Dietary Supplements and Mortality Rate in Older Women:  The Iowa Women's Health Study FREE

In the United States, the use of dietary supplements has increased substantially during the past several decades,^1- 3 reaching approximately one-half of adults in 2000, with annual sales of more than $20 billion.^1,3 Sixty-six percent of women participating in the Iowa Women's Health Study² used at least 1 dietary supplement daily in 1986 at an average age of 62 years; in 2004, the proportion increased to 85%. Moreover, 27% of women reported using 4 or more supplemental products in 2004.² At the population level, dietary supplements contributed substantially to the total intake of several nutrients, particularly in elderly individuals.^1- 2

Supplemental nutrient intake clearly is beneficial in deficiency conditions.⁴ However, in well-nourished populations, supplements often are intended to yield benefit by preventing chronic diseases. Results of epidemiologic studies^5- 9 assessing supplement use and total mortality risk have been inconsistent. Several randomized controlled trials (RCTs),^10- 11 concentrating mainly on calcium and vitamins B, C, D, and E, have not shown beneficial effects of dietary supplements on total mortality rate; in contrast, some^12- 13 have suggested the possibility of harm. Meta-analyses^14- 15 concur in finding no decreased risk and potential harm. Supplements are widely used, and further studies regarding their health effects are needed. Also, little is known about the long-term effects of multivitamin use and less commonly used supplements, such as iron and other minerals.

The aim of the present study was to assess the relationship between supplement use and total mortality rate in older women in the Iowa Women's Health Study. Our hypothesis, based on the findings of a previous study by some of us,² was that the use of dietary supplements would not be associated with a reduced rate of total mortality.

The Iowa Women's Health Study¹⁶ was designed to examine associations between several host, dietary, and lifestyle factors and the incidence of cancer in postmenopausal women. At the study baseline in 1986, a total of 41 836 women aged 55 to 69 years completed a 16-page self-administered questionnaire. Of these women, 99.2% were white and 98.6% were postmenopausal. Respondents were slightly younger, had lower body mass index (calculated as weight in kilograms divided by height in meters squared), and were more likely to live in rural areas compared with nonrespondents.¹⁷ The Iowa Women's Health Study was approved by the University of Minnesota Institutional Review Board; return of the questionnaire was considered to indicate informed consent, concordant with prevailing practice in 1986.

We included 38 772 women, excluding from all analyses those who did not adequately complete a questionnaire including food frequency and supplement use at baseline in 1986.² For the analyses starting in 1997, a total of 29 230 women who filled out the supplement use questionnaire (diet data were not assessed) were included. In the 2004 starting analysis, 19 124 women were included. The study flow is shown in the Figure.

Food intake was assessed at baseline and in the 2004 follow-up, using 2 nearly identical versions of the validated 127-food item Harvard Service Food Frequency Questionnaire.^18- 19 Food composition values were obtained from the Harvard University Food Composition Database derived from US Department of Agriculture sources, supplemented with manufacturer information, and updated to reflect marketplace changes.

Supplement use was queried in 1986, 1997, and 2004 and included the 15 supplements assessed at all 3 surveys: multivitamins; vitamins A, beta-carotene, B₆, folic acid, B complex, C, D, and E; and minerals iron, calcium, copper, magnesium, selenium, and zinc. Different forms of vitamin D, cholecalciferol (D₃) or ergocalciferol (D₂), were not distinguished. At the baseline and 2004 follow-up surveys, the supplement-related questions were part of the Food Frequency Questionnaire. In the 1997 follow-up survey, the supplement questions were asked without querying regarding diet. Dose was assessed uniformly across 3 surveys for vitamins A, B₆, C, D, and E and for minerals calcium, iron, selenium, and zinc with 5 supplement-specific response options (no dose information was collected for vitamin B₆ at baseline or for vitamin D in 2004). Although the dietary supplement portion of the Food Frequency Questionnaire used in the study was not validated separately,¹⁹ an evaluation²⁰ with similar instruments has reported validity correlations of approximately 0.8.

Deaths through December 31, 2008, were identified annually through the State Health Registry of Iowa or the National Death Index for participants who did not respond to the follow-up questionnaires or who had emigrated from Iowa. Underlying cause of death was assigned by state vital registries via the International Classification of Diseases (ICD). We defined all cardiovascular disease (CVD) by ICD-9 codes 390-459 or ICD-10 codes I00-I99, cancer by codes 140-239 or C00-D48, and “other cause of death” for all other deaths, excluding 231 of those related to injury, accident, and suicide, because it is unlikely that supplement use would be causally related to these outcomes. Follow-up duration was calculated as the time from the baseline date to the date of death or to the last follow-up contact or December 31, 2008, whichever came first.

The baseline questionnaire included questions concerning potential confounders including age, height, educational level, place of residence (farm, rural area other than a farm, or city), diabetes mellitus, high blood pressure, weight, hormone replacement therapy, physical activity, and smoking status. As previously described,² physical activity was characterized as participating in moderate or vigorous activities less than a few times per month, a few times per month or once per week, or 2 or more times per week. Waist and hip circumferences were measured by each participant using a fixed protocol.²⁰

The 1986 and 2004 questionnaires included the same questions and in a similar form except that educational level, place of residence, and waist and hip circumferences were not reassessed. The only questions that the 1997 questionnaire included in common with the 1986 and 2004 questionnaires were those regarding diabetes mellitus, weight, high blood pressure, hormone replacement therapy, and smoking status. Neither blood lipid levels nor blood pressure were measured in any survey.

Analyses were performed using statistical software (PC-SAS, version 9.2; SAS Institute, Inc, Cary, North Carolina). Continuous variables were compared using analysis of variance and categorical variables using χ² tests. Cumulative mortality rates by supplement use were examined. Absolute risk increase (ARI) and absolute risk reduction (ARR) were calculated by multiplying the absolute risk in the reference group by the multivariable-adjusted hazard ratio (HR) change in the comparison group. Cox proportional hazards regression analyses were used to explore the relationship between supplement use and outcomes. In the minimally adjusted model, we adjusted the association for age and energy intake; in multivariable adjusted model, version 1, we additionally adjusted for educational level, place of residence, diabetes mellitus, high blood pressure, body mass index, waist to hip ratio, hormone replacement therapy, physical activity, and smoking status. For multivariable adjusted model, version 2, we added intake of alcohol, saturated fatty acids, whole grain products, fruits, and vegetables.

Additional analyses were performed for shorter follow-up intervals; in each of the following periods, approximately 15% of deaths occurred: from 1986 until the end of 1996, from 1997 until the end of 2003, and from 2004 until the end of 2008. Data including supplement use from the corresponding interval questionnaire were used whenever available. Covariate adjustment was performed, as described in the previous paragraph. For analyses starting in 1997, current covariate data were available for diabetes mellitus, high blood pressure, body mass index, hormone replacement therapy, and smoking status. For analyses starting in 2004, current data were available for all covariates except educational level, place of residence, and waist to hip ratio. When current data were unavailable, information from 1986 was used.

Among the 38 772 women (mean [SD] age, 61.6 [4.2] years) followed up from the 1986 questionnaire data, 15 594 deaths (40.2%) occurred during the mean follow-up time of 19.0 years. Mean body mass index was 27.0 (5.1); 36.8% of the respondents reported high blood pressure; 6.8%, diabetes mellitus; and 15.1%, current smoking status. At baseline, compared with nonusers, supplement users had a lower prevalence of diabetes mellitus, high blood pressure, and smoking status; a lower BMI and waist to hip ratio; and were less likely to live on a farm. Supplement users had a higher educational level, were more physically active, and were more likely to use estrogen replacement therapy (Table 1). Also, supplement users were more likely to have lower intake of energy, total fat, and monounsaturated fatty acids, saturated fatty acids and to have higher intake of protein, carbohydrates, polyunsaturated fatty acids, alcohol, whole grain products, fruits, and vegetables. Similar patterns were seen in the 2004 questionnaire among 19 124 women (Table 1) and for individual supplements (eg, iron and calcium) (eTable 1).

Self-reported use of dietary supplements increased substantially between 1986 and 2004.² In 1986, 1997, and 2004, 62.7%, 75.1%, and 85.1% of the women, respectively, reported using at least 1 supplement daily. The most commonly used supplements were calcium, multivitamins, vitamin C, and vitamin E (eTable 2); the most common supplement combinations were calcium and multivitamins; calcium, multivitamins, and vitamin C; and calcium and vitamin C.

At baseline, in Cox proportional hazards regression models with full follow-up time and adjusted for age and energy intake, self-reported use of vitamin B complex; vitamins C, D, and E; and calcium had significantly lower risk of total mortality compared with nonuse; copper was associated with higher risk (Table 2). With further adjustment (dose in multivariable adjusted model version 1), only the use of calcium retained a significantly lower risk of mortality (HR, 0.92; ARR, 3.5%); the other inverse associations were attenuated to nonsignificance. In contrast, further adjustment for nonnutritional factors strengthened several associations to significance that had HR higher than 1 in the minimal model: multivitamins (HR, 1.06; ARI, 2.2%), vitamin B₆ (1.09; 3.5%), and iron (1.09; 3.8%). Further adjustment for nutritional factors (version 2) affected the associations further: multivitamins (HR, 1.06; ARI, 2.4%), vitamin B₆ (1.10; 4.1%), folic acid (1.15; 5.9%), calcium (0.91; 3.8%), copper (1.45; 18.0%), iron (1.10; 3.9%), magnesium (1.08; 3.6%), and zinc (1.08; 3.0%).

In sensitivity analyses excluding women who had CVD or diabetes mellitus (n = 5772) or cancer (n = 3523) at baseline, the results were not materially changed. For example, for iron, the multivariable adjusted HR for total mortality was 1.13 (95% CI, 1.05-1.22). Parallel to the situation with total mortality rate, most supplements were unrelated to or showed higher cause-specific mortality rate in multivariable adjusted model version 2, although risk patterns varied across causes (Table 3).

In multivariable adjusted analyses across the shorter follow-up intervals, beginning with the baseline and each follow-up questionnaire (Table 4), the most consistent findings in multivariable adjusted model version 2 were for supplemental iron (HR, 1.20, 1.43, and 1.56; ARI, 2.2%, 5.5%, and 6.6%, respectively) and calcium (0.89, 0.90, and 0.88; ARR, 1.4%, 1.5%, and 1.8%, respectively). Supplemental folic acid tended toward higher risk, significant only in the last interval (HR, 1.28, 1.19, and 1.27; ARI, 3.0%, 2.6%, and 3.4%, respectively).

Dose-response associations could be computed for selected supplements. The inverse association with calcium was lost at its highest dose (Table 5). For supplemental iron, a dose-response relationship was observed in the full follow-up cohort starting in 1986. In the dose-response interval analyses, significantly increased risk was seen at progressively lower doses as the women aged through baseline in 1986, to baseline in 1997, to baseline in 2004. For vitamins A, C, D, and E, as well as minerals selenium and zinc, no dose-response association was found. These dose-response associations persisted after women with a history of CVD, diabetes mellitus, or cancer at baseline were excluded.

For supplemental iron, we also studied the consistency of reported use across surveys and total mortality among 16 841 women who completed all 3 questionnaires. Compared with measures for nonusers, the multivariable adjusted HRs for users were 1.35 (95% CI, 1.20-1.52) for use reported at 1 survey, 1.62 (1.30-2.01) for use reported at 2 surveys, and 1.60 (1.04-2.46) for use reported at all 3 surveys.

In agreement with our hypothesis, most of the supplements studied were not associated with a reduced total mortality rate in older women. In contrast, we found that several commonly used dietary vitamin and mineral supplements, including multivitamins, vitamins B₆, and folic acid, as well as minerals iron, magnesium, zinc, and copper, were associated with a higher risk of total mortality. Of particular concern, supplemental iron was strongly and dose dependently associated with increased total mortality risk. Also, the association was consistent across shorter intervals, strengthened with multiple use reports and with increasing age at reported use. Supplemental calcium was consistently inversely related to total mortality rate; however, no clear dose-response relationship was observed.

Previous studies summarized in a systematic review²¹ provide little support for our findings, suggesting beneficial effects of calcium on total mortality rate; in prospective cohorts and RCTs, vitamin D supplementation, but not calcium, was found to be associated with a nonsignificant reduction in CVD mortality. The pooled HR for the CVD risk in RCTs was 0.90 (95% CI, 0.77-1.05) for vitamin D and 1.14 (0.92-1.41) for calcium. We found no evidence for a benefit of vitamin D against total mortality.

The evidence regarding a possible harmful effect of supplemental iron is limited. Pocobelli et al⁶ found that men in the highest category of average 10-year dose of supplemental iron had a 27% increased risk of total mortality compared with nonusers in age- and sex-adjusted models. The association was, however, attenuated after multivariable adjustment. High iron stores, measured as serum ferritin, have been found to be related to increased risk of CVD in 2 studies^22- 23 but not in another.²⁴ Although we did not evaluate the possible mechanism, iron is suggested²⁵ to catalyze reactions that produce oxidants and thus promote oxidative stress. However, we cannot rule out the possibility that the increase in total mortality rate was caused by illnesses for which use of iron supplements is indicated. Chronic disease, major injury, and/or operations may cause anemia, which is then treated with supplemental iron. However, we could find no evidence for such reverse causality. Iron supplementation was related to future mortality rate even 19 years later in women free of CVD, diabetes mellitus, and cancer; baseline covariates of iron use were not greatly different from those of other supplements; and progressively lower doses were associated with excess risk as the women aged.

Increased blood homocysteine concentrations are considered to be a modifiable risk factor for CVD.²⁶ In RCTs,^14,27 folic acid, vitamin B₆, and vitamin B₁₂ or their combinations have decreased blood homocysteine concentrations but failed to reduce the risk of CVD. In contrast, use of B vitamins has been found to be related to an increased risk of CVD in one study. Ebbing et al¹³ found that the combination of folic acid and B₁₂ supplementation increased the risk of mortality from all causes and from cancer in an RCT setting.

We are not aware of any long-term RCTs studying the effects of daily multivitamins on total mortality rate; epidemiologic studies^5,7- 9 have not provided evidence of benefit. Observational findings regarding the antioxidant supplements selenium, beta-carotene, and vitamins A, C, and E and total mortality have been inconsistent,^5- 6,9 although the use of vitamins C and E has been found to be related with reduced risk of all-cause mortality in 2 studies.^5,9 For supplemental vitamin A and beta-carotene, an observational study⁶ has not provided evidence of benefit for total mortality rate. In RCTs,^10- 11 supplementation with selenium, beta-carotene, or vitamins A, C, or E has not been found to be beneficial relative to total mortality rate in well-nourished populations, and some studies^12- 13 have suggested this practice yields harm.

Strengths of the current study include the large sample size and longitudinal design. Also, the use of dietary supplements was queried 3 times: at baseline in 1986, 1997, and 2004. The use of repeated measures enabled evaluation of the consistency of the findings and decreased the risk that the exposure was misclassified.

Our study also has limitations. An intermediate event, such as CVD or cancer, can induce a change in supplement use and confound the exposure-outcome association. In our data, the use of supplements was not modified by a prebaseline diagnosis of CVD, diabetes mellitus, or cancer. Furthermore, intermediate cancer did not alter the supplement-taking pattern. It is possible that despite extensive adjustment, residual confounding remained. The use of dietary supplements is related to healthier lifestyle,^1- 2 thus leading to apparently inverse associations with total mortality rate. The associations found after adjustment for lifestyle factors are more accurate from a perspective of a causal relationship. However, we cannot exclude the possibility that some supplements were taken for reasonable cause in response to symptoms or clinical disease. We did not have data regarding nutritional status or detailed information of supplements used. Also, the study population consisted only of white women; thus, generalization to other populations, ethnic groups, or men could be questioned. Because our primary hypothesis concerning supplement use and total mortality rate with covariate adjustment included 15 separate tests, a conservative Bonferroni approach would require a P value of .05/15.00 = .003. However, many of the additional statistical tests were confirmatory, strengthening confidence that findings were not explainable by chance.

Among the elderly population, the use of dietary supplements is widespread,^1- 3 and supplements often are used with the intention of attaining health benefits by preventing chronic diseases. Although we cannot rule out benefits of supplements, such as improved quality of life, our study raises a concern regarding their long-term safety. Also, cumulative effects of widespread supplement use, together with food fortification, have raised concern regarding exceeding upper recommended levels and, thus, regarding long-term safety.¹ It is not advisable to make a causal statement of excess risk based on these observational data; however, it is noteworthy that dietary supplements, unlike drugs, do not require rigorous RCT testing, and observational studies are often the best-available method for assessing the safety of long-term use. Based on existing evidence, we see little justification for the general and widespread use of dietary supplements. We recommend that they be used with strong medically based cause, such as symptomatic nutrient deficiency disease.

In conclusion, in this large prospective cohort of older women, we found that most dietary supplements were unrelated to total mortality rate. However, several commonly used dietary vitamin and mineral supplements were associated with increased total mortality rate, most strongly supplemental iron; calcium showed some evidence of lower risk.