Magnesium Intake and Risk of Type 2 Diabetes in Men and Women

The characteristics of the Nurses’ Health Study (NHS) and the Health Professionals’ Follow-up Study (HPFS) have been described elsewhere (17,18). Briefly, the NHS was initiated in 1976, when 121,700 female registered nurses, aged 30–55 years, completed a mailed questionnaire on their medical history and lifestyle characteristics. Every 2 years, follow-up questionnaires have been sent to update information on potential risk factors and identify newly diagnosed cases of diabetes and other chronic diseases. The HPFS began in 1986 when 51,529 U.S. health professionals, aged 40–75 years, answered a detailed questionnaire on lifestyle and medical history. Similar to the NHS, this cohort has been followed through biennial questionnaires. In both cohorts, the response rate to the follow-up questionnaires has exceeded 90%.

Diet was first evaluated in 1980 in the NHS and in 1986 in the HPFS. Repeated dietary assessments have been carried out every 2–4 years. From participants who returned the baseline dietary questionnaire, we excluded those who had >10 blanks in food items or did not satisfy our a priori criteria of plausible daily caloric intake. For this analysis, we also excluded participants who at baseline reported history of diabetes, cardiovascular disease, or cancer. These exclusions left 85,060 women followed over 18 years (1980–1998) and 42,872 men followed over 12 years (1986–1998) for the present analysis.

In the NHS, a 61-item semiquantitative food frequency questionnaire (FFQ) was used to collect dietary information in 1980. In 1984, the questionnaire was expanded to 131 items. Similar FFQs were used to update diet in subsequent follow-up in the NHS (1986, 1990, 1994, and 1998) and the HPFS (1986, 1990, 1994, and 1998). In the FFQ, a common unit or portion size for each food was specified and participants were asked how often they had consumed that amount on average during the previous year. The nine responses ranged from “never or less than once per month” to “six or more times per day.” Nutrient intake was computed by multiplying the frequency of consumption of each food by the nutrient content of the specified portions. Composition values for dietary magnesium and other nutrients were obtained from the Harvard University Food Composition Database (22 November 1993), derived from U.S. Department of Agriculture sources (19), and supplemented with manufacturer information. A detailed description of dietary questionnaires and their validity in these cohorts have been published elsewhere (20,21). Correlation coefficients between FFQ and dietary record for magnesium intake were 0.76 in women and 0.66 in men after within-person variation was taken into account.

Use of specific brand and type of multivitamins was ascertained at baseline and updated every 2 years, asking current users about weekly number of multivitamins taken. This information was included in total magnesium intake computation. Questions on separate magnesium supplements were first asked in 1984 in the NHS and in 1986 in the HPFS, with information updated at least every 4 years. Although we did not have information on the exact magnesium content of these supplements, we estimated the content based on the most frequently used magnesium supplements on the market in the year of the questionnaires and used that amount for the calculation of total magnesium intake. In a separate analysis, we examined the association between magnesium supplement use and diabetes risk.

In both cohorts, body weight was self-reported on baseline questionnaires and updated every 2 years. In validation studies, self-reported weights were highly correlated with measured weights (22). In the NHS, to be consistent with the baseline evaluation, we used the cumulative average of hours per week spent in moderate to vigorous activity. In the HPFS, we had detailed information on the hours per week spent in leisure-time physical activities since baseline and through follow-up. We calculated total weekly energy expenditure from physical activity expressed as metabolic equivalents (METs). The validity and reproducibility of the physical activity questionnaires have been previously documented in these cohorts (23,24). Every 2 years, we updated participants’ smoking status (past, current, and number of cigarettes per day if smoking currently). Family history of diabetes (in first-degree relatives) was assessed on multiple occasions in both cohorts. We inquired about physician-diagnosed hypertension and high cholesterol every 2 years; these self-reports were highly accurate compared with medical records in a validation study (25).

On each biennial questionnaire, we asked the participants if and when they had ever been diagnosed with diabetes. To confirm self-reported diagnoses, we mailed a supplementary questionnaire regarding symptoms, diagnostic tests, and therapy. After excluding participants with type 1 and secondary diabetes, the diagnosis of type 2 diabetes was established when at least one of the following criteria was reported in the supplementary questionnaire: 1) at least one classic symptom of type 2 diabetes and elevated plasma glucose (≥140 mg/dl [7.8 mmol/l] fasting or ≥200 mg/dl [11.1 mmol/l] random measure), 2) elevated plasma glucose concentrations on at least two different occasions in the absence of symptoms, or 3) treatment with hypoglycemic therapy (insulin or oral hypoglycemic agents). These criteria accord with those proposed by the National Diabetes Data Group (NDDG). The new guidelines from the American Diabetes Association (ADA) for diagnosing diabetes (fasting plasma glucose ≥126 mg/dl [7.0 mmol/l]) were announced in June 1997 (26) and have been incorporated into the confirmation and documentation of diabetes in subsequent follow-up in both cohorts.

The validity of the method for confirming type 2 diabetes by supplementary questionnaire using the NDDG criteria has been previously documented in these cohorts (27,28). To document the reliability of reports of diabetes in the most recent cycle (1996–1998), an additional validation study was carried out only in the NHS. In this study, we reviewed medical records in two separate groups: women who satisfied NDDG criteria by the supplementary questionnaire and women who satisfied only ADA criteria (fasting plasma glucose between 126 and 139 mg/dl). Medical record review confirmed the diagnosis of diabetes by NDDG criteria in 94 of 95 (98.9%) subjects for the former group. The number of women reporting that they met ADA but not NDDG criteria was small (<5% of cases in this cycle); medical record review confirmed the diagnosis of diabetes by ADA criteria in all but one person, thus confirming its validity using the new criteria.

Person-time of follow-up for each participant was computed from the date of return of the baseline questionnaire (1980 for women and 1986 for men) to either the date of diabetes diagnosis, death, or the end of follow-up (January 1998 for HPFS or July 1998 for NHS), whichever occurred first.

In the primary analysis, participants were divided into quintiles of total magnesium intake (including magnesium from multivitamins), and incidence rates were calculated as the number of events divided by total person-time in each quintile. The relative risks (RRs) were computed as the incidence rates of diabetes in each category of magnesium intake divided by the incidence rate in the lowest quintile of intake (reference group).

To reduce within-person variation and best represent the long-term effects of magnesium intake, we calculated the cumulative average intake of magnesium from all the dietary questionnaires available up to the start of each 2-year period (29). For example, for men, to model diabetes incidence in the 1988–1990 period, we used the 1986 magnesium intake and for the 1990–1992 period, we used the average of 1986 and 1990 intakes. We also conducted a secondary analysis using baseline magnesium intake only.

Cox proportional hazards models stratified by age and time period were used in all multivariate analyses to estimate RRs. To control for multiple confounders, we adjusted for history of hypertension and hypercholesterolemia at baseline and biennially updated information on smoking status, BMI (in eight categories), level of physical activity, family history of diabetes (first-degree relatives), and alcohol intake (four categories). We also adjusted for several dietary variables (30), including glycemic load and intakes of cereal fiber, polyunsaturated fat, trans fat, and processed meat, all in quintiles. Finally, we performed stratified analyses according to levels of BMI, physical activity, and family history of diabetes.

All P values were two sided. Tests for trend were conducted using the median value for each quintile of magnesium intake analyzed as a continuous variable in the regression models. Likelihood ratio χ² was used to assess the significance of the interactions between magnesium intake and the variables used in the stratified models. All analyses were done with SAS version 8.2 (SAS, Cary, NC).

At baseline, compared with those in the lowest quintile of magnesium intake, both women (in 1980) and men (in 1986) with higher intakes of magnesium tended to be leaner, more physically active, and more likely to take multivitamins and magnesium supplements (Table 1). Magnesium intake was positively associated with intakes of fiber and inversely associated with intakes of fat and processed meat. Averaged over the entire follow-up, the median intake (min-max) of magnesium was 290 mg/day (79–1,110 mg/day) in women and 349 mg/day (102–1,593 mg/day) in men.

During a follow-up of 18 years in the NHS (1,456,362 person-years) and 12 years in men (472,730 person-years), we documented 4,085 incident cases of type 2 diabetes in women and 1,333 in men. After adjusting for age and total energy intake (Table 2), we observed a significant inverse association between magnesium intake and risk of type 2 diabetes in both cohorts, with RRs (95% CIs) comparing the top versus bottom quintiles of 0.55 (0.50–0.61) and 0.56 (0.47–0.67) in women and men, respectively. After additional adjustment for BMI, the RRs were somewhat attenuated in both cohorts. However, the RRs were practically unchanged after further adjustment for other nondietary covariates. The RRs remained significant after the addition of dietary variables in the multivariate models. Further adjustment for caffeine slightly attenuated the association between magnesium intake and diabetes risk. The RRs (95% CIs) between extreme quintiles was 0.83 (0.73–0.95) in women and 0.76 (0.61–0.94) in men. Moreover, the adjustment for other minerals, such as calcium, potassium, and phosphorous, did not change the estimate of the association among women (RR comparing extreme quintiles 0.74 [0.63–0.88]), and the inverse association for magnesium was stronger among men (0.62 [0.48–0.81]). Analyses with the single baseline diet assessment instead of updated cumulative average of repeated measurements yielded similar results: 0.79 (0.71–0.88) in women and 0.73 (0.60–0.90) in men. Excluding participants with a history of hypertension or hypercholesterolemia at baseline, using only symptomatic or nonsymptomatic cases as an outcome, or modeling dietary rather than total magnesium intake did not materially change the results. Finally, the inclusion of diuretic use in the final model did not modify our results.

As shown in Table 3, the inverse association was persistent in subgroup analysis according to BMI, physical activity, and family history of diabetes. We did not identify any significant interactions between magnesium intake and these covariates. The inverse association was also similar between drinkers and nondrinkers and between participants with or without hypertension (data not shown).

Finally, we assessed the association between magnesium supplements and risk of type 2 diabetes. The proportion taking magnesium supplements in the entire follow-up period was 3.1% in women and 3.6% in men. There were relatively few cases in the supplement user group (111 in women and 52 in men). We found a significant inverse association in the age-adjusted model only in women (RR 0.82, 95% CI [0.68–0.99] in women and 1.01 [0.76–1.33] in men). However, in the multivariate models, we found no statistical association between use of magnesium supplements and diabetes risk in both women and men: 0.93 (0.77–1.12) and 1.07 (0.81–1.41), respectively. The use of multivitamins was not significantly associated with diabetes risk.

In these two large prospective studies, we observed a consistent inverse association between magnesium intake and risk of type 2 diabetes in men and women. This association was independent of other risk factors for type 2 diabetes, including several dietary factors. Moreover, the inverse association with magnesium intake was consistent across different subgroups defined by the main predictors of type 2 diabetes, such as BMI, physical activity, and family history of diabetes.

The prospective design reduces the possibility of recall and selection bias, and the high rate of follow-up reduces bias due to loss to follow-up. Another advantage is that diet was assessed multiple times during follow-up, which not only reduces measurement error (29), but also takes into account changes in eating behaviors.

Our study has several limitations. Given the size of these cohorts, screening for blood glucose was not feasible, thus some cases of diabetes may have been undiagnosed. However, our validation study showed that undiagnosed diabetes was rare in our cohort because the participants are health professionals (31). It is possible that participants with “unhealthy” diets are more likely to be screened for diabetes. However, the analysis using only symptomatic cases did not substantially change our results, arguing against surveillance bias. On the other hand, the diagnostic criteria for type 2 diabetes were changed in 1997 such that lower plasma glucose levels would now be considered diagnostic. If these criteria were used since baseline, some noncases would have been reclassified as cases. However, this would bias the estimates toward the null.

The inverse association between magnesium intake and diabetes risk was observed in all multivariate models, including the main dietary and nondietary risk factors for diabetes. Moreover, the observed association was consistent within different subgroups, which further supports the idea that confounding by these factors was unlikely to explain our results. However, the effects of residual confounding cannot be completely ruled out in observational studies.

Besides earlier analyses within the NHS and HPFS (12–14), which were consistent with our present results, two other large prospective studies have specifically explored the association between magnesium intake and type 2 diabetes risk. Findings in older women (15) were very similar to our results, with an RR comparing extreme quintiles of 0.76 (95% CI 0.62–0.95) in a multivariate model, including whole grain and cereal fiber. In the other study, Kao et al. (16) found an inverse association between serum magnesium levels and type 2 diabetes, but did not find a significant association between dietary magnesium and subsequent incidence of diabetes. Unlike our study, both of the other studies used only single baseline dietary assessment.

Several experimental studies suggest a protective role of magnesium intake against diabetes. Using a rat model of spontaneous type 2 diabetes, Balon et al. (32) demonstrated a significant reduction in the incidence of diabetes after 7 weeks of feeding with a magnesium-rich diet. In humans, some (11,33,34) but not all (35–37) experimental studies have shown benefits of magnesium supplements on glucose metabolism and/or insulin sensitivity. Some of the inconsistencies among these studies can be explained by differences in treatment periods, doses of magnesium, and parameters used to evaluate the effect. Moreover, most of these studies have been conducted on diabetic subjects, in whom the underlying insulin resistance could interfere with magnesium uptake at the cellular level (38). In one study (11), elderly nondiabetic subjects participated in a double-blind, randomized, crossover study comparing magnesium supplements (4.5 g/day) versus placebo during 4 weeks. This study showed a beneficial effect on insulin response to glucose and insulin action. Whether long-term magnesium supplementation decreases the risk for type 2 diabetes in the general population is unclear, and the hypothesis merits testing in clinical trials. In our observational analysis, magnesium supplement use was not significantly associated with diabetes risk in multivariate models. However, the power of our study was limited by the low prevalence of magnesium supplement use in these cohorts.

Several mechanisms, including insulin secretion, binding, and action, have been proposed to explain the effect of intracellular or plasma magnesium on diabetes pathogenesis (6). Intracellular magnesium is a critical cofactor for several enzymes in carbohydrate metabolism, especially those involved in phosphorylation reactions such as tyrosine-kinase. In animal models (9), hypomagnesemia induced by low magnesium intake triggers severe insulin resistance, which was shown to be partially dependent on deficient tyrosine-kinase activity on the post-receptor pathway of insulin in muscle cells. In healthy humans, a study of short-term low magnesium diet (39) showed that it reduced serum and intracellular magnesium and produced insulin resistance, using a minimal model. Consistent with the effect of magnesium on insulin resistance, Fung et al. (40) found an inverse association between magnesium intake and fasting insulin level, a good marker of insulin resistance, in a cross-sectional sample of the NHS.

Higher magnesium intake is likely more beneficial among individuals with some degree of magnesium deficiency. However, there is no generally accepted test for magnesium status. Also, our subgroup analysis suggests that higher magnesium consumption is likely beneficial for all groups, regardless of their BMI, physical activity levels, and hypertension status.

In conclusion, these two large prospective cohorts provide strong and consistent evidence to support an inverse association between magnesium intake and diabetes risk. The effect of magnesium supplementation in general or high-risk populations requires further research, ideally in randomized clinical trials. This study supports the dietary recommendation to increase consumption of major food sources of magnesium, such as whole grains, nuts, and green leafy vegetables.

This study was supported in part by National Institutes of Health Grants nos. CA55075, HL35464, CA87969, and DK58845 and by National Institute of Diabetes and Digestive and Kidney Diseases training grant no. DK07703. R.L.-R. is also supported by a scholarship from the Consejo Nacional de Ciencia y Tecnología (CONACyT), Mexico.