Several research approaches, particularly observational studies, suggest that diets rich in antioxidant nutrients, particularly vitamins C and E, β-carotene, and selenium, and/or supplements containing one or more of these nutrients are associated with a reduction in the risk of several age-related chronic diseases, including some forms of cancer, cardiovascular, eye, and neurodegenerative diseases, e.g., references [1,2,3,4,5,6,7]. These relatively consistent results stimulated the initiation of several large scale, randomized clinical trials (RCTs), including the Nutrition Intervention Trials (NIT)[8]; Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) study [9]; Gruppo Italiano per lo Studio della Sopravvivenza nell’Infarto miocardico-Prevenzione trial [10]; Heart Outcomes Prevention Study (HOPE) and Heart Outcomes Prevention Evaluation-The Ongoing Outcomes (HOPE-TOO)[11,12]; Supplementation in Vitamins and Mineral Antioxidants (SU.VI.MAX) study [13]; Women’s Health Study (WHS)[14]; Physicians’ Health Study (PHS) and Physicians’ Health Study II (PHS-II)[15,16]; Age-Related Eye Disease Study (AREDS)[17,18]; and Selenium and Vitamin E Cancer Prevention Trial (SELECT)[19]. The overall results of most of these RCTs have been presented as equivocal or null.

These RCTs were mostly designed to test the efficacy of antioxidant supplementation in the primary or secondary prevention of cancer and/or cardiovascular disease. Due largely to issues of cost and statistical power, most of these RCTs have tested only one or two antioxidant nutrients and, thus, may not have fully benefited from the dynamic interrelationships between these nutrient and other dietary and endogenous components of the antioxidant defense network. Importantly, none of these studies employed inclusion criteria such as low antioxidant nutrient intake or status and/or elevated biomarkers of oxidative stress, or determined outcome measures as to whether these parameters were affected by the intervention, elements necessary to testing the hypothesis that the putative benefit derived from these nutrients is due to their antioxidant actions [20,21].

The limitations of RCTs in testing the efficacy and safety of nutrients are not often fully appreciated. The RCT research approach was developed for drugs, where interventions are designed to cure a disease, not produced by their absence; nutrients prevent dysfunction that would result from their inadequate intake. Further, drug effects are generally intended to be acute, large, and with a specific target for action, while nutrient effects are typically chronic, modest, and polyvalent in scope. In addition, drug effects can be tested against a non-exposed (placebo) contrast group, whereas it is impossible and/or unethical to attempt a zero intake group for nutrients [22]. Nonetheless, RCTs are currently considered the “gold standard” for evaluating dietary interventions and, thus, receive the attention of most meta-analyses on this topic.

In addition to using RCTs to test the potential health benefits of antioxidant supplements, meta-analyses may also provide an approach to determining their safety. Although specific adverse events have been reported for most nutrients, as reflected in established Tolerable Upper Intake Levels [23], all-cause or total mortality has sometimes utilized as a global indicator of safety in meta-analyses of RCTs [24,25,26]. In their meta-analysis, Bjelakovic et al. [27,28] concluded that vitamins A and E and β-carotene were associated with an increased risk of all-cause mortality with a relative risk (95% confidence interval) of 1.16 (1.10–1.24), 1.04 (1.01–1.07), and 1.07 (1.02–1.11), respectively. This report was based on 66 RCTs utilizing >1000 participants and excluded any study absent a report of mortality (747 articles) or considered by the authors to be of low methodologic quality (21 articles). The results from the RCTs selected by Bjelakovic et al. [27] may not be readily extrapolated to all other studies with low power due to too few subjects. Thus, we have selected a power-independent approach to reevaluate this meta-analysis with respect to its primary outcome of total mortality as well as examining their impact on the efficacy of the intervention.

We focus here on the primary endpoints of these RCTs. These data were extracted from all but two of the studies used in the Bjelakovic et al. [27] meta-analysis. The report by Chandra [29] (reference no. 39) has been excluded because of documented deficiencies in the integrity of data [30]. The article by Bonelli [31] (reference no. 57) has been excluded because a copy of this publication could not be obtained. The remaining 66 studies were reviewed and classified into three groups with separate analyses conducted for studies with primary prevention, secondary prevention, and therapeutic efficacy as their primary goal. Studies of primary prevention are defined as interventions in generally healthy people intended to reduce the risk of a disease or disability. Secondary prevention actions described here involve dietary interventions (nutrient supplementation) in a diagnosed group to slow the pathogenesis of the disease and/or prevent recurrent events or cause-specific mortality. Studies of therapeutic efficacy are directed to improving the quality of life, limiting complications, and/or providing rehabilitation to a patient group. Importantly, the therapeutic efficacy of dietary supplements is always tested concurrently with established standards of care, often including polypharmacy regimens, as withdrawing these other treatments would be unethical.

All the studies were independently classified by the authors and rated with regard to the primary study goal as: positive or goal achieved (+1), null outcome (0) or adverse effect (−1). Differences in classification or rating between the authors were documented and discussions held to resolve the conflicts until unanimous agreement was reached. The percentage of studies with positive results in the three groups of studies was compared using a χ² test for homogeneity [32].

Assessing the primary goal (i.e., the determination of benefit) of the RCTs analyzed by Bjelakovic et al. [27] places their conclusion about the risk of mortality in a context relevant to the consideration of using antioxidant supplements in health promotion and therapeutic treatments. Clearly, increases in disease-specific or all-cause mortality are never objectives of intervention studies but, when observed, are a reason to stop the protocol early. Nonetheless, it is worth noting that the risk of mortality in any RCTs of nutrients will be substantially dependent upon the nature of the cohort, including parameters such as advanced age, severe disease status, toxicity of drug treatments, etc. Importantly, in our analysis, we did not evaluate as positive, null or adverse the outcomes of the 405 studies excluded from the Bjelakovic et al. [27] study because no death occurred. This is a limitation to our study as it does not provide a full assessment of the potential benefits versus the risk of total mortality from antioxidant supplementation.

With regard to the efficacy of antioxidant supplementation in the RCTs included in the meta-analysis by Bjelakovic et al. [27], we find that the benefit of the intervention was statistically significant principally in those populations generally characterized at risk for micronutrient deficiencies, including those of vitamins C and E, selenium, and beta-carotene as well as other nutrients such as zinc that contribute to the antioxidant defense network. This relationship may suggest that dietary supplementation for the prevention or treatment of chronic diseases is likely to be most effective in those with inadequate intakes, though absent overt deficiency syndromes. Further, this relationship also suggests there is a threshold nutrient status above which additional intake via supplementation might provide no further benefit. However, the threshold for adequate intake of any nutrient is recognized as dependent on an individual’s specific requirements as affected by parameters such as age, sex, health status, and nutrigenomic factors such as polymorphisms. The contribution of polymorphisms to antioxidant defenses can be illustrated by the manganese-dependent superoxide dismutase polymorphism associated with a lower risk of prostate cancer in the presence of high intakes of selenium to optimize glutathione peroxidase activity [91].

Critical to the validity of a meta-analysis is the statistical accounting for study variability, particularly when information from multiple RCTs is combined, including the requirement for considering model uncertainty and trial effect. For example, Berry et al. [92] employed a Bayesian hierarchical meta-analytic method rather than the frequentist approach to synthesize results from the RCTs of vitamin E supplementation used by Bjelakovic et al. [27] and Miller et al. [26] and concluded that vitamin E intake is unlikely to affect mortality regardless of dose. Other criticisms of the Bjelakovic et al. [27] meta-analysis have been noted, such as including small RCTs with few deaths; attributing deaths occurring after only a few months of antioxidant treatment; combining different nutrients in different forms with a large range of doses in a wide variety of population groups with varying health status; misclassifying some studies as being of high risk of bias; and basing the conclusion about antioxidant-specific effects on models that excluded selenium trials. However, no study before has considered evaluating the same studies selected by Bjelakovic et al. [27] for their benefits.

Our approach to considering overall beneficial outcomes of the RCTs selected by Bjelakovic et al. [27] provides a balancing perspective to their meta-analysis focused exclusively on the risk of all-cause mortality. It is worthwhile noting limitations associated with the use of meta-analyses of total mortality as an indication of harm without determining the cause of death. Bjelakovic et al. [26] did not investigate cause-specific mortality and thus could not eliminate those that lack any biological plausibility to antioxidant (or pro-oxidant) toxicity such as accidental deaths and homicides.

It should be noted that we did not include in our evaluation benefits found only in subgroups from the RCTs selected by Bjelakovic et al. [27]. For examples, we judged the WHS to have an overall null outcome among its 39,876 women >45 years of age receiving a 600 IU vitamin E supplement every other day for 10 years despite the observation by Lee et al. [14] of a significant reduction in the relative risk (RR) of the primary outcome of myocardial infarction, cardiovascular death or stroke among women >65 years of age (RR = 0.74; 95% Confidence Interval [CI]: 0.59–0.93; P = 0.009). Similarly, we judged the Vitamin E Atherosclerosis Study (VEAPS) to have a null effect (P = 0.08) on its primary trial endpoint of the rate of change in right distal common carotid artery intima-media thickness in 332 healthy men and women ≥40 years receiving 400 IU vitamin E for three years despite the observation by Hodis et al. [67] of significantly reduced concentrations of circulating oxidized low density lipoprotein (LDL) and increased resistance of LDL to oxidation, both biomarkers of cardiovascular disease risk. While such subgroup analyses typically lack the statistical power from which to draw definitive conclusions, these results do suggest that further RCTs are warranted to test the potential benefit of the intervention in that population.

As the relative number of positive outcomes evaluated in our analysis was based only on those studies analyzed by Bjelakovic et al. [27], it is possible we have underestimated the proportion of studies demonstrating benefit. For example, Bjelakovic et al. [27] selected the mortality data only from the AREDS study [18] report examining the effect of a combination antioxidant supplement on the development of cataracts and vision loss which had a null outcome (Table 2). However, the same mortality data, albeit provided in more detail, is available from the AREDS study [17] report examining the effect of the same supplement on age-related macular degeneration and vision loss in which a significant positive outcome was achieved. Further, in evaluating the positive, null and adverse outcomes of this set of studies, we did not take advantage of follow-up reports published after publication of the Bjelakovic et al. [27] article. For example, Watters et al. [93] recently examined prospectively the serum antioxidant status in men diagnosed with prostate cancer during the ATBC Study of 29,133 Finish smokers and in their 20 years follow-up found improved survival among the men with higher serum α-tocopherol at baseline (Hazard Ratio [HR] = 0.67; 95% CI: 0.45–1.00; P = 0.03), especially among cases who had received the vitamin E supplement and who were in the highest quintile of α-tocopherol at baseline (HR = 0.51; 95% CI: 0.20–0.90; P = 0.04) and at the 3-year follow-up measurement (HR = 0.26; 95% CI, 0.09–0.71; P = 0.02). Thus, a review including follow-up of earlier RCTs and new RCTs, e.g., of the SELECT [19] would offer a more comprehensive consideration of the benefit in relation to a potential (mortality) risk of antioxidant supplementation than provided here.

As noted above, the outcome of human studies on dietary antioxidants will depend on the initial status of the antioxidant defense network and oxidative stress in each subject, the dose(s) of the nutrient(s), and the concentration threshold for action of each nutrient. One of the major challenges in conducting RCTs to test the efficacy and safety of antioxidants (and other nutrients) as dietary supplements in reducing the risk of chronic disease, especially in primary prevention, is the need for very long durations of the intervention as we lack validated intermediary biomarkers of these conditions. The RCT conducted by Milman et al. [94] demonstrating the reduction of cardiovascular events (myocardial infarct, cardiovascular death, and stroke) by 400 IU/d vitamin E in type 2 diabetics with a haptoglobin 2-2 genotype but not other alleles suggests as well the need to consider nutrient/gene interactions in the study design. Total deaths within a single study or as part of a meta-analysis may be a utilized as one measure of the toxicity of a chronic intervention; however, collection of specific mortality data is essential as RCTs of dietary supplements cannot otherwise establish either biological plausibility or causality due to the inability to determine the response to discontinuation and the response to rechallenge.

The operational definition of dietary antioxidants provided by the U.S. Institute of Medicine states that these nutrients significantly decrease the adverse effects of reactive species, such as reactive oxygen and nitrogen species, on normal physiological function in humans [95]. These adverse effects are essentially described as oxidatively modified products of DNA, lipids, and protein such as 8-hydroxy-2’-deoxyguanosine, F_2α-isoprostanes, and protein carbonyls, respectively. However, the efficacy of different antioxidants in achieving this effect is critically dependent on a variety of factors, including the baseline value of these biomarkers and the dose and duration of treatment. While lowering these biomarkers has been associated in many experimental and observational studies with a reduced risk of some age-related pathologies and chronic diseases, no evidence appears available to suggest they are correlated with premature mortality. Importantly, the classification of vitamins C and E, carotenoids, flavonoids and related polyphenols as dietary antioxidants fails suggest their other various mechanisms of action, including anti-inflammation, induction of phase 2 detoxification enzymes, and modulation of redox sensitive signal transduction and gene expression. Thus, new studies of the clinical efficacy and safety of antioxidants as dietary supplements must consider appropriate forms and doses of the ingredients, a duration relevant to the pathogenesis and/or progression of the disease, and the biologically relevant molecular targets in populations most likely to respond in a fashion providing a high benefit/risk ratio.