A number of lines of evidence point to a role for selenium in the prevention of cancer. Animal studies have shown an anti-tumorigenic effect of selenium supplementation. Epidemiologic studies have shown lower incidence of cancer in regions with higher soil and crop selenium levels. A number of case-control studies have also shown lower blood selenium levels in cancer patients (prior to the onset of cancer) than in controls, although there have been some conflicting results.
This double-blind, placebo controlled study was designed to look at the effect of selenium supplementation on the incidence of skin cancer in patients with a history of squamous cell carcinoma (SCC) or basal cell carcinoma (BCC) of the skin.
Endpoints: The primary endpoint, defined at the outset, was the occurrence of skin cancers. Safety endpoints included signs of selenosis and other illnesses.
Secondary endpoints were added in 1990, after 7 years, when more funding became available and it became apparent that there was a significant effect on the incidence of cancer. These secondary endpoints were total cancer mortality and the incidence of the most frequently occurring cancers (lung, colorectal and prostate). Starting in 1990, more intense efforts were also made to ascertain the vital status and cancer status of all participants.
Study termination: The study, initially designed to run for 15 years from 1983 to 1998, was terminated in 1993, after 10 years. This decision was made by the monitoring committee for a number of reasons, including lack of effect on the primary endpoint and a significant effect on secondary endpoints.
Mortality: Unadjusted, all cause mortality did not differ significantly between the groups (selenium 108 vs. 129 placebo). After adjustment for sex, smoking and age, there was a significant reduction in all cause mortality in the selenium group, RR=0.79, due to a reduction in cancer deaths. Cardiovascular mortality was the same in the two groups.
This study was carried out in regions of the US with low selenium levels and high rates of SCC, BCC and overall cancer mortality. Previous case-control studies had suggested an association between low selenium levels and the incidence of skin cancers, and this was the basis for the design of this study. Why was no association found between selenium supplementation and the primary endpoints of skin cancer incidence?
The authors speculate that apoptosis of transformed cells, which may be enhanced by selenium, occurs early in the course of skin cancer. Thus, the beneficial effect of selenium in preventing skin cancers would take many years to become evident and the duration of this trial may have been too short to demonstrate it. On the other hand, apoptosis occurs much later in the course of other cancers, which could allow a beneficial effect of selenium to become apparent much earlier for these tumors, as revealed in this study.
The authors note that the secondary endpoints were added well after the study was initiated (in fact, after it was 70% completed) and were the result of multiple comparisons. Such comparisons after the fact are always subject to chance findings of significance. In particular, the strength of the risk reduction must be looked at with great caution. They concede that further studies will be necessary to confirm these findings.
In favor of their results not being due to chance are the fact that the risk reduction was seen in multiple subgroups, and that is was just as strong during the 3 years after the secondary endpoints were defined as it was during the 7 years preceding the definition.
Finally, they note that the study's population was a very specific one: patients from low-selenium and high cancer regions, with a history of skin cancer and who were 75% male. Other studies will have to examine the effect of selenium in a more general population.
One piece of data that would have been interesting but was not provided is the breakdown of risk-reduction by sex. Only 25% of the participants were women, but the incidence of breast cancer was insignificantly higher in the selenium group. Even if there were not enough women to allow statistically significant conclusions to be drawn, the numbers would have been interesting, especially since, in some previous epidemiologic and case-control studies, a benefit of selenium was only noted in men.
The fact that the secondary endpoints, which are the interesting results of this study, were only defined after 70% of the study was completed is obviously a drawback, as the authors freely admit. In a sense, these should almost not be called "endpoints" but rather represent analysis after the fact.
Nevertheless, the results of this study are important. They do not occur in a vacuum, as pure "data-dredging", but rather in the context of multiple lines of evidence (animal, epidemiological and case-control) pointing to a beneficial effect of selenium, at least in certain populations. Previous epidemiological and case-control studies looking at vitamin A and beta-carotene suggested a beneficial effect in the prevention of lung cancer, but this was not borne out in a placebo-controlled trial. This points out the importance of randomized trials, such as the one reported here.
Selenium is toxic in excess, but the dose used in this trial produced no toxicity and led to blood-levels well below those felt to be potentially toxic.
Because of the post-hoc quality of the analysis, because of conflicting results concerning the effect of selenium in men and women noted in other studies and because of the specific population studied here, these results will need to be confirmed in other populations by further trials.
Dr. Clark, corresponding author of this study, submitted the following reply:
From: "Larry C. Clark" <SELENIUM@CCIT.ARIZONA.EDU>
Your summarization of the article is excellent. However, there are two related misconceptions presented in your comments: the timing of the additional endpoints and the post-hoc nature of the analysis.
2) While the full analysis of the 1983-93 period could be considered post-hoc because the additional endpoints were not formally defined at the beginning of the trial, the information presented in table 5 for the 1990-93 period after the formal definition of the endpoints should not be considered a post hoc analysis or data dredging. The endpoints were defined because they represented the leading cancers which were likely to have enough events for a valid statistical analysis. The evolutionary nature of the trial contributed to this design, since the original grant that funded the research provided only 2 years of support for the project and involved only 3 of the clinics.
We did not present any information on gender differences for treatment because they would have represented a sub-group analysis. The addition of selected sub-group analysis would have introduced information bias into the paper, which was already complex. It also would have made the presentation and interpretation of the information more difficult. We are planning a paper on sub-group analysis that would present treatment information on age, gender, smoking, and pre-randomization alpha tocopherol and selenium levels.
One additional point: the intervention agent was a high selenium brewers yeast which is almost entirely organic Se, about 40% is selenomethionine. (Nutrition 21, La Jolla, CA).
Date: Thu, 30 Jan 1997
1. It might have been appropriate to utilize some form of correction (ie., Bonferroni) to appropriately lower the P value associated with significance for the secondary end points, since they were added late (as readily admitted by the authors), and because with multiple analyses of data, it becomes more likely to find a statistically significant P value by chance alone.
2. If I read the paper correctly, it seems that a power calculation was not performed before the study was underway, but rather after the data were collected. This is important since this became a negative study for the primary endpoints. Is it possible that a small difference existed that was not found because the numbers of patients were too small?
3. Why was the study terminated early? Conservative analysis of the secondary end points suggests no real differences (post hoc, perhaps needed a correction lowering the P value, many confidence intervals included or very nearly included 1). So I don't think it was an ethical issue. Perhaps pressing on with the study might have revealed something more in terms of primary end points, since the mean follow-up was only 4.5 yrs and 4.3 yrs ( Se group/placebo group respectively). I guess I'm just suggesting that I don't see a substantial reason to stop early.
4. I think the most compelling thing about this study is that it suggests that Se in this dosage is safe (at least for this relatively short time-period).
Thank you for considering these comments. I actually think that the authors were actually more than fair in summarizing their data conservatively. It never ceases to amaze me how often the lay press vastly over-interprets this kind of study.
John Foxworth, PharmD
Clinical Pharmacology Section, Dept of Medicine, Truman Medical Center-West
Associate Professor, University of Missouri-Kansas City School of Medicine
Submit a comment about this article
Selected references cited in the article from the NLM's PubMed database.
References related to this article from PubMed.
Return to Journal Club home page