Since the publication of our vitamin E dose–response meta-analysis, we have received hundreds of e-mails, letters, and phone calls as well as more than 40 Letters to the Editor submitted electronically. Annals has asked us to prepare a written response to 11 selected letters.

Blatt and Pryor and Krishnan and associates hypothesize that natural vitamin E supplements have greater benefit than synthetic vitamin E supplements, even though no trial has directly compared them on mortality outcomes. In response, we have performed a subgroup analysis comparing the 4 trials using natural supplements, all high-dosage (400 IU/d), with high-dosage trials using the synthetic form. The relative risks for all-cause mortality in the 4 trials that provided natural vitamin E were 1.00 (95% CI, 0.89 to 1.12) in the Heart Outcomes Prevention Evaluation (HOPE) (1), 1.83 (CI, 0.88 to 3.78) in the VECAT study (2), 1.22 (CI, 0.86 to 1.73) in CHAOS (3, 4), and 1.09 (CI, 0.72 to 1.66) in the SPACE study (5). The pooled relative risk comparing vitamin E with control was 1.04 (CI, 1.00 to 1.07) in the high-dosage trials of synthetic vitamin E and 1.05 (CI, 0.97 to 1.13) in the high-dosage trials of natural vitamin E. There was no evidence for a differential effect of synthetic versus natural sources of vitamin E (P >0.2 for heterogeneity).

Hemilä, Krishnan and associates, Lim and coworkers, Marras and colleagues, and Meydani and associates were concerned about the generalizability of our findings to healthy populations, since many of the high-dosage vitamin E trials enrolled participants with chronic diseases. In general, trials enroll high-risk individuals to increase study power. When interventions are shown to be effective in high-risk populations, subsequent trials are conducted in low-risk populations, and the same relative effects are often observed (as in statin trials in primary prevention). As for vitamin E, the findings of the Women's Health Study, a large primary prevention trial presented after the publication of our meta-analysis (6), suggest that the mortality increase that we observed is likely to apply to healthier groups. In this study, 39 876 healthy women were randomly assigned to receive 600 IU of vitamin E on alternate days or placebo for 10 years. At the end of follow-up, there were 636 deaths in the vitamin E group and 615 in the placebo group. This increase in mortality (relative risk, 1.04 [CI, 0.93 to 1.16]), although nonsignificant in this individual trial, is consistent with the findings of our meta-analysis.

Meydani and associates argue that we should have used a model with a constant effect of vitamin E up to a certain change-point dose and a linear effect above it. This model is problematic for a variety of reasons. First, it forces the effect of vitamin E to be constant over a wide range of low doses, an implausible assumption from a biological standpoint. Second, it forces a sharp change in effect at the chosen change-point dose, which is implausible in population studies even if there were sharp change-point effects in individual patients (7). Finally, the conclusions from this model largely depend on the chosen change-point, but this choice is difficult from a statistical perspective, and the sequential evaluation of change-points with selection of the best-fitting model underestimates the uncertainty in the final model (8). Our quadratic linear-spline model overcomes these difficulties and, contrary to the implications of Meydani and associates, it does not force a harmful effect above 150 IU of vitamin E per day. Meydani and associates argue that higher risk for death is not evident until the dosage exceeds 400 IU/d, instead of 150 IU/d as we describe. We indicated in our paper that establishing the precise dose of vitamin E at which the relative risk for death increases above 1 is very difficult. However, as also shown by the findings of the large Women's Health Study, it is highly likely that this threshold is below 400 IU/d.

DeZee and coworkers question our use of a hierarchical logistic regression model as opposed to traditional meta-regression. We believe that our model is more appropriate because it produces the same results that would be obtained if we had individual patient data on randomized treatment assignment, trial dose, and survival status. In fact, the results of their reanalysis of our data using traditional meta-analytic techniques are very similar to ours, and it is hard to believe that different conclusions can be obtained from their analysis and from ours. As argued in our paper, the dose of vitamin E is likely to be the explanation for the heterogeneity of the study results identified by DeZee and coworkers in their letter.

Possolo reanalyzed our data using a nonparametric, locally quadratic weighted regression model and also found a positive but statistically nonsignificant association between vitamin E dose and increased mortality. It is difficult to evaluate this analysis on the basis of the information provided in the letter, but we note that weighted regression routines available in general statistical packages are inadequate for meta-analysis, and specialized meta-regression programs are needed to obtain correct standard errors and confidence intervals (9).

DeZee and coworkers, Jialal and Devaraj, and Lim and coworkers criticize our decision to exclude trials with fewer than 10 deaths. We decided a priori to exclude studies with less than 1 year of follow-up or fewer than 10 deaths because we suspected that a variety of small or short-term trials designed primarily to evaluate the effect of vitamin E on physiologic intermediate outcomes would not collect or report mortality data systematically. On the other hand, it is hard to imagine trials designed to evaluate the effect of vitamin E supplements on clinical outcomes or mortality, the objective of our meta-analysis, with fewer than 10 deaths. In addition, there is no a priori reason to believe that trials with fewer than 10 deaths would quantitatively differ from larger trials or that excluding such trials would result in biased estimates of effect. Furthermore, when we compiled data from 11 trials with fewer than 10 deaths, the number of deaths among participants assigned to vitamin E exceeded the number of deaths among participants assigned to placebo (22 vs. 18 deaths, respectively; reference list available upon request).

Baggott suggests that some cardiovascular disease risk factors in some of the trials were higher among participants assigned to vitamin E, which would create a bias against vitamin E. However, it is unlikely that randomized experiments involving more than 135 000 participants would have imbalances in risk that systematically favor one randomization group.

Jialal and Devaraj mention 2 trials of vitamin E supplementation showing a beneficial effect on surrogate markers of atherosclerosis. The Transplant-Associated Atherosclerosis Study, a small trial (n = 40) that reported no deaths, showed ultrasonographic evidence for benefit among heart transplant recipients (10). The ASAP study, which used a 2 x 2 factorial design and examined the use of vitamin E (272 IU/d) and vitamin C (1000 mg/d) in hypercholesterolemic smokers, reported that vitamin E supplementation reduced progression of carotid artery disease (11, 12). However, the mortality data from the ASAP trial were consistent with our findings, since there were 4 deaths in the vitamin E groups and 2 deaths in the nonvitamin E groups after 3 years of follow-up. At year 3, all participants in the vitamin groups were given an open-label combination of vitamins C and E, while the placebo group continued without supplementation. At 6 years, the relative risk for death was higher in those assigned to the combination of vitamins compared with those assigned to placebo (19 deaths in 390 participants taking supplements vs. 3 deaths in 130 participants taking no supplements). On a final note, other trials have shown greater progression of atherosclerosis among those assigned to vitamin E (13-15).

Carter suggests that HATS, which we excluded because it had only 2 deaths, demonstrated that antioxidant vitamin supplementation (including 800 IU of vitamin E per day) provided angiographic evidence of slowed progression of coronary artery disease (16). On the contrary, HATS showed that antioxidant supplementation alone did not slow the progression of coronary plaque and that, surprisingly, antioxidants diminished the protective effect of simvastatin–niacin at slowing the progression of coronary disease.

Several authors have also expressed concerns over our choice of total mortality as an end point in view of the beneficial effects of vitamin E supplementation on physiologic variables related to oxidative stress. Although these surrogate markers can provide mechanistic insights, their clinical relevance is uncertain. In contrast, all-cause mortality, the outcome used in our meta-analysis, has unambiguous clinical relevance.

In conclusion, 19 randomized trials that together enrolled more than 135 000 participants have failed to document a survival benefit with vitamin E supplementation. In contrast, we have provided evidence that high-dosage vitamin E supplementation may increase total mortality. While future trials will refine the estimates of the effect of vitamin E supplementation and the dose at which the relative risk for death exceeds 1, we stand by our conclusions that use of high-dosage vitamin E supplementation should be avoided.