Reexamination of a Meta-Analysis of the Effect of Antioxidant Supplementation on Mortality and Health in Randomized Trials

A recent meta-analysis of selected randomized clinical trials (RCTs), in which population groups of differing ages and health status were supplemented with various doses of β-carotene, vitamin A, and/or vitamin E, found that these interventions increased all-cause mortality. However, this meta-analysis did not consider the rationale of the constituent RCTs for antioxidant supplementation, none of which included mortality as a primary outcome. As the rationale for these trials was to test the hypothesis of a potential benefit of antioxidant supplementation, an alternative approach to a systematic evaluation of these RCTs would be to evaluate this outcome relative to the putative risk of greater total mortality. Thus, we examined these data based on the primary outcome of the 66 RCTs included in the meta-analysis via a decision analysis to identify whether the results provided a positive (i.e., benefit), null or negative (i.e., harm) outcome. Our evaluation indicated that of these RCTs, 24 had a positive outcome, 39 had a null outcome, and 3 had a negative outcome. We further categorized these interventions as primary (risk reduction in healthy populations) or secondary (slowing pathogenesis or preventing recurrent events and/or cause-specific mortality) prevention or therapeutic (treatment to improve quality of life, limit complications, and/or provide rehabilitation) studies, and determined positive outcomes in 8 of 20 primary prevention studies, 10 of 34 secondary prevention studies, and 6 out of 16 therapeutic studies. Seven of the eight RCTs with a positive outcome in primary prevention included participants in a population where malnutrition is frequently described. These results suggest that analyses of potential risks from antioxidant supplementation should be placed in the context of a benefit/risk ratio.

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 -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.

Methods
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 χ 2 test for homogeneity [32].

Efficacy of Antioxidant Interventions
Study outcomes were listed with regard to their design as primary or secondary prevention or therapy. When studies were reported to have more than one study outcome both were listed. Overall, 24 of 66 RCTs (36%) reported a positive outcome of the intervention indicating a benefit of antioxidant supplementation. Studies testing the effect of primary prevention by antioxidants showed positive outcomes in 8 of 21 studies (38%) while a single study reported a negative outcome. Reports from secondary prevention RCTs showed positive outcomes in 10 of 29 studies (34%) while 2 obtained negative outcomes. In studies directed to therapeutic outcomes, 6 of 16 RCTs (38%) achieved positive results. No significant difference (p = 0.78) was observed in secondary prevention trials regarding outcome compared with other studies. Most of these RCTs (60%) found neither benefit nor harm from the antioxidant supplement related to their primary goal (Tables 1 and 2). Seven of the eight RCTs with a positive outcome in primary prevention included participants in a population where malnutrition is frequently described [8,13,[33][34][35][36][37]. For example, five of these studies were conducted in the elderly, a population often described as at risk of micronutrient deficiencies [33][34][35][36][37]. The positive outcome in the SU.VI.MAX study of Hercberg et al. [13] was obtained only in males who, compared to the female participants, presented with a lower baseline status of antioxidants during enrolment into the study. The participants in the Nutrition Intervention Trials were residents of Linxian, China, an area with a high prevalence of malnutrition [8]. Penn et al. [36] The effect of dietary supplementation with vitamins A, C and E on cell-mediated immune function in elderly long-stay patients: A randomized controlled trial x x 40 Murphy et al. [41] Impact of vitamin A supplementation on the incidence of infection in elderly nursing-home residents: A randomized controlled trial x x 41 Blot et al. [8] Nutrition intervention trials in Linxian, China: Supplementation with specific vitamin/mineral combinations, cancer incidence, and disease-specific mortality in the general population x x 42 Li et al. [42] Nutrition intervention trials in Linxian, China: Multiple vitamin/mineral supplementation, cancer incidence, and disease-specific mortality among adults with esophageal dysplasia x x 43 Wenzel et al. [43] Alcohol-induced toxic hepatitis-a "free radical" associated disease. Lowering fatality by adjuvant antioxidant therapy x x    Takagi et al. [75] Pilot clinical trial of the use of alpha-tocopherol for the prevention of hepatocellular carcinoma in patients with liver cirrhosis x x 83 Virtamo et al. [9] Incidence of cancer and mortality following alpha-tocopherol and beta-carotene supplementation: A postintervention follow-up x x 84 Allsup et al. [76] Can a short period of micronutrient supplementation in older institutionalized people improve response to influenza vaccine?
x x

Discussion
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 -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 -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 2 -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.