Iron Intake and the Risk of Type 2 Diabetes in Women: A prospective cohort study

The NHS cohort was established in 1976, when 121,700 registered nurses aged 30–55 years from 11 U.S. states responded to a mailed questionnaire on risk factors for cancer and cardiovascular disease. This study was approved by the human research committee of the Brigham and Women’s Hospital. Every 2 years, follow-up questionnaires are sent to update information on potential risk factors and health outcomes. In 1980, a 61-item food frequency questionnaire (FFQ) designed to assess dietary intake was added. In 1984, 1986, 1990, 1994, and 1998, an expanded (131-item) FFQ was used. The reproducibility and validity of the dietary data have been reported in detail elsewhere (30–32). Women without FFQ data at baseline or those with unreasonably high (>3,500 kcal/day) or low (<500 kcal/day) intakes and those who had left >10 items blank were excluded from the analyses. In this study, we also excluded participants with a history of cancer (other than nonmelanoma skin cancer), cardiovascular disease (angina, coronary bypass or angioplasty, myocardial infarction, or stroke), and diabetes at baseline. Therefore, we conducted these analyses among 85,031 women.

The semiquantitative FFQs provide information on the average frequency of consumption of selected foods and beverages during the preceding year. The food composition database used to calculate the nutrient values is based primarily on U.S. Department of Agriculture data (33) supplemented with manufacturers’ data. In addition to multivitamin supplements, participants also reported the use of any specific iron supplements including their dose. Total iron intake was calculated as the sum of dietary and all supplemental intakes. All nutrient intakes (except supplemental iron and alcohol) were energy adjusted by the residual method (34). The Pearson correlation coefficient for total iron intake estimated from FFQs and four 1-week diet records (3 months apart) in a validation study among a subset of 150 women was 0.55 (31). In our previous analysis, heme iron and supplemental intake were significant predictors of body iron stores measured by serum ferritin (35).

In addition to total, dietary, and supplemental iron, we estimated the intake of heme iron, which is found only in animal products. We calculated nonheme iron (derived from fortified cereals, plant-based foods, and supplements) as the difference between total and heme iron intake.

In the NHS, information on body weight, smoking, postmenopausal hormone use, and physical activity is updated every 2 years. We have previously evaluated the reproducibility and validity of these self-reported measures including body weight in a subset of the cohort participants with technician-assessed measurements (36). Self-reported weight and the average of two technician measures were highly correlated (Pearson’s correlation coefficient = 0.96). Women provided information regarding family history of diabetes in their first-degree relatives in 1982 and 1988. During the 20-year follow-up, participants reported physical activity (hours per week) in 1980, 1982, 1986, 1988, 1992, 1996, and 1998. We estimated the physical activity level of an individual by using the cumulative average number of hours per week spent in moderate or vigorous activity (brisk walking, vigorous sports, jogging, cycling, and heavy gardening and housework).

If a woman self-reported a diagnosis of type 2 diabetes on any of the 2-year follow-up questionnaires, we mailed a supplementary questionnaire regarding her symptoms, diagnostic testing, and current treatment. Diagnosis was confirmed if at least one of the following criteria was reported: 1) at least one of the classic symptoms (polydipsia, polyuria, polyphagia, weight loss, or coma) in addition to a fasting plasma glucose level of ≥140 mg/dl (7.8 mmol/l) or a random level of ≥200 mg/dl (11.1 mmol/l); 2) at least two elevated plasma glucose concentrations on different occasions (fasting ≥140 mg/dl [7.8 mmol/l], random ≥200 mg/dl [11.1 mmol/l]) or random ≥200 mg/dl (11.1 mmol/l) after at least 2 h of oral glucose tolerance testing in the absence of symptoms; or 3) treatment with oral drugs for hyperglycemia or with insulin. Because most of cases of diabetes were diagnosed before 1997, our diagnostic criteria correspond to those of the National Diabetes Data Group (37). Women with gestational or type 1 diabetes were not included in this study. A study in a subsample of women demonstrated a high validity of the supplemental questionnaire in the confirmation of diabetes diagnosis (38). A random sample of 84 women was selected from among those confirmed to have type 2 diabetes based on a supplemental questionnaire. Medical records were obtained for 62 of these women and a physician blinded to the questionnaire data reviewed the records. We confirmed diagnosis of type 2 diabetes in 61 (98%) of the 62 women.

All statistical analyses were performed with SAS software (SAS Institute, Cary, NC). Participants contributed follow-up time from the date of questionnaire return in 1980 to the date of diagnosis of type 2 diabetes, death, or 1 June 2000, whichever came first. For a more accurate measure of long-term diet, we calculated the cumulative average intake for each nutrient using FFQ data provided until the beginning of each 2-year follow-up interval (39). We stopped updating diet when a woman reported a diagnosis of cardiovascular disease, hypertension, hypercholesterolemia, or cancer during follow-up because diagnosis of these conditions may lead to changes in diet.

We divided participants into five categories (quintiles) according to their cumulative iron intake. Incidence rates for type 2 diabetes were calculated by dividing the number of cases by the person-years of follow-up for each quintile of iron intake. Relative risks (RRs) of type 2 diabetes were calculated as an incidence rate ratio by dividing the rate in each quintile by the rate in the lowest quintile of intake. We used the Cox proportional hazards model (40) to calculate RRs for diabetes adjusted for potential confounders including age, BMI, family history of diabetes (first-degree relative), smoking, alcohol intake, postmenopausal hormone use, use of multivitamin supplements, and physical activity. Further, we adjusted for dietary factors including trans fat, the polyunsaturated-to-saturated fat ratio, cereal fiber, total calories, whole grains, fruits and vegetables, red meat, caffeine, glycemic load, and magnesium. We evaluated the proportional hazard assumption by conducting likelihood ratio tests comparing models with and without interaction terms between exposure categories and follow-up time. None of these tests was statistically significant, i.e., there was no violation of the proportional hazards assumption. Tests of linear trend across quintiles of iron intake were conducted by assigning the median value for each quintile and fitting this continuous variable in the model.

During the 20-year follow-up period (1980–2000; 1,578,982 person-years), we documented 4,599 incident cases of diabetes. Women in the highest quintile of total iron intake at baseline were more physically active and were less likely to smoke or to consume alcohol or to have hypertension or elevated cholesterol compared with women in the lower quintiles (Table 1). High iron intake was also associated with higher intake of cereal fiber, fruits and vegetables, whole grains, magnesium, and multivitamin supplements. For most of these factors, we found an opposite trend with quintiles of heme iron intake at baseline. In addition, higher intake of heme iron was associated with higher intake of fat (total and saturated), red meat, and protein and with lower intake of carbohydrates and glycemic load.

In our study population, intake of total and dietary iron at baseline correlated highly with that of nonheme iron, because most iron in the diet is in this form. As the results for nonheme iron were similar to those for total iron, we did not include them separately in the article.

Both total and dietary iron were inversely associated with the risk of type 2 diabetes in the models adjusted for age, BMI, and nondietary factors (Table 2). However, this association disappeared when we also adjusted for dietary factors. Although this confounding was not attributable to any one covariate in particular, the important factors were glycemic load, caffeine, red meat, and cereal fiber. Excluding women who reported current use of iron supplements (4.2%) did not change the results when we evaluated dietary iron. In addition, we did not find any significant association between supplemental iron and the risk of diabetes.

Heme iron intake was associated with an elevated risk of diabetes (Table 3). The age-adjusted RR between extreme quintiles of cumulative heme iron intake was 1.87 (95% CI 1.69–2.07; P_trend < 0.0001), which was attenuated to 1.24 (1.12–1.37; P_trend < 0.0001) when we added BMI in the model. When we adjusted for other nondietary and dietary factors, this RR was 1.28 (1.14–1.45; P_trend < 0.0001). This association remained statistically significant even after adjustment of red meat intake (1.20 [1.05–1.37]; P_trend = 0.003). When we modeled heme iron as a continuous variable, the multivariate RR for every 1-mg increase in intake was 1.23 (1.13–1.34). In a model in which heme iron was used in seven categories, the RR comparing women who consumed ≥2.25 mg/day and those whose consumed <0.75 mg/day was 1.52 (1.22–1.88). In a separate multivariate model with all forms (heme, dietary nonheme, and supplemental) of cumulative iron intake mutually adjusted, only the RR between extreme quintiles of heme iron was significant (1.16 [1.01–1.33]; P_trend: 0.01). In addition, we conducted analyses for heme iron using different methods of updating the exposure and obtained similar results. When we used the simple update method with 4 years of latency (e.g., diet in 1986 was used to predict the risk between 1990 and 1994), the results were slightly stronger (Table 3). However, using only baseline heme yielded nonsignificant association (RR between extreme quintiles 1.08 [0.97–1.21]; P_trend 0.07), suggesting the advantage of using repeated measures for dietary assessment.

Given that red meat is the main source of heme iron in the diet and its intake is positively associated with the diabetes risk in this cohort of women (41), it may confound the association between heme iron and the risk of diabetes. When we added red meat in the model, the estimates for heme iron remained significant. However, the simultaneous control for red meat may not be appropriate given that heme iron is mainly derived from red meat but suggests that the association for heme iron is not fully explained by red meat intake. To address this issue further, we conducted analysis by subdividing heme iron into red meat heme and heme derived from other sources (fish and poultry). Both of these sources were independently associated with an elevated risk of diabetes in the multivariate models, suggesting that the association of heme iron and diabetes is unlikely to be confounded by the intake of red meat.

Body iron may be associated with obesity, menopausal status, alcohol (42), and intake of phytates, which impair heme iron absorption (43). Therefore, we conducted multivariate analyses within strata of these factors but did not find any significant effect modification by these factors.

In this prospective cohort study, we found no association between total, dietary, supplemental, or nonheme iron intake and the risk of diabetes. However, heme iron intake was positively associated with this risk, and this association was not entirely explained by red meat intake.

The extraordinary capacity of the human body to accumulate iron is seen in hereditary hemochromatosis (44). In this genetic disorder, iron deposition occurs in most tissues including the liver, heart, musculoskeletal system, and pancreas (resulting in secondary diabetes). High body iron stores less extreme than those in hemochromatosis have been linked to insulin resistance (10). Iron is a strong prooxidant that catalyzes the formation of hydroxyl radicals, and the increase in oxidative stress may be associated with the risk of diabetes (45). Some researchers have also suggested that high iron levels impede insulin extraction in the liver, leading to peripheral hyperinsulinemia (46,47). Another potential mechanism is direct iron deposition in the pancreatic β-cells that can impair insulin secretion (48). In addition, the hemochromatosis C282Y and H63D mutations in the HFE gene (chromosome 6) that are associated with high iron stores may also play a role in the development of diabetes (49,50).

Serum ferritin is now proposed to be a component of the metabolic syndrome (49). Interestingly, Moirand et al. (51) described a new non–HLA-linked iron overload syndrome, which suggests a relation between iron excess and metabolic disorders. This condition of insulin resistance–associated hepatic iron overload is characterized by the presence of hyperferritinemia in addition to the entities of metabolic syndrome. In a cross-sectional analysis using National Health and Nutrition Examination Survey (NHANES) data, the indexes of body fat distribution including waist-to-hip ratio, were positively associated with serum ferritin levels in Mexican-American men aged between 20 and 49 years (23). Further, several clinical trials of venesection therapy showed beneficial effects on the signs and symptoms of metabolic syndrome (52–54). In a clinical trial among 28 diabetic participants randomly assigned to undergo blood letting (three times, 500 ml, 2 weeks apart), the phlebotomy group had improved insulin sensitivity at 4 and 12 months compared with the observation group (53).

Although several cross-sectional and case-control studies have linked elevated iron indexes with diabetes (23,25–27), similar data from prospective studies are limited. Salonen et al. (55) first reported an association between serum ferritin levels and incidence of diabetes in a group of Finnish men (RR between extreme quartiles 2.4 [95%CI 1.03–5.50]; P_trend = 0.04). In a recent nested case-control study (22) (698 case patients and 716 control subjects), we reported an elevated risk of diabetes in women with high ferritin levels (RR between quintile 5 vs. 1 2.68 [1.75–4.11]; P_trend < 0.001) and those with a low transferrin receptor–to-ferritin ratio (RR between quintile 1 vs. 5 2.44 [1.61–3.71]; P_trend = 0.01).

Because dietary intake of iron is an important determinant of body iron stores (56), higher intake may be associated with an elevated risk of diabetes. It is useful to evaluate dietary iron as intake can be readily modified. In a sample of 620 healthy postmenopausal women in this cohort, higher intakes of heme and supplemental iron were associated with higher serum ferritin levels (35). However, the specificity of serum ferritin as a marker of body iron store is controversial because it is affected by other factors including chronic inflammation (16,57).

In our study, heme iron was associated with a higher risk of incident diabetes. This form of iron is more bioavailable because of its higher absorption independent of body iron status, unlike the absorption of nonheme iron, which is well regulated (56,58). Therefore, it is probable that a chronically high intake of heme iron can lead to high body iron stores and thus may elevate the risk of diabetes. Some reports have suggested that alcohol affects iron metabolism by some unknown mechanisms that involve the iron-binding proteins (42). Lee et al. (29) found a stronger positive association between heme iron and risk of diabetes among women who consumed alcohol compared with nondrinkers. However, we found no significant effect modification by alcohol.

Supplemental iron has also been associated with higher serum ferritin levels in a subset of these women (35). However, we found no association of supplemental iron and the risk of diabetes. In this cohort, current use of iron supplements was relatively rare (<5%). It is possible that women who used iron supplements may have had a much lower baseline iron status.

Because of the observational nature of our study, we cannot completely exclude the possibility that the association was due to residual confounding by unmeasured factors. However, the large sample size, prospective design with repeated dietary assessments, high rate of follow-up, and information on a multitude of covariates minimize major sources of such bias. Because BMI is the strongest predictor of diabetes, we controlled for it in 10 categories. Modeling BMI as a continuous variable did not alter our results. Another potential concern is the measurement error associated with dietary assessment; however, this error is most likely to bias our results toward the null. In addition, by using repeated dietary measurements with a validated FFQ to account for dietary changes, we reduced the extent of this error (39).

In summary, we found no association between total, dietary, supplemental, and nonheme iron and the risk of diabetes. However, heme iron intake was positively associated with the risk of diabetes independent of BMI and other risk factors for diabetes.

This study was supported by National Institutes of Health Grants DK58845 and CA87969. F.B.H. is partly supported by an American Heart Association Established Investigator Award.