Magnesium dietary intake and physical activity in Type 2 diabetes by gender in White, African‐American and Mexican American: NHANES 2011‐2014

Abstract Aims To analyse the causal relationships of nutrition intake and physical activity on haemoglobin A1c (HbA1C) in patients diagnosed with type 2 diabetes mellitus (T2DM) stratified by gender and ethnicity. Materials and Methods An historical cohort of patients with diagnosed T2DM (n = 2831) was extracted from the National Health and Nutrition Examination Survey (NHANES) 2011‐2014 public database, including but not limited to, measurements of physical activity, nutrition, body mass index (BMI) and HbA1c. Multivariate analyses and path analyses were employed to estimate the regression coefficients and path coefficients (ρ) of causal path models of physical activity and nutrition intake on HbA1c stratified by gender and three ethnicity groups (ie non‐Hispanic white, non‐Hispanic black and Mexican American). Results A significant causal path from increased physical activity to increased magnesium (Mg) intake to decreased HbA1c was found. In addition, increased physical activity significantly decreased BMI, which further decreased HbA1c. These results varied by gender and ethnicity but were directionally consistent. Physical activity decreased HbA1c through BMI for males and through Mg intake for females. Mexican American decreased HbA1c through Mg intake, while non‐Hispanic black had an increased HbA1c due to its ethnicity and through increased BMI. Conclusions The beneficial effects of physical activity on decreased HbA1c were mediated through the increased Mg intake and decreased BMI. This aligned with recent investigations of the inverse causal association of Mg intake with insulin resistance and with decreased inflammation.


| INTRODUC TI ON
Type 2 diabetes mellitus (T2DM) is a major public health problem worldwide and is one of the major causes of mortality in the United States (US) outstripping cancer, HIV/AIDS, and cardiovascular diseases. In addition, negative economic and psychosocial outcomes are associated with T2DM.1-3 The complications from diabetes affect several organ systems and increase the risk of premature death. 4 T2DM also results in an increased economic burden on the individual and the healthcare system. 3 Globally, it is estimated that 422 million adults had T2DM in 2014, which was 12% of the adult population. 5 In the USA, diabetes prevalence rose sharply over the past three decades. Diagnosed T2DM accounted for 3.3% of the US population in 1995, rose to 5.6% in 2005 and was 12.1% in 2015, not including undiagnosed diabetes. 6 Variation in prevalence of diabetes exists between ethnic groups due to hereditary and environmental factors. The prevalence of T2DM is higher in non-Hispanic blacks (19.1%) and Mexican Americans (14.5%) compared with non-Hispanic whites (10.1%). 1,6 Of the two types of diabetes, T2DM comprises approximately 90% to 95%, and Type-1 diabetes ranges from 5% to 10%. 7 Type 1 diabetes has a different aetiology from T2DM and is characterized by the total inability of the pancreas to produce insulin, while type 2 diabetes is caused by a combination of genetic and environmental factors related to impaired insulin secretion, insulin resistance, obesity, overeating, lack of exercise, stress, and ageing. 8 Management of T2DM includes modified nutritional intake, physical activity, and medication as the three key counteracting strategies. 9 Associations of these factors with haemoglobin A1c (HbA1C) in T2DM have been investigated using regression models [9][10][11][12][13] and indicated that (a) lifestyle interventions (diet and physical exercise) reduce incidence of T2DM, (b) physical exercise can manipulate blood magnesium levels, particularly in patients with T2DM,and (c) obesity is one of the most important modifiable risk factors for the prevention of type 2 diabetes. The causal relationships of nutrition intake and physical activity on HbA1C in patients diagnosed with T2DM have not been studied using the causal path analysis. [14][15][16] Path analysis is an extension of multiple regression and a special case of causal structural equation models. Path analysis and causal inference procedures are based on ideas developed originally in biology and economics 14,15 for cross-sectional data, where parent-offspring regressions were used at a single point in time. Path analysis uses a system of structural equations and allows analysis of the relationship between dependent variables as well as between independent variables and dependent variables in a complex system analysis. 17 Path analysis is very powerful for examining complex models and for studying the causal relationship and mediation effects. Although there are several randomized clinical trials (RCTs) showing that increased physical activity and nutrition (eg Mg supplement) can improve HbA1c in T2D patients 18,19 , these studies are specific for certain populations. The current study is based on NHANES data, and the results could be applicable to the USA general population. In the present investigation, we apply path [14][15][16] to examine these causal relationships using the National Health and Nutrition Examination Survey (NHANES) 2011-2014 dataset. The current study's goal is to provide insights into possible mechanisms of how dietary intake of specific nutrients and physical activity affect T2DM management.

| Data and materials
The NHANES datasets are a publicly available national survey con- individual. Subjects 20 to 79 years old were included in the present study. We used two dummy variables for three ethnic groups: non-Hispanic black, Mexican American and non-Hispanic white.
Listwise deletion of cases with missing values was employed. All the subjects included had a complete record for demographic characteristics (ie age, gender, race and ethnicity), physical activity, BMI, HbA1c and all 65 nutritional intake measures. We only included the subjects who were taking oral hypoglycemic agents (ie biguanides, sulphonylureas and thiazolidinedione-TZDs). Insulin-dependent subjects were excluded. Thus, we obtained a cohort of patients with diagnosed T2DM who were taking only oral hypoglycemic medications. The flowchart of the study cohort was presented in Figure S1 in the online supplementary material.

| Variables used in the analyses
The distribution of the diet and physical activity in the US adults with T2DM based on NHANES III has been reported 13, but NHANES dietary data have not been used in a causal path model to analyse the effect of physical activity and nutrition intake on HbA1c. In the present study, path analysis [14][15][16] was used to analyse a system of causal structural models to predict HbA1c using the data on demographics, dietary intake, BMI and physical activity. Causal analysis of nutrition intake, physical activity and BMI on T2DM, across three major US ethnic groups (ie non-Hispanic white, non-Hispanic black and Mexican American) may provide insight into medical management of T2DM.
The variable 'physical activity' was defined using the NHANES physical activity questionnaire data on weekly work activities and recreational activities. Physical activity in the analysis was a dummy variable where '1' indicates subjects who had either TA B L E 1 Descriptive statistics for the variables included in the path analyses for the entire study cohort, stratified by gender and ethnicity  SES is treated as a numerical variable taking scores from 1 to 3 with a higher score indicating a higher SES. Mean and standard deviation (SD) are reported for each variable (see Table 1).

| Statistical analysis
Path analysis [14][15][16] is used to estimate the causal relationships between demographic variables, physical activity and dietary information to the outcome variable, HbA1c. HbA1c indicates T2DM control, with well-controlled T2DM (ie < 6.5%) or poorly controlled diabetes (8.0% or over). We hypothesize that demographic characteristics (X 1 ) and physical activity (X 2 ) affect nutrition intake (N 1 and N 2 ), which in turn affects BMI and HbA1c. Physical activity and nutrition intake also directly affect HbA1c Figure  Single-headed arrows indicate causal paths. [14][15][16] Path analysis has not been used previously to analyse the relationship of physical activity and nutrition to HbA1c. Path analyses use causal structural equation models [14][15][16] to fit a multivariate nonexperimental dataset to a complex causal model. The path coefficient (ρ) is determined by fitting the causal structural equations and the significance level is adjusted using Bonferroni corrections. 16 Path analysis is a powerful technique to analyse and compare different complex hypothetical causal models, and to find the model with the most consistent fit to the data. 15,16 Thus, the causal importance of the variables in the pathways to the outcomes can be quantita- was considered significant if the absolute value of the product of the single path coefficients is greater than 0.01 using the path product property. 15,16 Path analysis was performed for the entire population first, and then in each homogeneous group (gender-specific and gender-ethnicity-specific). Path analyses were performed using R statistical software with package 'lavaan'. 21

| RE SULTS
The descriptive statistics of variables included in the path analysis are reported in The path diagram Figure 2 showed the significant paths, and Table 2 showed the estimated significant path coefficients. Physical  Table 2). To further investigate the difference in the causal structure between male and female, separate path analyses were performed for males and females (see Figure 3 and Table S1 for males, and Figure 4 and Table S2   Mexican Americans (see Figure 5 and Table S3 in the supplement).
Estimated path coefficients for the non-Hispanic black males (see Table S4 in the supplement) showed that nutritional variables were not significant for HbA1c. For non-Hispanic black females, physical activity had a significant protective effect on HbA1c (ρ = −0.073) through Mg intake. Similarly, education also had reduced HbA1c (ρ = −0.052) through Mg intake. Age had a strong direct effect on increasing HbA1c (ρ = 0.340) for non-Hispanic black females (see Figure 6).  Figure 7). For non-Hispanic white females, only age had a significant direct effect of ρ = 0.273 on HbA1c (see Figure 8). For white males, education had a direct effect of ρ = −0.115 on HbA1c, while education was not significant for white females. However, physical activity lowered BMI and resulted in an indirect effect of ρ = −0.081 on HbA1c for white females (see Figure 8 and Table S5 in the Supplement). Physical activity had no significant effect on HbA1c for white males (see Figure 7 and Table S5).  all the causal effects on HbA1c from independent variables through BMI were through obesity (see Figure S2 in the Supplement). These findings agreed with previously published results 11 that the prevalence of diabetes increased as the severity of obesity increased.

| D ISCUSS I ON AND CON CLUS I ON
Our study found higher BMI was associated with increased HbA1c.
The indirect effect from physical activity through obesity but not overweight to HbA1c is also supported by the finding that physical activity lowered HbA1c in subjects with high BMI. Literature 22 suggests that adjusting for total energy intake is important. The use of BMI as an intermediary variable between nutrients and HbA1c in the present model apparently was a proxy for total energy intake. In this analytical approach, total energy expenditure or its derivatives were not significant (P > .05) and dropped from the subsequent path analysis.
In the present study, the effect of physical activity on HbA1c varied across different homogeneous groups (gender, ethnicity).
The total effect of physical activity on HbA1c was through several indirect paths through dietary intake. Males had higher path coefficients for intake of all nutritional components than females, especially for protein and total fat. However, females had a higher BMI (ρ = −0.248, Table 2) causing a higher HbA1c, compared with males. Differences in body composition between females and males may explain this difference (ie a higher per cent body fat in females than males). Non-Hispanic blacks had a higher HbA1c compared to non-Hispanic whites (ρ = 0.347, Table 2). However, in this study, Mexican Americans had a lower HbA1c than non-Hispanic white (ρ = −0.065) mediated through Mg intake.
Our study revealed the importance of Mg intake and physical activity in decreasing BMI and HbA1c. Physical activity increased Mg intake (ρ = 0.205, Table 2). Higher Mg intake lowered HbA1c significantly (ρ = −0.161, Table 2), particularly in females (ρ = −0.478, Table S2 in the Supplement). Increased Mg intake caused lower BMI (ρ = −0.161, Table 2) and lower BMI caused lower HbA1c (ρ = 0.214, Table 2). The path coefficient from Mg intake to HbA1c was large (ρ = −0.127) and highly significant (P < .0001), indicating that the relationship of Mg intake and HbA1c is a robust and true observation, not by chance. The mean and standard deviation for Mg levels across different variables in Table 3 provided additional evidence for the strong association between Mg intake and BMI, ethnicity, gender and physical activity in control of HbA1c level. Therefore, the present study indicated the relationships between physical activity and Mg intake were statistically significant and biologically plausible predictors of HbA1c across all ethnic and gender subgroups. The causal association between physical activity, Mg intake and HbA1c control was supported by prior research. 12,23,24 Dietary intake of Mg was higher among those who were more physically active, and both factors were associated with lower HbA1c. 25,26 Meta-analyses of more than 20 investigations showed that increased Mg intake was associated with higher physical activity and lower HbA1c. [27][28][29] Higher physical activity and Mg intake were associated with better glycemic control and lower BMI. 30,31 These published findings aligned with the causal models developed in the present investigation in which physical activity was associated with higher Mg intake, lower BMI and decreased HbA1c. Mg intake was also associated with lower levels of inflammation markers in diabetics 32, and with reduced comorbidities in diabetics such as myocardial infarction. 33 Ultimately, one benefit on increased Mg intake was de- The mechanisms that underly the apparent benefit of Mg in type 2 diabetes involve several pathways (see Figure 9). Lower Mg intake is associated with poorer beta-cell insulin secretion, which may be partly compensated through Mg supplementation. 35,36 Hypomagnesemia causes impaired carbohydrate and other nutrient metabolism. 37,38 Inflammatory markers synergistically impaired insulin signalling, contributing to insulin resistance (see double-headed arrows in Figure 9). 39,40 Mg plays an important role in T2DM glucose control. The patients with T2DM should take the foods with good sources of Mg, which include leafy vegetables and other vegetables (peas, broccoli, cabbage, green beans, artichokes, asparagus, Brussel sprouts), cereals, whole grains, nuts, seeds, legumes, seafood, dark chocolate, tofu and bananas (see Figure 9). data are based on self-report of food and liquid intake. From these data, nutrient intake is estimated after adjusting for day-to-day variation. A further limitation is that the quantity of food intake must also be estimated, further confounding the data. Therefore, the actual measurement of dietary intake did not occur, and it was self-reported intake and quantity using cups, spoon and rulers provided by NHANES.
The limitations of path analysis include (a) relations among variables in the model are assumed to be linear, additive and causal; b) there is one-way causal flow, that is, reciprocal causation between variables is ruled out; and c) the variables are measured on an numerical scale and are measured without error. Regardless of these F I G U R E 9 Relationship between magnesium intake and T2DM (modified from 35,40 ) limitations, the findings here for the effect of Mg intake and physical activity on control of glucose in T2DM are robust.

E TH I C S S TATEM ENT
NHANES data are collected by the US Public Health Service for the purpose of surveillance research. The NHANES datasets are completely anonymized and contain no protected health information (PHI). The IRB classification of this data is Exempt and is in accordance with the Declaration of Helsinki.

CO N FLI C T O F I NTE R E S T
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.