good qualities.The specific reason was that we wanted to select a factor that was associated with local PM 2.5 concentration and, as far as possible, not associated with health outcomes.Wind speed has been repeatedly shown to be a good instrumental variable for PM 2.5 in previous studies [26,27].We obtained gridded monthly wind speed data from ERA5-Land covering the entire study period in China at a resolution of 1 km [28].We calculated wind speed data for the twelve months before the respondent interviews.The variables are in meters (m).

Metabolism syndrome definition
The definition of MetS was based on the previous studies for Chinese adults [12,29].For the Chinese population, MetS are defined as participants with a waist circumference of 90 cm for men or 80 cm for women, which is the primary condition.Meanwhile, they must meet two or more of the following criteria: (1) triglyceride (TG) levels ⩾1.7 mmol l −1 ; (2) high density lipoprotein-cholesterol (HDL-C) levels <1.03 mmol l −1 for men or <1.29 mmol l −1 for women; (3) glucose levels ⩾5.6 mmol l −1 or receiving anti-diabetic therapy; (4) abnormal blood pressure: systolic blood pressure (SBP) ⩾130 mmHg/diastolic blood pressure (DBP) ⩾85 mmHg or receiving antihypertensive therapy.

Collection and assessment of dietary intakes
In the CHNS, dietary assessment is based on a combination of data collected at the individual level and a food inventory taken at the household level.The dietary data of each participant were collected by using 24 h of dietary on three consecutive days.The daily intake on average for each food item was then calculated according to the product of the intake frequency and the amount consumed at each time (in grams per day).
The dietary quality of each participant was evaluated by using the Chinese diet balance index 2007 (DBI-07), a scale for assessment of the overall dietary quality in the Chinese population in the previous works [30,31].There were eight indicators calculated in the DBI including the intakes of cereal, intakes of vegetables and fruits, intakes of dairy and soil products, intakes of animal food, intakes of oil, salt and alcoholic beverage, and components of dietary variety.In this study, information on drinking water was unavailable.Thus, this indicator was excluded for further analysis.Based on the dietary guideline recommendations for Chinese people, each type of food is weighted with a score (table S2).Based on the diet quality index for China [32], the dietary variety was further divided into four components for evaluation, including cereals and tubers, vegetables and fruits, dairy products, soybean and soybean products, and animal food (table S3).The dietary quality distance (DQD) was mainly used to show the dietary pattern of each participant, which could assess unbalanced food intake by adding the absolute values of both positive and negative scores (the possible range of DQD was 0-76).

Model specification and estimation technique
To explore the robust effect between long-term PM 2.5 exposure and MetS in adults, we first fitted the multivariable binary probit model controlling a series of confounders and the unobservable but fixed effects of provincial.Besides, the multivariable binary probit model is a useful benchmark, the link between longterm PM 2.5 exposure and the risk of MetS can be modeled as: where MetS i represents the MetS status of individual i.PM 2.5i refers to the long-term PM 2.5 exposure concentration of individual i. γ 1 is the coefficient, and a positive value indicates that PM 2.5 is positively associated with increased risk of MetS.X i is a set of covariates; 1 (•) is an indicator function that takes value 1 if the inequality inside it is true.ε i is a standard normal random effect term.Taking the expectation of the dependent variable given the independent variable yields the conditional expectation in probit form.
However, an important methodological problem for equation (1), which leads to the difference of the coefficients between the latent variable form and conditional expectation form, is the potential endogeneity due to omitted variables or the possible bidirectionality between PM 2.5 and health outcomes [16,33].A frequently raised concern is that when faced with air pollution, individuals with higher risks are more likely to take steps to avoid potential environmental risks, such as relocating to a location with better air conditions [34].An instrumental variable (IV) probit model could be a good approach to solve this issue.Since the effect of modification of dietary patterns cannot be clearly defined in the basic IV probit model, the control function approach (CF, also known as the two-stage residual inclusion) is the best choice [35].
The CF model can correct the endogeneity problem by applying the idea of the IV approach to model the endogeneity in the error term [36].We use the contemporaneous annual wind speed at the residential address as an instrument variable.Wind speed is directly related to PM 2.5 exposure and not directly related to individual MetS, except through air pollution [26,37].In other words, the only channel for wind speed to affect MetS is through its impact on PM 2.5.Therefore, we used equation (2) to model the relationship between wind speed and PM 2.5 concentrations, while equation ( 3) was an extended version of equation ( 1) to obtain probit CF estimates.
where Z i represents the instrumental variable, V i refers to the residuals fitted by equation ( 2), Φ refers to the cumulative normal distribution function, other definitions are the same as in equation ( 1), and the ρ subscript denotes that each coefficient is multipled by The ideal exogenous instrument (Z) needs to satisfy two prerequisites.(1) It can adequately explain the change in PM 2.5 exposure (also known as the relevant prerequisite), and (2) it cannot also independently explain whether the respondent experiences MetS (also known as the valid exclusion restriction).We firstly simply test the relationship between wind speed and PM 2.5 concentrations, which was reported in figure S1.We then used the Anderson-Rubin test (AR test) and the Wald test to test the first prerequisite, and the relevant prerequisite was valid when the null hypothesis of both the AR test and the Wald test was rejected at the 5% level, and both exceeded 3.2.As for the second prerequisite, since the number of instruments equals the number of endogenous regressors in this paper, this prerequisite cannot be directly verified by statistical methods, such as sargan tests, and is usually supported by relevant evidence from previous literature, which has been corroborated by sufficient evidence [27,37].
On the other hand, we were also interested in whether maintaining good dietary patterns and sex differences could mitigate the risk of MetS from PM 2.5 .Therefore, we additionally explored the modification analysis as modeled below: where ε′ i = ηV i + e i , with ε′|Z i , X i ∼ Normal(0, 1), and e i |V i , Z i , X i ∼ Normal(0, 1 − η 2 τ 2 ).MOD represents the modification factors.X * i represents control variables other than MOD.PM c 2.5 and MOD C are PM 2.5 and MOD after separate normalization process, which is done to facilitate the interpretation of the results.The ρ′ subscript denotes that each coefficient is multipled by (1 − η 2 τ 2 ) −1/2 .λ 3 is the coefficient of the interaction term.
In the cohort, there is intermittent missing data and wear and tear during data collection and followup.Therefore, the multiple imputation method was used to impute data for those participants that had missing variables other than key variables and with a missing rate of less than 10%.We also conducted the same regression analysis based on the data with the missing values removed to test the robustness.The details of data missing and data distribution after multiple imputations are shown in figures S2 and S3.Robust standard errors obtained by bootstrapping were used to avoid potential sampling errors in the two-stage estimation.In addition, to address concerns that the results may be affected by the individuals' hypertension and diabetes status, we further investigated the results without controlling for prevalence of diabetes and hypertension.To control for the effects of other pollutants, we further adjusted for long-term exposure levels to PM 10 on top of the main regression, aiming to examine the robustness of the results.All data management was performed in R 4.0.2.Statistical analyses were performed in Stata (version 15.0 SE, Stata Crop, Chicago, IL, USA).P < 0.05 was used to determine statistical significance.

Descriptive statistics
We finally included 4,277 participants from 144 counties/districts with a mean age of 50 years and slightly more females than males (53.17% vs. 46.83%).The basic socio-economic and demographic characteristics of the included participants are shown in tables 1 and S4.Specifically, the total incidence of MetS was 22.38% and the mean score of DBI was 40.23 (standard deviation [SD]: 8.11).The mean long-term PM 2.5 level of each individual was 65.79 µg m −3 (SD: 23.5) and the mean wind speed was 0.96 m s −1 (SD: 0.24).

Causal effect of long-term PM 2.5 exposure on MetS
The result of classic probit model is displayed in column (1) of table 2. As discussed in the part of methods, the multivariable binary probit model cannot avoid the endogeneity of PM 2.5 caused by reverse causality, and the estimates may be biased, leading to false conclusions.
Column (2) of table 2 presents the findings from the probit CF model.The result reveals that the residual of PM 2.5 , obtained by incorporating instrument, is statistically significant even at the 1% level.This suggests the presence of endogeneity in the analysis, which calls for an appropriate empirical strategy to address it.Therefore, the following discussion primarily focuses on the results derived from the probit CF model.Specifically, the coefficient of 0.058 (95% confidential interval [CI]: 0.014, 0.102) for PM 2.5 signifies that long-term exposure to PM 2.5 is associated with an increased risk of MetS in adults.We further reported the marginal effect based on the same model.The results indicate a marginal effect of 0.013 (95% CI: 0.003, 0.022), indicating that the average partial effect of long-term PM 2.5 levels on the risk of MetS in adults is 1.3 percentage points.
Furthermore, upon comparing the results of the multivariable binary probit model and the probit CF model, we observed a contrasting direction in the key coefficient, specifically PM 2.5 .This indicates that neglecting the endogeneity would substantially impact the results, potentially leading to a different conclusion.These findings underscore the importance of employing causal inference methods in this study.

Modification analysis
To further investigate whether maintaining healthy dietary patterns and sex differences could mitigate the risk of MetS associated with PM 2.5 exposure, we examined models (4) and ( 5), and the corresponding marginal results are presented in tables 3. To facilitate interpretation with real-world meaning, we centered continuous variables and converted the results to the effects of one SD change in PM 2.5 and DBI.Specifically, one SD increase in DBI score from the mean value would lead to a 1 percentage point (95% CI: 0.2, 1.9) increase in the average partial effect of PM 2.5 on the risk of MetS.Furthermore, we found that the average partial effect of PM 2.5 on the risk of MetS would be 4.2 percentage points (95% CI: 2.1, 6.3) higher for male compared to female.

Robustness check
We used data with missing values directly removed to test the robustness of the results based on the same analysis.All results were consistent with those we obtained in the main regression, thus testing the robustness of our analysis.The detailed results of the robustness tests are presented in tables S5-S7.

Discussion
To the best of our knowledge, this is the first study to determine the direct causal effect of long-term PM 2.5 exposure on MetS occurrence in a large populationbased cohort in China.In addition, a significant modifying role of dietary habits was found.Poor dietary habits may exacerbate the influence of PM 2.5 on ( MetS risk.Male may be at greater risk of PM 2.5 -related MetS. The mean age of the population in our study was middle-aged.22.38% of the participants were considered to have MetS.The prevelance of MetS in our study was similar to that reported in the two recent studies (20.7% and 22% in the European regions) [6,38] but lower than that reported in the population studied by Voss et al [39].However, the mean PM 2.5 level was much higher than in developed countries with a value of 65.79 µg m −3 (compared with 11.8 µg m −3 and 18.4 µg m −3 in Germany) [39].In addition, we found that a one-unit increase in PM 2.5 concentration could increase the risk of MetS in adults by 1.3%.
Reviewing the literature to date, the association between PM components and the risk of MetS has been preliminarily investigated in several clinical observational and experimental studies [7,12,38].However, these findings have been inconsistent.For example, Feng et al reported a positive association between MetS occurrence and PM 2.5 exposure (OR = 1.38, 95%CI: 1.23-1.55)based on a population-based observational study with 72 248 participants [12].In contrast, another cross-section study [13]found no statistically significant association between PM 2.5 exposure and prevalent MetS (OR = 1.03, 95% CI: 0.99-1.07).The pooled analysis also failed to confirm the significant positive association between these two events (OR = 1.34, 95%CI: 0.96-1.89) in another meta-analysis of nine studies [15].In our study, the classic probit model results found that long-term exposure to PM 2.5 would not increase the risk of MetS in adults but would protect them from this condition.However, the opposite results were found in the CF-probit model.This divergence revealed the existence of endogeneity to a considerable extent, and reasons for this result might include omitted variables and measurement error, thus making the results obtained from the probit model less credible in terms of evidence.Thereby, by using the causal model, this study noticed a significant positive causal effect of long-term PM 2.5 exposure on MetS risk, providing more prospective evidence in supporting the reduction of ambient PM 2.5 pollution in preventing the MetS occurrence.
Meanwhile, although the underlying mechanisms between PM 2.5 exposure and the risk of MetS had not been fully documented, some potential pathways could help to strengthen our findings.Of note, systemic inflammation was regarded as one of the pivotal factors in increasing the risk of the MetS, with a series of dysfunctions of blood pressure and lipid metabolism as well as insulin resistance [40][41][42].Thereby, the elevated systemic inflammation and oxidative stress in PM 2.5 exposure would trigger metabolic dysfunction and then increase the risk of MetS [43].Additionally, modification of DNA methylation at the promoter region caused by PM 2.5 pollution would remarkably lead to modifications of gene expression, which could increase the risk of multiple diseases like systemic metabolic disorders and even malignancies [44,45].In addition, our population-based evidence suggests that male may be more susceptible to PM 2.5 -induced MetS.Previous evidence has shown that there are large sex differences in the adverse effects of air pollution, but the underlying mechanisms of PM 2.5 -related MetS are still unclear [46].Laboratory evidence suggests that hormonal differences may contribute to this difference [46,47].Thereby, future experiments analysis should pay more attention to potential mechanisms and could help to better understand the role of PM 2.5 pollution on MetS development.
It is noteworthy that a growing body of work has found that dietary patterns can have a significant impact on systemic metabolism and nutritional status, as well as influencing systemic disorders [19,48].In particular, recent studies have observed that an energy-restricted dietary pattern could significantly improve metabolic disturbances.On the other hand, a high-fat diet and other poor dietary habits would increase the risk of MetS in the following life course [21,49].However, this fact was frequently ignored in several previous studies evaluating the association between long-term PM 2.5 exposure and MetS [13,50].Here, we addressed this research gap.We determined the adverse modification of PM 2.5 exposure by poor dietary habits in relation to MetS in the Chinese population.Specifically, we revealed that participants with poorer dietary patterns had a higher risk of MetS compared with adequate groups at the same PM 2.5 exposure levels.Similarly, a recent study from the US found a modifying role in dietary patterns [51].In their report, compared with participants with passive smoking exposure alone, poor dietary patterns combined with passive smoking exposure were associated with a significantly higher risk of MetS.Another study found that dietary patterns rich in animal products significantly modify the association between prenatal exposure to air pollutants and the risk of gestational diabetes mellitus [52].In addition, two recent welldesigned randomized controlled trials showed the feasibility of time-restricted food habits in improving the systemic metabolic quality, especially in terms of MetS-associated biomarkers, in healthy or MetS participants [53,54].
Our study has several strengths.First, it is a national study from a developing country with a wide range of PM 2.5 exposures.In addition, a rigorously tested causal model was used to explore a more reliable relationship between long-term PM 2.5 exposure and Mets risk.Furthermore, we also observed the modifying role of dietary patterns and sex on the risk of the MetS under long-term PM 2.5 exposure.
Admittedly, the current study has some limitations.First, this study has not controlled for the influence of the range of activities and migration of the population, which means that it is assumed that the respondents are largely within the district and county during the study period.Second, the dietary score of each participant was calculated based on an earlier version of the DBI, as the records of water intake were not collected.Thus, future dietary assessment scales with more comprehensive daily dietary information may be more useful for the assessment of the regulatory role of dietary patterns in air pollution-mediated MetS.In addition, CHNS is not a nationally representative survey, and the majority of our subjects were also from rural areas and from southern China.This means that we need to be cautious when generalizing our results to the national level.Finally, some other air pollutants of interest, such as SO 2 , NO 2 , and PM 10 , were not further analyzed in our study.Although we found a causal effect of long-term exposure to PM 2.5 on the risk of MetS, and previous laboratory evidence indirectly supports this association, we still advocate a cautious interpretation of these findings, particularly given that the instrumental variable could potentially have a pathway to health through its impact on other air pollutants, which is generally minor but should not be overlooked.Future work is needed to comprehensively assess the potential mechanisms of PM 2.5 -related MetS, including how PM 2.5 gradually leads to changes in more detailed earlier biomarkers, and to identify more potential practical strategies for environmental-related human health.

Conclusion
In summarize, the current large-scale populationbased study highlights the adverse effects of long-term PM 2.5 exposure on the risk of MetS in the Chinese populations.Additionally, poor dietary patterns may exacerbate this effect.Males may be at a higher risk of PM 2.5 -related MetS.The strong evidence of our findings provides important national policy implications for developing countries to monitor and reduce air pollution as well as the modification of the dietary patterns that protect the susceptible population.

Figure 1 .
Figure 1.Panel A: the study population selection process; Panel B: the distribution of PM2.5 concentrations and participants in the study.Abbreviation: MetS: metabolic syndrome.

Table 1 .
Descriptive statistics for the variables in this study (Mean/SD or Number/%).
(1)e:(1)For the continuous variables, statistics reported are the sample mean with the standard deviation in parentheses.(2) For the binary variables, statistics reported are the sample frequency with the percentage in parentheses.

Table 3 .
IV-Probit CF marginal effect of modification factors.