Randomized Trials of Antioxidant Supplementation for Cancer Prevention: Title and subTitle BreakFirst Bias, Now Chance—Next, Cause

In 1996, a wave of hope arose when the Nutritional Prevention of Cancer trial reported a 65% reduction in prostate cancer incidence in men receiving selenium supplementation.¹ This came only 2 years after the ATBC (α-Tocopherol, Beta Carotene) Cancer Prevention Trial had reported a 35% reduction in prostate cancer occurrence among men taking vitamin E supplements.² Suddenly, it appeared to make sense that this cancer could be prevented by bolstering antioxidant defenses in middle-aged and older men. Prostate cancer was not a prespecified end point in either trial, and although both results were based on post hoc analysis, randomization had worked and it seemed unlikely that the encouraging findings were due to confounding bias. Indeed, confounding stood as the likely explanation for discordance between the earlier beta carotene trials that failed to demonstrate any benefit and numerous observational studies reporting that men who consumed beta carotene–containing foods were at lower risk for developing cancer, particularly those of the respiratory tract.

Now, 12 years later, comes the disappointing news that 2 major trials conceived during the wave of hope found that neither selenium nor vitamin E supplementation, alone or in combination, produced any reductions in prostate cancer or cancer of any type. The results of these important studies, the SELECT (Selenium and Vitamin E Cancer Prevention Trial) by Lippman and colleagues and the Physicians' Health Study (PHS) II by Gaziano and colleagues, are reported in this issue of JAMA.^{3 - 4}

Despite the null findings, it is important to recognize what these trials have accomplished. SELECT had a simple, cost-effective design, completed accrual of more than 35 000 participants ahead of schedule—making it the largest individually randomized cancer prevention trial ever conducted—and maintained high rates of adherence and retention for 4 to 7 years. Given its statistical power, it is unlikely that the study missed detecting a benefit of even a very modest size. The PHS II, with more than 14 000 male physicians, was also remarkably well conducted and was especially noteworthy for its cost-efficiency, with most of the study having been conducted through the mail. Neither study has delivered its final word on the blinded, randomized results. Preplanned subgroup analyses are expected, particularly those aimed at baseline serum levels, smoking status, and genetic factors that might have modified response. After that, like Voyager space probes, each study will continue to contribute a wide range of valuable data as observational cohorts, before they fade away.

What can clinicians, researchers, and men concerned about prostate cancer prevention learn from the main results of these phase 3 trials? Three overarching points are worth consideration.

First, widespread use of prostate-specific antigen (PSA) testing has substantially restricted the design and interpretation of prostate cancer prevention trials, perhaps permanently. Neither trial mandated regular PSA testing; thus, end points were determined by routine clinical management. This greatly reduced the cost of each study; however, off-study PSA testing was extremely common among these healthy, motivated volunteers who had good access to care. In SELECT, 83% to 86% of men reported receiving PSA tests each year. Because of this intense surveillance, the geometric mean PSA at diagnosis was low (<4.81 ng/mL) and, more importantly, the proportion of cases with nonlocalized tumors was very low. Of 1758 prostate cancers detected, at diagnosis only 10 were extracapsular, 1 was regional node positive, and 9 were metastatic. Even assuming that these represent nonoverlapping cases, that would amount to only 20 nonlocalized cancers, or 1.1%. By comparison, among 13 740 cases of prostate cancer reported to the CAPSURE database from 1990 to 2007, 3.7% were at clinical stage T3bN0M0 or greater, excluding T3a cases with extracapsular extension only.⁵ Perhaps more strikingly, there was only 1 death from prostate cancer in SELECT despite nearly 200 000 person-years of observation. Even allowing for the requirement of a normal PSA and rectal examination at entry, that is an enormous deficit in expected mortality. Applying conservative assumptions about the exact age and follow-up time distributions in SELECT, approximately 75 to 100 deaths would have been expected based on the US age-specific prostate cancer mortality rates for 2000 through 2005.⁶ There also were indications of a deficit in advanced prostate cancer in PHS II, although a much smaller one.

How can an agent be shown to prevent serious, clinically significant prostate cancers when PSA testing may be rapidly removing those cancers from the population at risk before they progress? The effects on early stage lesions may still be observed, but if an agent has important effects on progression, those effects may be obscured. What to do? Trials could be conducted only in populations that shun PSA testing, or mandatory biopsies could be performed after some reasonable interval of exposure. Mandatory biopsies have the advantage of at least gathering end points more quickly and also solve problems due to the influence of an agent on PSA levels, as seen for finasteride and now, apparently, statins.⁷ Prostate cancer detection rates are exquisitely sensitive to any factor that affects PSA levels, biopsy rates, or biopsy sensitivity independent of an anticancer effect. In that regard, the nonsignificant increase in diabetes (presumably type 2) in the selenium group of SELECT deserves further scrutiny because of the recognition that obesity has a small suppressive effect on PSA and may have depressed prostate cancer ascertainment.⁸ The increase in prostate cancer in the vitamin E group, although not statistically significant (P = .06), is also worrisome. However, this could have been due to a chance elevation of the biopsy rate or biopsy sensitivity in that group, and the absence of any prostate cancer increase in the combination group in SELECT or in PHS II is reassuring.

Second, single-agent interventions, even in combinations, may be an ineffective approach to primary prevention in average-risk populations. It may be time to give up the idea that the protective influence of diet on prostate cancer risk—which is clearly observed in migrant studies and in populations transitioning to a Western diet—can be emulated by isolated dietary molecules given alone or in combination to middle-aged and older men. Evidence from the Prostate Cancer Prevention Trial suggests that pharmacological inhibition of intraprostatic androgen is capable of retarding the early growth of prostate cancers.⁹ For cost and safety reasons, it is unreasonable, however, to recommend that all men begin taking a 5α-reductase inhibitor daily once they reach age 50 years. On the other hand, nonpharmacological dietary prevention of prostate cancer is probably more complex and may involve certain inconvenient truths. Fortunately, no dietary change this profound is likely to be beneficial for prostate cancer alone. If it requires whole foods, extracts, or dietary patterns, it may be necessary to give up the reductionist need to know which molecule is most responsible and perhaps give up the notion of placebo controls as well. If it requires starting exposure early in life and sustaining it for decades, it may mean having to give up the idea of phase 3 trials altogether. This does not mean that whole food or complex mixture studies cannot be sound and biologically based. One can start, for example, by asking how diet affects intraprostatic androgen activity, a proven chemopreventive mechanism.

Third, it may be time to critically examine the methods used to vet hypotheses for some phase 3 trials. In hindsight, and now with more recent information available, it is easy to second-guess the rationale for these trials but more difficult, and much more important, to determine how to develop better phase 3 studies. Since the mid-1990s the tools available for preclinical science, including new transgenic models and genomic technologies, have vastly improved. Phase 2 trials, which should be determining which few interventions make it through the pipeline, will improve as intermediate end-point biomarkers become validated. Moreover, improvements in patient risk stratification, notably through multi–single-nucleotide polymorphism genetic testing,¹⁰ will help design targeted trials with higher statistical efficiency and more favorable risk-to-benefit ratios.

It now appears more likely that the unexpected results on prostate cancer from the earlier selenium and vitamin E trials were due to chance. In fact, the final report covering the entire blinded treatment period of the Nutritional Prevention of Cancer Trial, published in 2003, found that for unexplained reasons (chance), participants in the selenium group were less likely than those in the placebo group, by 14% vs 35%, to undergo biopsy when the PSA level was elevated.¹¹ After taking this into account, the reduction in prostate cancer incidence was restricted to the subgroup with low baseline PSA levels and the lowest tertile of plasma selenium, a subgroup comprising fewer than 17 cases between treatment groups combined.

Epidemiology teaches that every statistical association has only 3 possible explanations: bias, chance, and cause. Regarding nutritional prevention of prostate cancer, first-generation phase 3 trials were too reliant on biased interpretation of prior research; second-generation trials may have been too reliant on chance; yet there is every reason to believe that the next generation will have a firmer basis for causal hypotheses. Until then, physicians should not recommend selenium or vitamin E—or any other antioxidant supplements—to their patients for preventing prostate cancer.