All-cause cirrhosis and cancer of the liver are two of the four top leading causes of death from gastrointestinal and liver diseases in the USA( Reference Peery, Crockett and Barritt 1 ). Globally, the mortality rate from cirrhosis and cirrhosis-related diseases has increased over the past 35 years( 2 ). A large portion of liver cirrhosis and cancer is caused by non-alcoholic fatty liver disease (NAFLD) and alcoholic liver disease( Reference Blachier, Leleu and Peck-Radosavljevic 3 ). NAFLD is the most common liver disease in the world( Reference Bedogni, Miglioli and Masutti 4 – Reference Lazo, Hernaez and Eberhardt 6 ) and includes a spectrum of liver injury ranging from steatosis to severe steatohepatitis that can progress to fibrosis, cirrhosis, liver failure or even liver cancer( Reference Matteoni, Younossi and Gramlich 7 ). Unlike alcoholic liver disease which is caused by chronic heavy alcohol use, the aetiology of NAFLD is not clear but may include obesity, type 2 diabetes, use of drugs and exposure to toxic substances( Reference Godos, Federico and Dallio 8 – Reference Fierbinteanu-Braticevici, Sinescu and Moldoveanu 10 ). NAFLD is considered a feature of metabolic syndrome( Reference Marchesini, Brizi and Bianchi 11 ).
Mg may be a factor that is related to the aetiology of both alcoholic liver disease and NAFLD. People who chronically drink heavy amounts of alcohol are at high risk of Mg deficiency( Reference Rivlin 12 ) and prolonged exposure to alcohol leads to a substantial reduction in Mg homeostasis in the liver( Reference Young, Cefaratti and Romani 13 ). Furthermore, as many as 50 % of type 2 diabetic patients have hypomagnesaemia( Reference Gommers, Hoenderop and Bindels 14 ). As such, one previous study found that serum Mg levels were significantly lower in patients with either alcoholic or non-alcoholic liver steatosis( Reference Turecky, Kupcova and Szantova 15 ). A meta-analysis of randomized trials indicated that Mg supplementation improves insulin resistance in patients with type 2 diabetes( Reference Song, He and Levitan 16 ). Mg intake has also been linked to a reduced risk of metabolic syndrome( Reference Champagne 17 ) and type 2 diabetes( Reference Song, Manson and Buring 18 , Reference Dong, Xun and He 19 ). Very recently, we reported that high intake of Mg may be associated with a reduced risk of mortality due to liver disease, particularly among alcohol drinkers and those with hepatic steatosis( Reference Wu, Zhu and Fan 20 ).
In the present study, we examined whether the intake of Mg was associated with the prevalence of fatty liver disease or prediabetes. Although previous studies have examined the association between intake of Ca and type 2 diabetes( Reference Pittas, Lau and Hu 21 ), few studies have examined the role of Ca intake in association with prediabetes( Reference Drouillet, Balkau and Charles 22 ). Our recent studies indicated that Ca intake may interact with Mg intake in relation to diseases of the gastrointestinal tract, such as colorectal adenoma( Reference Dai, Shrubsole and Ness 23 ), adenoma recurrence( Reference Dai, Sandler and Barry 24 ), reflux oesophagitis, Barrett’s oesophagus( Reference Dai, Cantwell and Murray 25 ) and other chronic disease( Reference Dai, Shu and Deng 26 ). Thus, we hypothesized that intake of Ca may also be related to risk of fatty liver disease and prediabetes.
To test these novel hypotheses and to examine whether these associations differ by alcohol drinking status, we analysed data from the Third National Health and Nutrition Examination Survey (NHANES III) follow-up cohort.
Methods
Study population
NHANES III was conducted in the USA from 1988 through 1994 by the National Center for Health Statistics of the Centers for Disease Control and Prevention. The institutional review board of the Centers for Disease Control and Prevention approved the investigation, and all participants provided informed consent. It was conducted in two phases, each of which comprised a national probability sample. In total, 39 695 participants were selected from a complex, multistage, stratified, clustered probability sample representative of the civilian, non-institutionalized population. Of these participants, 33 994 (86 %) were interviewed in their homes. All interviewed participants were invited to the mobile examination centre for a medical examination. In our study, we excluded 15 169 participants younger than 20 years old. We also excluded 2210 participants with (i) a physician’s diagnosis of diabetes, (ii) glycolated Hb (HbA1c) ≥6·5 % or (iii) fasting glucose ≥126 mg/dl. Furthermore, 2178 participants did not complete dietary or supplemental intake assessment and 948 participants had no information on fatty liver disease or prediabetes, thus they were not included in the study. As a result, 13 489 participants were included in the final analyses.
Classification of key health outcomes
1. Fatty liver disease cases: An ultrasound examination was performed using a Toshiba Sonolayer SSA-90A and Toshiba video recorder among participants aged 20–74 years in NHANES III between 1988 and 1994. In 2009–2010, archived gallbladder ultrasound video images were reviewed to assess the presence of fat within the hepatic parenchyma using standard criteria. Followed by five criteria (liver to kidney contrast; brightness of the liver parenchyma; deep beam attenuation; echogenic walls in the small intrahepatic vessels; definition of the gallbladder walls), hepatic steatosis was categorized as normal, mild, moderate or severe. To avoid potential overlap between mild and moderate hepatic steatosis, a categorization of fatty liver disease as ‘yes’ or ‘no’ was generally used. ‘Yes’ indicated moderate or severe hepatic steatosis, while ‘no’ indicated the liver was normal or had mild hepatic steatosis( 27 ).
2. Prediabetes cases: According to the American Diabetes Association, prediabetes meets the following criteria: (i) no diagnosis of diabetes from a doctor; and (ii) fasting plasma glucose level between 100 and 125 mg/dl or HbA1c between 5·7 and 6·4 %( 28 ).
3. Both prediabetes and fatty liver disease cases: Those who had both conditions defined above.
4. Controls: Those who had neither fatty liver disease nor prediabetes were regarded as controls.
Nutrient intake assessments
Detailed dietary and supplemental intakes including Mg and Ca were derived from a single 24 h dietary recall and a 30 d supplemental interview which participants completed at the mobile examination centre. For the current analyses, only dietary recall data determined by NHANES to be ‘reliable (i.e. the individual food files contain records only for participants with complete intake records that were considered to be reliable)’ were used. The total intake of these nutrients was calculated by summing up the Mg and Ca intakes from the dietary and supplemental intakes.
Covariates
We considered a number of factors as potential confounding factors, including: age (years), sex (men and women), race and ethnicity (non-Hispanic Whites; non-Hispanic Blacks; Other), educational attainment (lower than high-school education; high-school diploma; college graduate or above), ratio of poverty to income (≤1; 1–3; >3), cigarette smoking status (never; former; current), alcohol drinking status (never alcohol drinker: had less than twelve drinks of any kind of alcoholic beverage in entire life; former alcohol drinker: had more than twelve drinks of any kind of alcoholic beverage in entire life but in the past 12 months had less than twelve drinks of any alcoholic beverage; current alcohol drinker: had more than twelve drinks of any kind of alcoholic beverage in entire life and in the past 12 months had at least twelve drinks alcohol), physical activity status (yes: ‘have done one or more activities in the past month: jogging/running, swimming, riding a bicycle, aerobics activity, garden/yard work or other activity’; no: ‘no activity was done in the past month’), BMI (kg/m2), waist-to-hip ratio, daily intakes of total energy (kcal/d), Ca (mg/d) and Mg (mg/d), use of Ca supplements (yes; no) and use of Mg supplements (yes; no).
Statistical analysis
All analyses were performed using the Survey package in the SAS statistical software package version 9.4 to account for the applicable weighting in the multistage clustered, probability sampling design in the NHANES III cohort. Covariates were compared between cases and controls to evaluate potential confounding factors using the Rao–Scott χ 2 test for categorical data and Survey regression models for continuous variables. Survey logistic regression models with fatty liver disease or prediabetes or both prediabetes and fatty liver disease as the dependent variable were used to analyse the association of Ca or Mg intake with fatty liver disease or prediabetes or both prediabetes and fatty liver disease adjusting for potential confounders. Total Ca or Mg intake was included in the models as a categorical variable, using quartiles based on the controls’ distribution. To assess the linear trend in the odds of Ca or Mg quartile, the median value for each quartile was entered in the logistic regression model as an ordinal variable. Stratified analyses by sex (men or women), ratio of Ca intake to Mg intake (Ca:Mg <2·6 or ≥2·6; to be consistent with our previous studies, the cut-off point of 2·6 was used( Reference Dai, Sandler and Barry 24 )), daily Ca intake (<1200 mg or ≥1200 mg) and alcohol drinking status (never, former or current alcohol drinker) were conducted. All reported P values were two-sided with statistical significance evaluated at 0·05.
Results
We compared demographic characteristics and potential confounding factors of cases with fatty liver disease, cases with prediabetes and cases with both prediabetes and fatty liver disease to normal controls (Table 1). Compared with controls, cases were older and were more likely to be men, former smokers, former alcohol drinkers, non-Hispanic Black and to have lower educational attainment, higher poverty and higher BMI. Cases with prediabetes consumed lower amounts of Ca and Mg compared with controls.
Table 1 Baseline demographic characteristics and selected risk factors by disease status: US adults aged ≥20 years (n 13 489), Third National Health and Nutrition Examination Survey (NHANES III), 1988–1994Footnote *,Footnote †
* Values are presented as weighted mean and standard deviation for continuous variables; or as unweighted frequency and weighted percentage for categorical variables.
† P values calculated using survey regression model for continuous variables or the Rao–Scott χ 2 test for categorical variables; significant P values are indicated in bold font.
‡ P value for the comparison between fatty liver disease cases and controls.
§ P value for the comparison between prediabetes cases and controls.
║ P value for the comparison between prediabetes & fatty liver disease cases and controls.
After adjusting for potential confounders, we found that intake of Ca was not related to the odds of fatty liver disease, prediabetes or both prediabetes and fatty liver disease. On the other hand, we found that intake of Mg was associated with approximately 30 % reduced odds of fatty liver disease (P for trend=0·05) and prediabetes (P for trend=0·02). The association pattern was similar between intake of Mg and risk of both prediabetes and fatty liver disease although not statistically significant (Table 2).
Table 2 The association of intakes of calcium and magnesium with fatty liver disease, prediabetes and both prediabetes and fatty liver disease, among all subjects: US adults aged ≥20 years (n 13 489), Third National Health and Nutrition Examination Survey (NHANES III), 1988–1994Footnote *
Ref., reference category.
Significant P values are indicated in bold font.
* Survey logistic regression models were used after adjustment for age, sex, race, educational attainment, household income, smoking status, alcohol drinking, physical activity, BMI, daily intakes of total energy, Mg or Ca, supplemental Ca intake (yes or no) and supplemental Mg intake (yes or no).
In stratified analyses, we found that the intake of Ca was marginally associated with increased odds of fatty liver disease among women (Table 3). Also, the intake of Ca may be related to increased odds of prediabetes among those with Ca:Mg ≥2·6, with an OR of 1·98 (95 % CI 1·07, 3·67) for the highest quartile intake v. the lowest. In the stratified analysis, no significant association was found by drinking status (P for trend >0·05). None of the interactions were statistically significant.
Table 3 The association of intake of calcium with fatty liver disease, prediabetes and both fatty liver disease and prediabetes, stratified by sex, ratio of calcium intake to magnesium intake and drinking status: US adults aged ≥20 years (n 13 489), Third National Health and Nutrition Examination Survey (NHANES III), 1988–1994Footnote *
Ref., reference category.
* Survey logistic regression models were used after adjustment for age, sex, race, educational attainment, household income, smoking status, alcohol drinking, physical activity, BMI, daily intakes of total energy and Mg, supplemental Ca intake (yes or no) and supplemental Mg intake (yes or no).
The inverse association between the intake of Mg and risk of prediabetes was significant only among those with Ca:Mg ≥2·6 (Table 4). However, the P for interaction were not statistically significant (Table 4). On the other hand, there was a significant interaction between Mg intake and Ca intake (P for interaction=0·04) in relation to odds of prediabetes. It appears that the intake of Mg may be related to the reduced odds of fatty liver disease (P for trend=0·04) and prediabetes (P for trend=0·09) only when the intake of Ca is <1200 mg/d. In the stratified analysis according to alcohol drinking status, we found that the intake of Mg may be associated with reduced odds of fatty liver disease only among former drinkers (P for trend=0·04) and current drinkers (P for trend=0·04). However, P for interaction was not statistically significant. We did not find that the association between intake of Mg and risk of fatty liver and prediabetes differed by sex.
Table 4 The association of intake of magnesium with fatty liver disease, prediabetes and both fatty liver disease and prediabetes, stratified by gender, ratio of calcium intake to magnesium intake, intake of calcium and drinking status: US adults aged ≥20 years (n 13 489), Third National Health and Nutrition Examination Survey (NHANES III), 1988–1994Footnote *
Ref., reference category.
Significant P values are indicated in bold font.
* Survey logistic regression models were used after adjustment for age, sex, race, educational attainment, household income, smoking status, alcohol drinking, physical activity, BMI, daily intakes of total energy and Ca, supplemental Ca intake (yes or no) and supplemental Mg intake (yes or no).
Discussion
In the NHANES III cohort, a nationally representative sample of the US general population, we found that the intake of Ca was overall not associated with the odds of fatty liver disease, prediabetes or both. On the other hand, we found that higher intake of Mg was significantly associated with lower odds of fatty liver disease and prediabetes. Due to limited sample sizes in stratified analyses, we consider the stratified analyses as exploratory. In the stratified analysis, the only significant interaction was between Mg intake and Ca intake (P for interaction=0·04) in relation to odds of prediabetes. None of the other interactions were statistically significant. We found that the inverse association between Mg intake and prediabetes appeared primarily in those with a Ca intake <1200 mg/d; and in the same subgroup, intake of Mg was also significantly related to a reduced odds of fatty liver disease. We also found Mg intake was related to reduced odds of fatty liver disease only among former and current alcohol drinkers.
Our finding of an inverse association between the intake of Mg and prediabetes is consistent with that of previous studies which have shown that high Mg intake is associated with a reduced risk of type 2 diabetes( Reference Song, Manson and Buring 18 , Reference Sjögren, Florén and Nilsson 29 , Reference Paolisso, Sgambato and Pizza 30 ), metabolic syndrome( Reference Champagne 17 , Reference He, Song and Belin 31 ), insulin resistance( Reference Pagano, Pacini and Musso 32 ) and prediabetes( Reference Hruby, Meigs and O’Donnell 33 ). Furthermore, we found an inverse association between the intake of Mg and the risk of fatty liver disease. This finding was not consistent with a null association found in a previous study( Reference Da Silva, Arendt and Noureldin 34 ). However, that cross-sectional study conducted in Canadians had a very small sample size, which may have limited the power to detect an association. We also found that the inverse association between the intake of Mg and fatty liver disease appeared primarily in alcohol drinkers. Although novel, this finding is consistent with the observation that heavy alcohol drinkers are at high risk of Mg deficiency( Reference Rivlin 12 ). Future large-scale studies are needed to confirm the findings. This is important because previous studies have found that alcohol causes a substantial reduction in Mg homeostasis in the liver( Reference Young, Cefaratti and Romani 13 ). It was reported by NHANES 1999–2000 that 79 % of US adults do not meet the RDA for Mg( Reference Dai, Shu and Deng 26 ). One study observed that serum concentrations of Mg were significantly reduced in patients with either alcoholic or non-alcoholic liver steatosis( Reference He, Song and Belin 31 ).
In a cohort study, the investigators found that intake of Mg was associated with a reduced risk metabolic syndrome( Reference He, Liu and Daviglus 35 ). Mg intake was also inversely associated with individual components of metabolic syndrome, particularly fasting glucose level, waist circumference and HDL cholesterol. Thus, the inverse association between the intake of Mg and metabolic syndrome is likely mediated through waist circumference or waist-to-hip ratio. Similarly, in our study, after additionally adjusting for waist-to-hip ratio, the significant associations disappeared. As such, waist-to-hip ratio may serve as a pathway to the relationship between Mg intake and the disease outcomes (i.e. fatty liver disease and/or prediabetes) or it could be an over-adjustment.
In the stratified analyses, we found intake of Mg may be more significantly related to reduced odds of prediabetes and fatty liver disease when Ca intake was <1200 mg/d. This finding suggests that the beneficial effect of Mg may be suppressed when Ca intake is higher than the Dietary Reference Intake. This finding is consistent with our recent finding indicating that Ca intake may interact with Mg intake in relation to risks of multiple common diseases( Reference Dai, Shrubsole and Ness 23 – Reference Dai, Shu and Deng 26 ). Some( Reference Hardwick, Jones and Brautbar 36 – Reference Abrams, Grusak and Stuff 38 ), but not all( 39 ), previous human studies indicate that high Ca intake may affect the absorption rate of Mg. It is known that over 80 % of plasma Mg is ultrafiltrated and reabsorbed in the kidneys. Thus, kidney reabsorption plays a key role in regulating Mg homeostasis( Reference Hoenderop and Bindels 40 ). Likewise, 10 g of Ca is filtered daily on average, of which 98 % is reabsorbed in the kidneys( Reference Monnens, Starremans and Bindels 41 , Reference Guéguen and Pointillart 42 ). Thus far, clinical trials have consistently found that high Ca intake leads to significantly increased excretion of Mg in the urine( Reference Hoenderop and Bindels 40 , Reference Green, Booth and Bunning 43 , Reference Nielsen, Milne and Gallagher 44 ). Thus, it is likely that high intake of Ca may lead to relative deficiency of Mg. We also found Mg intake may only be associated with reduced odds of fatty liver disease among former and current drinkers. This finding is possible because alcohol drinkers are at high risk of Mg deficiency( Reference Rivlin 12 ). However, further large studies, particularly longitudinal studies, are necessary to replicate the findings.
A strength of our study is that it is used data from NHANES, a population-based study with a nationally representative sample. However, although multiple 24 h dietary recalls are used as a gold standard measure in nutritional epidemiological studies, a one-time 24 h dietary recall as used in the current study may not have adequately captured long-term dietary intakes of Mg and Ca. Since inter-day variation in intakes of Mg and Ca is random, any residual inter-day variation in the current study would lead to non-differential misclassification, which usually biases the result to the null. Thus, the true association between intakes of Mg and Ca and risk of prediabetes and fatty liver disease may be stronger than what we have observed. We cannot eliminate the possibility that the associations with intakes of Mg and Ca are due to residual confounding factors or healthy lifestyle in general. However, we have adjusted for physical activity and BMI as well as total energy intake. Furthermore, in the same analysis, we found the associations of Ca and Mg are in opposite directions. Thus, it is unlikely our findings are due to confounding by healthy lifestyle in general because those who possess healthy behaviours are likely to use Ca supplements. Finally, like all cross-sectional studies, the temporal sequence for the associations is not clear. However, the inverse association between intake of Mg and risk of prediabetes is consistent with the associations of Mg with insulin resistance( Reference Song, He and Levitan 16 ), metabolic syndrome( Reference Champagne 17 , Reference He, Song and Belin 31 ) and type 2 diabetes( Reference Song, Manson and Buring 18 , Reference Dong, Xun and He 19 ).
Conclusion
In conclusion, our findings suggest that high intake of Mg may be associated with lower odds of having fatty liver disease and prediabetes, whereas high intake of Ca was overall not related to the risk. The associations may appear primarily in those whose Ca intake is less than 1200 mg/d. Further studies, particularly prospective cohort studies, are warranted to confirm or refute these findings.
Acknowledgements
Financial support: X.Z. and Q.D. were supported by the National Cancer Institute, Department of Health and Human Services (Q.D., grant number R01 CA202936). The National Cancer Institute had no role in the design, analysis or writing of this article. Conflict of interest: All authors have no conflicts of interest. Authorship: Conceptualization, J.L. and Q.D.; methodology and formal analysis, X.Z.; writing – original draft, W.L. and Q.D.; writing – review and editing, X.Z., Y.S., L.F., L.W., E.K.K., L.H., M.J.S., J.L. and Q.D.; funding acquisition, Q.D. and X.Z. 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 were approved by the institutional review board of the Centers for Disease Control and Prevention. Written informed consent was obtained from all subjects.
