|Home | About | Journals | Submit | Contact Us | Français|
Selenium is a trace element essential to humans. Higher selenium exposure and selenium supplements have been suggested to protect against several types of cancers.
Two research questions were addressed in this review: What is the evidence for
We searched electronic databases and bibliographies of reviews and included publications.
We included prospective observational studies to answer research question (a) and randomised controlled trials (RCTs) to answer research question (b).
We conducted random effects meta-analyses of epidemiological data when five or more studies were retrieved for a specific outcome. We made a narrative summary of data from RCTs.
We included 49 prospective observational studies and six RCTs. In epidemiologic data, we found a reduced cancer incidence (summary odds ratio (OR) 0.69 (95% confidence interval (CI) 0.53 to 0.91) and mortality (OR 0.55, 95% CI 0.36 to 0.83) with higher selenium exposure. Cancer risk was more pronouncedly reduced in men (incidence: OR 0.66, 95% CI 0.42 to 1.05) than in women (incidence: OR 0.90, 95% CI 0.45 to 1.77). These findings have potential limitations due to study design, quality and heterogeneity of the data, which complicated the interpretation of the summary statistics.
The RCTs found no protective efficacy of selenium yeast supplementation against non-melanoma skin cancer or L-selenomethionine supplementation against prostate cancer. Study results for the prevention of liver cancer with selenium supplements were inconsistent and studies had an unclear risk of bias. The results of the Nutritional Prevention of Cancer Trial (NPCT) and SELECT raised concerns about possible harmful effects of selenium supplements.
No reliable conclusions can be drawn regarding a causal relationship between low selenium exposure and an increased risk of cancer. Despite evidence for an inverse association between selenium exposure and the risk of some types of cancer, these results should be interpreted with care due to the potential limiting factors of heterogeneity and influences of unknown biases, confounding and effect modification.
The effect of selenium supplementation from RCTs yielded inconsistent results. To date, there is no convincing evidence that selenium supplements can prevent cancer in men, women or children.
Selenium is a trace element that is important for human health, but might also be harmful for humans when the taken in excess.
Fifty-five studies with more than one million participants were included in this systematic review. Forty-nine studies observed and analysed whether healthy people with high selenium levels in blood or toenail samples or with a high selenium intake developed cancer more or less often than other people. We found that people with higher selenium levels or intake had a lower frequency of certain cancers (such as bladder or prostate cancer) but no difference for other cancers such as breast cancer. However, it was not possible to determine from these studies that selenium levels or selenium intake were really the reason for the lower risk of cancer in some people. Factors apart from higher selenium levels could also influence the cancer risk: They might have had a healthier nutritional intake or lifestyle, have had a more favourable job or overall living conditions.
Six randomised controlled trials (RCTs) assessed whether the use of selenium supplements might prevent cancer. In general, there are two types of selenium supplements: one type uses the salt of selenium as main ingredient, the other type uses organic selenium. These two types may act differently in the human body when ingested. We assessed the quality of each trial according to four established methodological criteria. The trials with the most reliable results found that organic selenium did not prevent prostate cancer in men and increased the risk of non-melanoma skin cancer in women and men. Other trials found that participants using selenium salt or organic supplements had a decrease in liver cancer cases. However, due to methodological shortcomings this evidence was less convincing. We advise further investigation of selenium for liver cancer prevention before translating results into public health recommendations. We also recommend that there should be further evaluation of the effects of selenium supplements in populations according to their nutritional status as they may differ between undernourished and adequately nourished groups of people.
To maintain or improve health, access to healthy food and a healthy diet is important. Currently, there is no convincing evidence that individuals, particularly those who are adequately nourished, will benefit from selenium supplementation with regard to their cancer risk.
Selenium is a trace element essential to humans. Humans usually ingest selenium with crop and animal products and sometimes as functional foods or supplements. Speciation and concentration of selenium in food sources vary considerably, depending on plant and animal metabolism and growth conditions or animal nutrition (Duffield 1999).
Selenium species can be classified into selenium-containing organic compounds (e.g. selenomethionine, selenocysteine) and inorganic forms (selenate, selenite) (Rayman 2008a). Selenium yeast refers to a selenium-enriched yeast medium which usually contains 80% to 90% organically bound selenium with a high proportion of selenomethionine (Rayman 2004). Whether selenium is linked to specific beneficial health effects in humans is suspected but unproven and the debate on those effects is controversial (Drake 2006; Rayman 2008a).
The recommended daily allowance differs between regulatory agencies. For example, the highest amount of daily intake (55 μg selenium for adults) has been recommended by the US Institute of Medicine (Institute of Medicine 2009), whereas the WHO (World Health Organization) recommendations range between 30 and 40 μg/day for men and women (WHO 2004).
To prevent the risk of developing selenosis, the US Institute of Medicine has set the tolerable upper intake level to 400 μg per day for adults (Office of Dietary Supplements 2009). Besides the acute and chronic toxicity of high selenium exposure, possible harmful effects of long-term intake of lower dosages have also been discussed. However, effects of long-term intake of lower dosages are not so well investigated or understood (Vinceti 2001) and there may be differences between organic and inorganic forms (Rayman 2008a). A recent publication has questioned the current upper limit of ’safe intake’ and proposed a far lower ’safe level’ for long-term usage (20 μg/day for organic selenium) and a differentiation between organic and inorganic selenium sources (Vinceti 2009).
An accurate estimation of selenium exposure in epidemiological research presents a challenge. Individual exposure is often assessed as the concentration in blood specimens or toenail clippings (Longnecker 1996) or as estimated dietary or supplemental intake, but the validity of dietary logs and recall questionnaires has been questioned (Patterson 1998). For the measurement in blood specimens, either whole blood or blood fractions (plasma = blood without the cells; serum = plasma without the clotting factors) are used.
Selenium levels found in human specimens (Rayman 2008b) as well as the estimated intake of selenium (Alfthan 1996) show a high global variability. Different selenium levels within populations have been found to be related to ethnicity (Kant 2007), gender, age or smoking behaviour. Smoking tends to lower selenium biomarker concentrations despite being a source of selenium exposure (Kafai 2003). Globally, however, there are also inconsistencies as to how these factors are associated with selenium levels. For example, selenium levels increased with age in women, but not in men, in the French SU.VI.M.AX cohort study (Arnaud 2007), decreased with age in a female population in Ohio (Smith 2000) and two studies from Switzerland and Austria could not find an association between age and selenium status in either gender (Burri 2008; Gundacker 2006). Gender-specific nutritional and health behaviours, as well as gender-specific differences in selenium metabolism, may contribute to the observed discrepancies in selenium levels between genders (Rodriguez 1995).
The hypotheses about the potentially anticarcinogenic mechanisms of selenium include its effects on DNA stability, cell proliferation, necrotic and apoptotic cell death in healthy and malignant cells and its effects on the immune system (Whanger 2004). Selenium is involved in these processes as a source of selenometabolites and is part of selenium-containing enzymes (Hatfield 2001). The optimum level for the prevention and retardation of carcinogenesis in human cells has been discussed to be higher than the level commonly achieved under a diet not deficient in selenium (Whanger 2004). Gender differences regarding the effects of selenium on health, including cancer diseases, have been increasingly debated in recent years. Apart from gender differences in selenium levels, gender differences in selenium distribution in tissue or tumour biology might be involved in the differential health effects in women and men (Waters 2004).
However, selenium has also been shown to promote malignant cell transformation (Kandas 2009; Novoselov 2005; Su 2005) and protect cancer cells against stress-induced apoptosis (Sarada 2008) in animal and in-vitro studies and might therefore work as carcinogen.
Cancer is a leading cause of death worldwide. According to WHO estimates, 11.3 million people developed and 7.9 million died of cancer in 2007, with more than half of all new cases occurring in middle-income or low-income countries (WHO 2008).
The role of diet and nutrition for carcinogenesis and, as a potentially modifiable factor, in cancer prevention is still under debate. The identification of a nutrient supplement with cancer preventive properties would be a major breakthrough for public health. However, cancer is not a uniform disease and the existence of such a nutrient, or combination of nutrients, has been debated.
Case-control studies as well as systematic and non-systematic reviews have found conflicting results on risks of specific cancers and selenium exposure. Zhuo 2004 found a summary risk estimate suggestive for a protective effect of higher selenium exposure against lung cancer in a systematic review and meta-analysis of epidemiological studies; Brinkman 2006 found similar results for prostate cancer. However, two other epidemiological reviews concluded that studies did not support an association between selenium status and breast cancer risk in women (Navarro Silvera 2007; Waters 2004).
Vinceti and colleagues observed an increased melanoma incidence (Vinceti 1998) and mortality from melanoma (Vinceti 2000a) in both genders in a cohort of people from Northern Italy who where accidentally exposed to long-term consumption of drinking water with a high content of inorganic selenium. The standardised mortality ratio for melanoma in comparison to the non-exposed individuals in the municipality was 4.15 (95% CI 0.21 to 20.47) in women and 10.98 (95% CI 1.84 to 36.27) in men.
The first indication from an RCT that selenium supplements may reduce risk of gastrointestinal (GIT) cancers came from the General Population Trial in Linxian, China (Blot 1993; Dawsey 1994; Wang 1994). Study participants were living in regions with a very high rate of GIT cancers and subclinical deficiencies of several nutrients. The RCT investigated the efficacy of vitamin and mineral supplements to reduce cancer incidence and mortality, especially of oesophageal and gastric cancer, in middle-aged adults with four treatment factors for a period of 5.25 years. Participants receiving one study supplement (containing 50 μg selenised yeast, beta-carotene, alpha-tocopherol) had a reduced mortality from, but not incidence of, gastric cancer. Oesophageal cancer risk was not altered.
In the more recent French SU.VI.M.AX trial (Hercberg 2004), a supplementation with beta-carotene, vitamin C, vitamin E and 100 μg selenium-enriched yeast did not alter the incidence of cancer of the digestive tract after a median period of 7.5 years in women. In men, the incidence rate was lower in the intervention group than in the placebo group, but risk ratios (RRs) with confidence intervals (CIs) were not calculated because of low numbers. Bjelakovic 2006 conducted a Cochrane Review on antioxidant supplements for the prevention of gastrointestinal (GIT) cancers. Nine RCTs that investigated mono-selenium or selenium-containing supplements were included in this review. The authors concluded that selenium may potentially possess beneficial effects, but the results required further research before any recommendation could be made.
Selenium is suggested to be involved in central anti-carcinogenic processes. Selenium supplements are widely marketed with many health claims, the prevention of cancer being one of them. There is a worldwide debate about the association between selenium exposure and cancer risk or whether selenium supplements are effective in decreasing the incidence or mortality of cancer. Epidemiologic and other data suggest differential effects in men and women and there are hints that selenium supplements might even have harmful effects, this especially being the case in certain populations. This review is timely and important as several meta-analyses and systematic reviews have been published, but a comprehensive summary providing evidence from both prospective studies and intervention trials which a) include all types of cancer and b) look for gender-related differences does not exist.
Two research questions were addressed in this review: What is the evidence for
Randomised controlled trials (RCTs) and prospective observational studies (cohort studies and nested case-control studies) were included, irrespective of publication year, publication status or language.
All adult participants (aged 18 years and over) at risk of malignant neoplastic diseases.
We considered prospective observational studies (cohort studies including sub-cohort controlled studies and nested case-control studies) for inclusion if they assessed baseline exposure to selenium in apparently cancer-free individuals either as biochemical selenium status or estimated selenium intake at study inception. We considered RCTs for inclusion if they used selenium supplementation at any dose or route of administration for a minimum of four weeks versus placebo or no intervention. We excluded trials using selenium supplementation as part of a multi-component preparation, without a study arm using selenium monotherapy supplementation.
The primary outcome measures were
We conducted electronic searches in the Cochrane Central Register of Controlled Trials (CENTRAL, The Cochrane Library Issue 1, 2011), MEDLINE (via PubMed, 1966 to February 2011), EMBASE (1980 to 2010 week 50), CancerLit (February 2004) and CCMed (February 2011). We conducted the initial search in 2004 and updates in July 2007, January 2009, October 2009, and February 2011. As MEDLINE now includes the journals indexed in CancerLit, no further searches were conducted in this database after 2004.
We also searched the following online clinical trials databases: Clinical Trials of the American Cancer Society (http://www.cancer.gov, February 2011), the metaRegister of Controlled Trials (mRCT, http://www.controlled-trials.com, February 2011) and the German Cancer Study Register (http://www.studien.de, February 2011). The search strategies are given in Appendix 1. We scanned conference abstracts to identify unpublished material and searched the database for grey literature SIGLE (February 2004). This electronic database was discontinued in 2005.
Two review authors independently checked all electronic search results for eligibility. When search results could not be rejected with certainty on the basis of title and/or abstract, we obtained full text material.
We scanned bibliographies of papers retrieved with the described search strategy, and included publications to identify additional studies. If additional information was necessary we tried to contact all the correspondent authors of the included studies and asked investigators for information about unpublished trials.
Two review authors (GD, MH) independently applied the inclusion and exclusion criteria, if necessary with the assistance of an interpreter. We resolved disagreements by discussion and with the involvement of a third review author.
We used pre-tested extraction forms for epidemiological studies and RCTs to document data from the original material and assess the quality of studies. GD and another review author independently extracted data unblinded. GD checked extracted data for discrepancies and any discrepancies were discussed between both extracting review authors. In a small number of cases, we sought the opinion of a third review author to reach a consensus. If several reports from the same study were available, we considered the most recent to be the primary publication, but study details available from other publications were also extracted if not reported in the primary study reference.
We entered data from the extraction forms into a Microsoft Access database by hand. GD double-checked completely for errors and MH and GD triple-checked using descriptive database methods and plausibility checks.
For comparisons of selenium exposure as measured in serum and plasma specimens, we converted all data into the unit μg/l. Results provided as ppm (parts per million) or μg/g were converted using the factor 1.026 g/ml (weight density of serum) and data provided as μmol/l using the factor 78.96 (molecular weight of selenium). In order to be included, prospective observational studies had to report estimates of relative risk (RR (e.g. odds ratio (OR)) for each selenium exposure level.
The risk of bias in observational studies was assessed using an assessment form adapted from the Newcastle-Ottawa Quality Assessment Scale (NOS) for cohort and case-control studies (Wells 2004). The NOS was developed using a ’star system’ in which cohort or case-control studies are judged on the selection and comparability of the study groups and the ascertainment of either the exposure or outcome of interest.
The NOS form for cohort studies was used for all included observational studies. In addition, we assessed the risk of bias of nested case-control studies with the NOS case-control form. Both forms must be adapted a priori for use in a systematic review according to the research question and the review topic. For each question within this standardised assessment procedure either a ’star’ or ’no star’ is assigned to a study. A ’star’ indicates that study design was considered adequate and less likely to introduce bias.
We used the questions as reported in the appendices for study assessment; (*) means that for the corresponding item a ’star’ according to the NOS was assigned to the study. Key domains are the selection and comparability of the exposed and non-exposed cohort, the ascertainment of exposure and outcome and the length of follow-up. A study could receive a maximum of 9 stars in the cohort assessment (Appendix 2) and 9 stars in the assessment of the case-control part (Appendix 3).
The risk of bias assessment was based on the data provided in the included publications. We did not check other, not included publications for details. If an included study encompassed more than one publication with divergent rating in the NOS, we used the highest score.
We categorised generation of allocation sequence, allocation concealment, blinding and completeness of outcome data as adequate (low risk of bias), inadequate (high risk of bias) or unclear following the criteria specified in the Cochrane Handbook of Review of Interventions (Higgins 2009a). We considered these four items to be the key domains for bias risk assessment. Studies that were categorised as “adequate” in all four domains were considered to have a low risk of bias; studies with inadequate procedures in one or more key domains were considered to have a high risk of bias. Studies with unclear procedures in one or more key domains were considered to have an unclear risk of bias.
We assessed the fulfillment of ethical standards as follows:
When data were missing or discrepancies in study publications were found, we tried to make contact with the study investigators for further information. Contacting study authors helped to clarify discrepancies in several publications, e.g. differing data in text and tables within the same report, however, we retrieved no missing data or study details.
We performed data synthesis and analysis separately for RCTs and observational studies.
We restricted meta-analyses to cancers for which at least 5 studies were available. There were two reasons for this restriction. The first was practical and was to limit the number of analyses to be performed. The second was that we expected that results are heterogeneous, but heterogeneity cannot be described and quantified well if only very few studies are available (Higgins 2009b). Although the cut-off at 5 studies is somewhat arbitrary, this decision was made very early in the review process and was declared in the protocol.
This review includes only binary outcomes. If five or more studies were available for a specific type of cancer, we conducted a meta-analysis.
Study authors defined cancer cases either as diagnosis (i.e. cancer incidence) or death from cancer (i.e. cancer mortality) or as a combination of both. The term ’cancer risk’ is used in this paper as a generic term and refers likewise to cancer incidence, cancer mortality and combined incidence/mortality data.
A meta-analysis of highest versus lowest selenium exposure category was performed using a random effects model and by analysing the natural logarithm of the OR or RR, using the squared standard error of the natural logarithm of the OR or RR as weights. The latter was calculated from the reported upper and lower boundaries of the 95% CI of the OR or RR. If a 95% CI was not reported, we used the total number of cases and the total number of controls as well as the number of categories of selenium exposure to estimate the number of cases and controls per exposure category. We then used the standard normal approximation formula to calculate the standard error of the OR (comparing the highest versus the lowest exposure category (lnOR = (1/a + 1/b +1/c +1/d) where a, b, c, d are the four counts needed to calculate the OR via (a*d)/(b*c)).
We took the OR from the analysis that included the most extensive adjustment in the publication. For the calculation of the summary risk estimate, gender-aggregated data of mixed-gender studies were used when available.
We performed a Chi2 test for heterogeneity of study results. Additionally, we used I2 statistics (Higgins 2003) to quantify inconsistency. Meta-analyses were conducted using STATA 10.0 and STATA 11.0 software. We repeated meta-analyses that were included in this review publication using the Review Manager 5 statistical tool; for this, logarithmic data for the OR and the standard error were copied from STATA into Review Manager 5 and results were double-checked for errors.
We conducted sensitivity analyses to assess the effect of the methods used to assess selenium status/intake. We used gender-disaggregated data from mixed-gender studies together with data from single-gender cohorts for subgroup analyses by gender. We conducted the latter subgroup analyses to account for potential gender differences in selenium health effects (see Background).
We did not perform a meta-analysis of summary statistics with RCT data in this review as the minimum number of five studies, required for meta-analysis according to our review protocol, was not reached for any type of cancer.
Risk ratios (RR) of intervention trials that were not reported in the original publication were calculated on the basis of the number of participants and cases using the statistical tool for meta-analysis included in Review Manager 5. We also calculated the RR of adverse outcomes and its 95% CI, if sufficient data were available.
Citation style: Please note that we reference the sources of relevant information in a certain way to increase traceability of our results for interested readers. When the source of information is not the primary publication of an included study, the specific publication of interest is also referenced. For example “Hakama 1990 in: Knekt 1990” indicates that the cited paper is “Hakama 1990” as part of the mentioned study.
Three full text theses published in the US, could not be accessed (Coates 1987, in: Coates 1988; Menkes 1986a, in: Menkes 1986; Schober 1986, in: Menkes 1986). However, later journal publications were available and included in this review as main study publications (Coates 1988, in: Coates 1988; Menkes 1986b, in: Menkes 1986; Schober 1987, in: Menkes 1986). Thus the retrieval of the full text theses was considered to be unnecessary.
After excluding duplicates, the electronic search retrieved 4082 hits (flow chart of literature search: Figure 1). Of these, we excluded 3802 references as being clearly irrelevant due to title and abstract. The reasons for exclusion were:
Twelve publications of potential relevance were identified in the latest update search in February 2011. Due to time restrictions, these publications could not be included in the current review version and are listed in the section “Classification pending references”.
The remaining 268 publications were considered of possible relevance and re-evaluated.
One hundred and thirty-seven papers were identified for inclusion in this review: 80 papers referred to one ongoing and 49 completed observational studies. Fifty-seven papers referred to five ongoing and six completed randomised controlled trials.
A detailed description of the studies included is given in the table Characteristics of included studies.
Forty-nine completed observational studies were included in this review. Thirty-six studies were nested case-control studies, the others were sub-cohort controlled or cohort studies and one study used a cohort together with a nested case-control design. Sub-cohort controlled studies used (random) samples of the cohort as controls. The original papers were published between 1983 and 2009. Five studies were conducted in Asia (China, Japan and Taiwan), one in Australia, 19 in Europe (including data from Belgium, Denmark, Germany, Greece, Italy, Netherlands, Norway, Spain, Sweden, Channel Islands, Finland, France and the UK) and 24 in the US Overall, the studies included more than 1,078,000 participants. European study populations made up 45%, the US 45%, Asia 9.7%, and Australia 0.2% of all study participants. The median size of the study populations was 10,494. Twenty-six studies included men and women, one did not report gender, 17 included only men and five only women. For a substantial proportion of the study populations (38%), gender was not reported. Forty-one percent of participants were men, 21% were women. Six studies with gender-mixed populations reported results stratified by gender. The study populations were derived from 42 different cohorts. Twenty-three cohorts were non-randomly recruited, e.g. included volunteers, and 19 cohorts consisted of a random (or total) sample of the population of interest, which was either a specifically exposed population such as male tin-miners in China or the general population.
Thirty-seven studies specified the age range of their included participants, the majority of which included adults over 40 years. Five studies investigated nutritional and/or supplemental selenium intake, using food-frequency questionnaires or interviews. Forty-three studies assessed biochemical selenium status:
One study measured both serum selenium levels and intake. The mean follow-up period was up to three years in five studies and longer than three years in the remaining studies. Generally, study authors grouped the cases following the ICD classification that was up-to-date at the inception of the cohort observation. The level of disaggregation of data varied remarkably between the studies. While some studies reported cancer risk according to organ systems (e.g. urinary tract, respiratory tract), others stratified their data by one or two organs (e.g. female breast, urinary bladder). Only in the case of skin cancer did studies also differentiate according to histological type (e.g. melanoma, basal cell carcinoma).
For the following outcomes, five or more studies were included in the review and observational data were meta-analysed:
Table 1 provides an overview on the studies for each outcome. Five studies gave data for the group of “other” cancers, which encompassed any type of cancer not reported separately in the study publications. The definition of the group of “other” cancers varied between studies including predominantly rare cancers but also cancers of unknown origin. The results of the studies within the category “other cancers” are mentioned for the sake of completeness, however, due to the diversity of outcomes the results were not included in further analysis or discussion of this review.
Six randomised controlled trials with a total of 43,408 participants (94% men) were included in this review. All used parallel group designs with either two arms ( Li 2000; NPCT 1996; Reid 2008; Yu 1991; Yu 1997) or four arms (SELECT 2009). Three were conducted in China ( Li 2000; Yu 1991; Yu 1997), two in the US (NPCT 1996; Reid 2008) and one in the USA/Canada/Puerto Rico (SELECT 2009).
Selenium supplements and placebos were administered daily. As an active intervention, trials used 200 μg (NPCT 1996; Yu 1991; Yu 1997) or 400 μg (Reid 2008) selenium in the form of selenised yeast tablets. Li 2000 used 500 μg sodium selenite and SELECT 2009 used 200 μg L-selenomethionine.
Three Chinese trials investigated the preventive efficacy of selenium supplementation against primary liver cancer in different high-risk populations. Participants were either carriers of the Hepatitis B surface antigen (HBs-Ag) with normal liver function or first-degree relatives of liver cancer patients. Two trials used se-lenised yeast (Yu 1991; Yu 1997) and one sodium selenite (Li 2000).
The US Nutritional Prevention of Cancer Trial (NPCT) investigated the influence of selenium on the development of squamous and basal cell skin cancer in a high-risk group (NPCT 1996). Participants were men and women between 18 and 80 years. All had a history of two or more basal cell carcinomas or of one squamous cell carcinoma. All results were reported for two periods of follow-up: the intended study period (from 15 September 1983 to 31 December 1993: Clark 1996 see NPCT 1996) and the entire blinded intervention period (from 15 September 1983 to 31 January 1996: Duffield-Lillico 2002; Duffield-Lillico 2003 see NPCT 1996).
A sub-study of the NPCT (Reid 2008) investigated the efficacy of a higher selenium dose, supplied as selenised yeast orally, in the prevention of non-melanoma skin cancer in one of the NPCT study sites. Study design was similar to the NPCT study, except that 423 participants at this study site were randomised to placebo or intervention with higher selenium content.
The Selenium and Vitamin E Cancer Prevention Trial (SELECT) investigated the preventive potential of selenium, as selenomethionine, and vitamin E supplementation in men of diverse ethnic backgrounds against prostate cancer (SELECT 2009).
In 1990 additional secondary endpoints were identified post-hoc in NPCT 1996 (total cancer mortality, total cancer incidence, incidence of lung, prostate, colorectal cancers). Furthermore, the incidences of female breast cancer, bladder cancer, oesophageal cancer, melanoma, haematological cancers and cancers of the head and neck were also reported in trial publications (NPCT 1996). Reid 2008 reported the incidence of internal cancers.
SELECT 2009 investigated several pre-specified secondary outcomes, including the incidence of and deaths from any type of cancer, lung cancer, colorectal cancer and other cancers (excluding prostate, basal cell and squamous cell skin cancer).
Of these, 131 papers did not fulfil the inclusion criteria. Eighty-eight of these publications were rejected on the basis of abstract and title in the second evaluation, 43 papers were retrieved as full-text and their reasons for exclusion are described in the table Characteristics of excluded studies.
The main reasons for exclusion were:
The median value of ’assigned stars’ was seven in the cohort study assessment and eight in the (nested) case-control study assessment out of a maximum of nine stars each (Figure 2 and Figure 3). A summary of the rating according to the Newcastle-Ottawa-Scale (NOS) is presented in Table 2.
All but one cohort study received five to nine ’stars’ in the NOS. The exception (two ’stars’) was an early investigation, which was only available in abstract form for assessment (Clark 1985). For three items of the NOS cohort assessment, less than 70% of the included studies were considered adequate: representativeness of the cohort for the target population (51% of the studies received a ’star’), demonstration that cancer was not present at study commencement (69%) and completeness of follow-up data (49%).
The representativeness of the cohort for the target population is a matter of external validity and generalisability of study results, but a systematic deviation of participants from the target population might also introduce bias to study results. The target population of included studies depended on the study objectives and could have been the general population, but also special occupational groups. Studies that did not identify their target population or recruited volunteers were not assigned a ’star’ to this question. Differential selection of study participants, e.g. volunteers, from the target population can lead to confounding by factors associated with selenium status and cancer incidence, e.g. nutritional behaviour or socio-economic position.
All included studies chose comparison groups (cases/controls or exposed/non-exposed) from the same study population. This approach increased the comparability between groups.
The presence of undiagnosed cancer at the beginning of the study, when specimens for selenium analysis were taken, might influence selenium levels. People with certain types of cancer have been found to have lower selenium levels than healthy controls. This might lead to an overestimation of the protective effect of higher selenium levels against cancer, if undiagnosed cancer cases are prevalent. Some studies tried to investigate this source of bias by excluding cases that occurred within a certain period from the beginning (mostly one or two years).
Follow-up data were considered either as complete or as missing data unlikely to introduce bias to study results in 50% of the included observational studies. In the other cohorts, losses to follow-up were more than 5% and a description of losses to follow-up was not provided. A high attrition may alter the characteristics of the population under investigation and impede generalisability of study results to the intended target population (external validity). The presence of attrition does not necessarily mean that the study results are biased. However, given the possibility that selenium status may be linked to sociodemographic variables and socio-economic position which may also influence participation in follow-up procedures, a differential effect of attrition may introduce bias towards under- or over-estimation of the true exposure effect.
Thirty-six included observational studies were nested case-control studies and therefore additionally assessed using the NOS case-control form. The number of ’stars’ in the NOS assessment of the case-control part of the studies ranged from six to nine, with more than 85% having received eight or nine ’stars’. Although the included prospective case-control studies were generally assessed as having a low risk of bias, in some studies concern arose due to case definition and the question of representativeness of the cases. Definition of cases was considered inadequate in 19% of the nested case-control studies as cases were identified by self-reporting, linkage to databases with unclear validity of data or procedures were not described. The magnitude and direction of bias that might have been introduced to the study results is unclear.
In 19% of studies, not all identified cases (or an appropriate sample of them) were included in the trial analyses or selection procedures for analysed cases were not reported. In some studies, blood specimens were lost due to technical problems (e.g. cooler breakdown in one study centre), others did not have enough material available for analysis or cases for analysis were otherwise selected in a non-random manner. This might bias the estimates of association in either direction.
An overview of the risk of bias in the included randomised controlled trials is presented in Table 3.
All three trials on liver cancer risk (Li 2000; Yu 1991; Yu 1997) were considered to have an unclear risk of bias. In these trials, the generation of allocation sequence and the allocation concealment were not reported. One study mentioned that the drop-out rate was similar in the intervention and control group, the remaining two studies did not report the completeness of outcome data. Blinding was judged as adequate in all three studies as the use of placebo supplements was reported. We inferred from this procedure that at least the study participants and the physicians directly involved were blinded towards treatment status.
We would like to point out that we are not entirely convinced that Li 2000 is a randomised controlled trial. The study investigators used the term ’randomization based on the residence area’ and did not describe the randomisation procedure any further. As participants were recruited from 17 villages, the villages and not the individual participants (stratified by village) could have been randomly assigned to the intervention and control group. However, we could not make contact with the study investigators and clarify these questions. A randomisation of villages instead of individuals would have introduced bias to the study results as the incidence of liver cancer is known to differ between areas as a result of environmental factors.
Studies with inadequate or unclear allocation concealment have been found to overestimate the benefit of interventions in RCTs, especially in trials with subjective outcomes (Pildal 2007; Wood 2008). In all three liver cancer RCTs, follow-up and case-detection procedures were not reported, so the influence of subjective factors on case detection, such as interpretation of bodily symptoms as trigger of further diagnostic tests, is unknown. Although we judged blinding as ’adequate’ in all three liver cancer trials, we do not know whether it was successful in practice for patients, healthcare providers and outcome assessors.
These uncertainties about study methods seriously weaken our confidence in the reported RCT results on liver cancer risk.
Both included RCTs on non-melanoma skin cancer (NPCT 1996; Reid 2008) were considered to have a low risk of bias with adequate generation of allocation sequence, allocation concealment, blinding and completeness of outcome data. Reid 2008 was a sub-study of NPCT 1996.
Both studies also reported data for several secondary outcomes that were introduced post-hoc, such as lung, prostate, colorectal cancer and total cancer incidence. Placebo and selenium groups were similar regarding the distribution of the risk factors smoking, age, gender and PSA levels at randomisation. New cases were identified in the biannual participants’ interviews and by documenting cancer screening and diagnostic procedures from their medical files. Detection bias might have been introduced by a different use of diagnostic procedures in both groups: Men in the placebo group were, despite similar PSA levels, more likely to have undergone prostate biopsy (NPCT 1996) than men in the selenium group (Duffield-Lillico 2003b see NPCT 1996) which might have led to an underestimation of prostate cancer incidence in the selenium group and an overestimation of the treatment effect. As the background of this difference is unclear, it cannot be ruled out that differential health behaviours and diagnostic activities in both study groups might have affected the detection of lung and colorectal cancer as well, at least in the 75% male participants.
SELECT 2009 was considered to have a low risk of bias with adequate generation of allocation sequence, allocation concealment, blinding and completeness of outcome data.
Placebo and selenium groups were similar regarding the distribution of age, education, smoking status, PSA levels and race/ethnicity at baseline. No inter-group differences were found in the utilisation of PSA tests, prostate biopsies or digital rectal examinations during the course of the study.
When comparing the risk of cancer in higher and lower levels of selenium exposure, a summary risk estimate of 1 suggests that there is no association between selenium exposure and cancer, a summary risk estimate below 1 suggests a possible protective effect of higher selenium exposure and a summary risk estimate above 1 suggests a possible harmful effect of higher selenium exposure.
Results of 13 prospective observational studies on total cancer risk including data of more than 143,000 participants were meta-analysed. The cohorts of Salonen 1984 and Salonen 1985 overlapped. Hence, only data from Salonen 1985 were included in the meta-analysis. Fex 1987 had to be omitted as the CI value was not reported and could not be calculated from the available data.
In participants in the highest category of pre-diagnostic selenium exposure, the summary risk estimate was OR 0.69 (95% CI 0.53 to 0.91) for cancer incidence and OR 0.55 (95% CI 0.36 to 0.83) for cancer mortality in both genders (Analysis 1.17) when compared with participants in the lowest exposure category. Statistically significant heterogeneity was observed both among studies on incidence (I2 = 49%) and mortality (I2 = 58%).
Analyses by gender found the risk to be lower in men (incidence: OR 0.66 (95% CI 0.42 to 1.05), mortality 0.56 (95% CI 0.38 to 0.81)) (Analysis 1.20) than in women (incidence: OR 0.90 (95% CI 0.45 to 1.77), mortality: OR 0.92 (95% CI 0.79 to 1.07) (Analysis 1.19).
All studies used either serum or serum and plasma biomarker levels for the assessment of selenium status. Analysis 1.18 shows the results in ascending order of baseline exposure of those studies that reported their category borders. The graph does not reveal a clear pattern of a relationship between baseline biomarker level and cancer risk.
Seven studies were included in the meta-analysis. No association was seen between baseline selenium levels and incidence of breast cancer with an overall risk estimate of OR 1.00 (95% CI 0.78 to 1.29) (Analysis 1.1). The heterogeneity of trial results (I2 = 5.4%) was low and not statistically significant.
Meta-analysis of bladder cancer incidence in five observational studies found an inverse association with an overall risk estimate of 0.67 (95% CI 0.46 to 0.97) suggesting a protective effect of higher selenium levels against bladder cancer (Analysis 1.2) (overall heterogeneity: I2 = 30%).
Gender-disaggregated data were only available from Michaud 2005 indicating a protective effect in women, but not in men in this study. However, two studies (Michaud 2002; Nomura 1987) had only male participants and both found a non-significantly reduced bladder cancer risk for higher selenium exposure (Analysis 1.2). Heterogeneity was not reduced by gender stratification (I2 = 40% in study results for men).
Eleven studies were included in this meta-analysis. Data from Menkes 1986 and Knekt 1990 were not meta-analysed as the study population of the former overlapped with another meta-analysed study (Comstock 1997) and results of the latter were presented in insufficient detail.
The summary risk estimate for lung cancer incidence in both genders was 0.75 (95% CI 0.54 to 1.03) (Analysis 1.3). Statistically significant moderate heterogeneity was seen between study results (I2 = 54%).
In the meta-analysis according to gender using gender-stratified study results (Analysis 1.4), the summary risk estimate for women was OR 0.83 (95% CI 0.43 to 1.61)) and for men OR 0.88 (95% CI 0.61 to 1.28)). Heterogeneity among study results was not reduced by stratification. However, we expected the results for gender-combined data to be more or less a combination of the separate results for women and men. This was not the case here with ’gender-neutral’ data suggesting a larger protective effect than gender-stratified data. This discrepancy might relate to differences in study designs or populations.
In Knekt 1998, 95% of the lung cancer cases occurred in men. We repeated the meta-analysis of gender-disaggregated data categorising Knekt 1998 as ’men-only’ study and found a slightly changed summary risk estimate for men (OR 0.81 (95% CI 0.56 to 1.18)). The only study using nutritional intake assessment for exposure classification (Kromhout 1987) found no association with lung cancer risk (Analysis 1.6). Two studies measured selenium content in toenails with inconsistent results: Participants of the (all women) Nurses’ Health Study (Garland 1995) showed an increased lung cancer risk with higher selenium toenail levels, while an inverse association was observed in the Netherlands’ Cohort Study (vd Brandt 1993). The remaining eight studies used serum or plasma selenium levels. The summary OR was 0.84 (95% CI 0.66 to 1.06) with low heterogeneity (I2 = 3.5).
We plotted the studies using serum/plasma in ascending order of baseline exposure level (Analysis 1.5). No clear pattern of a relationship between baseline exposure levels and lung cancer risk could be seen in this graph. The two studies, which suggested the largest protective effect of higher selenium levels, were Knekt 1998 and Kabuto 1994. However, two other studies with quite similar biomarker levels found discrepant results (Nomura 1987; Ratnasinghe 2000).
Fourteen epidemiological studies on prostate cancer incidence were included in the meta-analysis. The summary risk estimate for higher selenium exposure was OR 0.78 (95% CI 0.66 to 0.92) (heterogeneity: I2 = 37%) (Analysis 1.7).
Stratification by the method of selenium assessment showed a reduction in prostate cancer risk for higher baseline biochemical markers (OR 0.74 (95% CI 0.61 to 0.88)), but not for higher estimated selenium intake (OR 1.00 (95% CI 0.73 to 1.36)) (Analysis 1.8). The inverse association between selenium biomarkers and prostate cancer incidence was stronger for toenail levels (OR 0.53 (95% CI 0.35 to 0.81)) than for blood levels (OR 0.81 (95% CI 0.68 to 0.97)) (Analysis 1.9). Heterogeneity among study results was slightly reduced with these stratifications.
Overall, the strongest inverse associations were seen in studies from the US published before 2001. These findings cannot be explained by differences in baseline selenium levels alone. Analysis 1.12 shows the results for studies using serum or plasma measurements in ascending order of selenium levels. For quite similar categories of selenium concentration, studies indicated different effects (Goodman 2001 versus Clark 1985; Nomura 2000 versus Peters 2007 and Gill 2009 see Epplein 2009).
Five observational studies were included in the meta-analysis of gastric cancer incidence. The summary risk estimate for both genders was OR 0.66 (95% CI 0.43 to 1.01) in the highest exposure category when compared with the lowest (I2 = 50.8%) (Analysis 1.13). However, in this meta-analysis one cohort (Mark 2000 in: Wei 2004) is included twice because the results were reported stratified according to cardia and non-cardia gastric cancer.
We repeated the meta-analyses including alternately the results of Mark 2000 (see Wei 2004) for cardia and non-cardia gastric cancer. The summary OR was 0.75 (95% CI 0.47 to 1.21) when data for non-cardia cancer were included and OR 0.59 (95% CI 0.38 to 0.93) when data for cardia cancer were included.
Using the available gender-stratified results for meta-analysis, the risk estimate for men was OR 0.43 (95% CI 0.14 to 1.32) (I2 = 56.1%) and for women OR 0.73 (95% CI 0.12 to 4.35) (I2 = 62.3%) (Analysis 1.14).
Five observational studies reported data on colon or colorectal cancer incidence. The summary risk estimate was OR 0.89 (95% CI 0.65 to 1.23) for both genders (I2 = 3.8%) (Analysis 1.15), OR 0.69 (95% CI 0.42 to 1.12) for men and OR 1.06 (95% CI 0.57 to 2.00) for women (Analysis 1.16).
Three RCTs investigated the efficacy of selenium supplementation for liver cancer prevention. All RCTs were conducted in China and with participants of different high-risk groups in the Qidong province.
Yu 1997 investigated a 4-year supplementation period with 200 μg selenium yeast/day in 226 male and female hepatitis B-surface antigen (HBs-Ag) carriers. Eleven cases (person-time incidence rate: 1573.03/100,000) were detected in the placebo group and four cases in the selenium group (RR 0.36 (95% CI 0.12 to 1.11)) during the 8-year follow-up period. The mean blood selenium level was 152 ng/ml in the intervention group and 107 ng/ml in the control group (during the intervention period).
Yu 1991 reported on a trial with 2474 male and female first-degree relatives of liver cancer patients who also received 200 μg selenium yeast/day. During the study period of two years, 10 cases in the selenium and 13 cases in the placebo group were observed (RR 0.55 (95% CI 0.24 to 1.25)).
Li 2000 randomised 2065 male HBs-Ag carriers receiving 0.5 mg sodium selenite daily for 3 years. Thirty four cases of liver cancer occurred in the 1112 subjects receiving selenium and 57 cases in the 953 placebo subjects (RR 0.51 (95% CI 0.34 to 0.77)).
A higher risk for non-melanoma skin cancer was seen in the 200 μg/day selenium supplementation group of the NPCT (RR 1.27 (95% CI 1.11 to 1.45)) (Duffield-Lillico 2003 see NPCT 1996). The increase remained statistically significant after multivariate adjustment (HR 1.17 (95% CI 1.02 to 1.34)). No variation in this effect by age, gender or smoking status was statistically significant. Mean selenium plasma concentration of the participants was 114 ng/ml at the time of randomisation. An increased risk for total non-melanoma skin cancer was seen in all tertiles of baseline plasma levels (Reid 2008 see NPCT 1996). The increased risk could be observed more pronouncedly at one study centre (Macon, Georgia) than at the other study sites. The percentage of female participants was higher in Macon, but distribution of other factors, in particular baseline selenium levels, was similar to the other sites and the reason for the different effect, if not due to chance alone, remained unclear.
In the sub-study with 400 μg/day selenium supplementation (Reid 2008), no alteration of non-melanoma skin cancer risk was seen (HR 0.91 (95% CI 0.69 to 1.20)). Gender-stratified analyses found a decreased risk in women (RR 0.40 (95% CI 0.20 to 0.80)) and a non-statistically significant increased risk in men (exact data not reported). Distribution of baseline plasma selenium levels was similar in this sub-study to the participants of the NPCT main study and we found no evidence for an effect modification according to baseline selenium exposure.
Neither the intervention with 200 μg/day nor with 400 μg/day found evidence that supported a preventive efficacy of selenium yeast supplementation against non-melanoma skin cancer in these populations. The results of the NPCT raised concerns about harmful effects of selenium yeast supplementation.
The comparison of the results of both investigations did not show a dose-response relationship between the amount of supplemental selenium intake and outcome: While a harmful effect on non-melanoma skin cancer risk was seen with the 200 μg/day supplement, the 400 μg/day supplement showed a decreased risk in women and no effect in men. This might indicate differential biological mechanisms and health effects of selenium at different levels of supplemental intake by gender. The 400 μg/day supplemental intake led to a mean selenium level of approximately 250 ng/ml while 200 μg/day supplemental intake increased plasma levels to approximately 200 ng/ml.
At the end of the total blinded treatment period in the NPCT, the RR 1.17 (95% CI 1.02 to 1.35) for basal cell carcinoma (BCC) was increased in the 200 μg/day selenium group. Multivariate adjustment attenuated the effect to a HR of 1.09 (95% CI 0.94 to 1.26). Eliminating the cases that occurred within the first two years of supplementation had no further effect on the RR. We found no statistically significant variations in effects according to age, gender or smoking status.
Reid 2008 found an adjusted HR of 0.95 (95% CI 0.69 to 1.29) in the 400 μg/day selenium sub-study.
In the NPCT, selenium supplementation increased both the (unadjusted) risk for squamous cell carcinoma (SCC) (RR 1.32 (95% CI 1.09 to 1.60)) and the multivariate-adjusted HR 1.25 (95% CI 1.03 to 1.51). After exclusion of the cases that occurred within the first two years, a slight decline in the effect of selenium supplementation was seen (leading to statistical non-significance). We found no statistically significant variations in effects according to age, gender or smoking status. The adverse effect of selenium supplementation on SCC risk seemed to increase with increasing plasma selenium levels at baseline. A higher risk of non-melanoma skin cancer incidence was seen only in participants with baseline plasma levels in the highest two tertiles of baseline exposure (greater or equal to 105.6 ng/ml), which suggested a possible interaction between supplementation and baseline exposure.
In the 400 μg/day selenium sub-study (Reid 2008), we found no alteration of SCC risk by selenium supplementation (HR 1.05 (95% CI 0.71 to 1.56)).
In the SELECT trial, we found no evidence of benefit of L-selenomethionine supplementation (compared to placebo) for a median of 5.5 years on prostate cancer incidence (HR 1.04, (95% CI 0.90 to 1.18), (99% CI 0.87 to 1.24)) (SELECT 2009). The adjusted HR for prostate cancer in the selenium plus vitamin E group compared to placebo was HR 1.05 ((95% CI 0.91 to 1.20), (99% CI 0.88 to 1.25)).
SELECT was terminated early in 2008 following the recommendation of the data and safety monitoring committee. The committee had some concern over a statistically non-significant increase in prostate cancer in the vitamin E-alone group (HR 1.13, (95% CI 0.99 to 1.29), (99% CI 0.95 to 1.35)).
SELECT failed to replicate the findings of the NPCT, which observed a reduction in prostate cancer incidence in the selenium yeast group (adjusted HR 0.48 (95% CI 0.28 to 0.80)), and found no evidence for a statistically significant cancer preventive efficacy of selenium supplements. NPCT investigators argued that as randomisation worked, bias seemed unlikely to explain the positive findings for some cancers including prostate cancer in their trial. Regarding prostate cancer, however, a differential participation of men with elevated PSA levels in prostate biopsies was observed in the selenium and placebo group (35% versus 14%; Duffield-Lillico 2003 see NPCT 1996). This may have occurred by chance and could have contributed to an overestimation of the effect of selenium supplementation in the NPCT.
The NPCT (NPCT 1996) reported on a number of secondary outcomes identified post-hoc and other cancers, which had occurred in the trial population.
The total cancer incidence was reduced in the selenium group (adjusted HR 0.75 (95% CI 0.58 to 0.97)) after a mean follow-up of 7.4 years (Duffield-Lillico 2002 see NPCT 1996). Cancer mortality was also reduced (adjusted HR 0.59 (95% CI 0.39 to 0.87)). A gender-stratified analysis revealed that any possibly protective effect on total cancer incidence in the NPCT was confined to men (adjusted HR 0.67 (95% CI 0.50 to 0.89)). The adjusted HR for women was 1.20 (95% CI 0.66 to 2.2). Because of the predominance of male participants, these gender-differential effects added up to a net benefit for the total study population.
The observed risk of specific cancers in the selenium group was lower than in the placebo group for lung cancer (adjusted HR 0.74 (95% CI 0.44 to 1.24)), colorectal cancer (adjusted HR 0.46 (95% CI 0.21 to 1.02)), oesophageal cancer (adjusted HR 0.40 (95% CI 0.08 to 2.07)) and prostate cancer, as mentioned above. On the contrary, it was higher in the selenium group than in the placebo group for melanoma (adjusted HR 1.18 (95% CI 0.49 to 2.85)), bladder cancer (adjusted HR 1.28 (95% CI 0.50 to 3.25)), breast cancer (adjusted HR 1.89 (95% CI 0.69 to 5.14)), head and neck cancer (adjusted HR 1.27 (95% CI 0.47 to 3.42)) and lymphoma/leukaemia (adjusted HR 1.25 (95% CI 0.43 to 3.61)). Reid 2008 found no evidence that selenium supplementation altered total cancer incidence (RR 1.10 (95% CI 0.57 to 2.17)).
In the NPCT, 35 participants had withdrawn from the study because of adverse effects, mainly gastrointestinal upset. The RR for adverse events in the selenium group was 1.51 (95% CI 0.74 to 3.11) (own calculation, based on the number of all randomised participants).
In SELECT, men in the selenium group had an increased risk of alopecia (RR 1.28 (99% CI 1.10 to 1.62)) and dermatitis (grade 1 to 2) (RR 1.17 (99% CI 1.00 to 1.35)), but not of halitosis, nail changes, fatigue, nausea or dermatitis (grade 3 to 4). A statistically non-significant increase in diabetes mellitus type II in the selenium-alone group (HR 1.07 (99% CI 0.94: 1.22)) was seen.
An increased risk for diabetes mellitus type II was also observed in the NPCT (Stranges 2007 in: NPCT 1996). A secondary analysis of participants who did not have diabetes at the start of the study revealed an excess risk in the selenium group (adjusted HR 1.55 (95% CI 1.03 to 2.33)). We found no statistically significant interactions with age, gender, smoking status and BMI. The RR for developing type II diabetes mellitus was higher in participants in the upper two tertiles of plasma selenium levels, indicating a possible interaction with baseline exposure status.
Both the SELECT and the NPCT results suggest that long-term supplementation with selenium may adversely affect glucose metabolism and increase the risk for diabetes mellitus type II.
The three trials on liver cancer and Reid 2008 did not mention the occurrence of adverse effects. One paper stated that no case of selenosis had been observed during the trial.
The aims of this review were to examine the efficacy of selenium supplements in preventing cancer and possible associations between selenium exposure and the risk of cancer incidence and mortality.
From our meta-analyses of 13 prospective observational studies on total cancer risk, we found a reduced cancer incidence and mortality with higher selenium exposure. The risk of cancer disease was 31% (95% CI 9% to 47%) lower in the highest category of selenium exposure than in the lowest, the risk of death from cancer was 45% (95% CI 17% to 64%) lower. Subgroup analyses by gender suggested that a beneficial effect of higher selenium exposure, if existent, could be higher in men than in women.
The risk of developing bladder cancer was reduced by 33% (95% CI 3% to 54%) and that of prostate cancer by 22% (95% CI 8% to 44%). The risk of lung, gastric or colorectal cancers were also found to be reduced with higher selenium exposure; however the confidence intervals of the summary risk estimates overlapped the 1.
No association was seen between selenium and the risk of breast cancer.
For all other types of cancer, data were only available from less than five epidemiological studies; thus results were narratively summarised. None of the study results supported an association between selenium exposure and gynaecological cancer risk, while results for cancers of the gastrointestinal, respiratory or urological tract were inconsistent. For respiratory and urological cancers (other than bladder, prostate or lung cancer), studies reported either no associations or increased risks for participants with a higher selenium exposure. For gastrointestinal cancers, studies found either no associations or reduced risks with a higher selenium exposure.
As is the case with all meta-analyses of epidemiological data, our findings have potential limitations resulting from study design as well as quality and heterogeneity of the data. These limitations can complicate the interpretation of the summary statistics.
We identified six randomised controlled trials, which investigated mono-selenium supplements in the prevention of non-melanoma skin cancer, liver cancer and prostate cancer. There was no convincing evidence that selenium supplementation can prevent non-melanoma skin cancer or liver cancer in women or men or prostate cancer. The results of the Nutritional Prevention of Cancer Trial (NPCT) raised concerns about possible harmful effects of selenium supplements.
The NPCT was considered to have a low risk of bias and found a statistically significant increase in the incidence of squamous cell carcinoma as well as a trend towards an increased basal cell carcinoma incidence with selenium yeast supplementation. The RR increase was 17% for total non-melanoma skin cancer and 25% for squamous cell carcinoma in both genders after a mean follow-up of 7.4 years. The number needed to harm in the study population was 19 (95% CI 10 to 143) after a duration of five years selenium supplementation. A sub-study of the NPCT, which used supplements with a higher selenium content, found no differences in non-melanoma skin cancer risk between the active and the control group in both genders combined (HR 0.91, 95% CI 0.69 to 1.20), but an indication of a possible modification of effect by sex or gender, which was not seen in the main part of the NPCT.
Secondary outcomes of the NPCT indicated a lower total cancer incidence and mortality in the selenium group in men, but not in women. Analyses stratified according to cancer type found a statistically significantly reduced risk for prostate cancer. These results were not seen in the sub-study of the NPCT and could not be replicated in the SELECT trial, although it should be noted that this trial used a different intervention.
The SELECT trial was a large prostate cancer prevention trial in the general population of North America. It was considered to have a low risk of bias and found no alteration of prostate cancer incidence by L-selenomethionine supplements after a median follow-up of 5.5 years (HR 1.04, 95% CI 0.90 to 1.18).
One out of three liver cancer prevention trials reported a statistically significantly reduced risk of liver cancer (RR 0.51 (95% CI 0.34 to 0.77)) for male carriers of the hepatitis B surface antigen taking inorganic selenium supplements (sodium selenite) for three years. This was in contrast to the other two studies reporting no statistically significant effect of organic selenium supplements (selenium yeast) for the same cancer site. Due to several methodological concerns relating to randomisation and completeness of outcome data, the risk of bias was unclear for all three RCTs. Therefore, we cannot conclude that there is strong support for selenium supplements as agents for the prevention of liver cancer.
We reviewed data from prospective observational studies, where selenium exposure was measured in populations without evidence of cancer, and which were then followed-up for a specified period of time. This approach minimised the risk of reverse causality if an association between selenium exposure and cancer was observed in the study.
The included studies differed in terms of selenium exposure measurement, types of outcomes, study designs and study populations. The low number of studies for most of the meta-analysed types of cancers prevented a thorough investigation of the sources of heterogeneity between study results. In particular, we could not explore the influence of specific sources of bias or the methodological quality of epidemiological studies on heterogeneity.
The investigations included over 1,078,000 individuals from diverse study populations predominantly from Europe and the USA and to a lesser extent, Asia and Australia) (also see: Dennert 2008). No prospective observational study on selenium and cancer risk could be identified from Africa or South America. This regional distribution reflects the under-representation of non-Western and resource-poor countries in epidemiological research (Pearce 2004). Differential regional representation in epidemiological studies is of special interest for this review, as selenium levels in humans vary significantly around the world. The selenium levels measured in the included cohorts reflect a broad range of naturally occurring selenium exposure as measured in cross-sectional studies worldwide. However, some of the lowest as well as the highest selenium levels in humans were reported in literature from South American populations (Jaffé 1992), a region which was not investigated in any of the reviewed observational studies.
More than half of the studies included mixed gender populations, but the majority of them did not report gender-disaggregated data. In the available gender-specific results, men are over-represented, which inhibits the further understanding of a possible gender-specific association between selenium exposure and cancer risk in epidemiological data. This also limits the generalisability of the review results that evidence for a clinically relevant sex or gender-differential effect exists.
This review investigated a diverse range of cancers, but cancer is not a uniform condition and malignant neoplasms show great differences in tumour biology. Only non-melanoma skin cancer, liver cancer and prostate cancer were investigated in the included prevention trials as primary outcomes. The results cannot be generalised to other types of cancers.
Regarding the three main outcomes, specific characteristics of the study populations may also limit the generalisability of the results to non-participants. Participants of the included RCTs on skin and liver cancer belonged to populations with a high risk for the outcome under investigation. The participants of the NPCT were mostly older and white, predominantly male inhabitants of the US Mean plasma selenium concentration was in the lower range of U.S. levels, but still well above the average selenium level of Europeans. A possible interaction with baseline selenium levels was found resulting in an increased risk for developing squamous cell cancer and diabetes mellitus type II in participants receiving the selenium supplement who had higher baseline levels. An indication of interaction and modification of effect was also found for gender regarding some of the study results. When applying the study results to other populations, characteristics of the population regarding gender and selenium exposure should be considered.
Apparently healthy men over 50 years of age from the general population of North America participated in the SELECT trial on prostate cancer prevention. The large sample size and the inclusion of non-white participants from different socio-economic backgrounds supported the generalisability of study findings to other adequately nourished populations.
Selenium supplements contain either organic or inorganic species of selenium or a mixture of both, e.g. in the form of selenised yeast. The different species of selenium may exhibit differential effects on human health. Four included RCTs used selenised yeast supplements and found either a harmful or no effect of supplementation on the main study outcome. The SELECT trial used L-selenomethionine supplements, which is the major component of selenium yeast, and also found no preventive efficacy. The only RCT investigating sodium selenite supplements found a protective effect against liver cancer, but was considered to have an unclear risk of bias. It is also unclear how applicable these results are in other settings and populations with a different nutritional status.
Interpretation of the results of clinical trials using selenium supplements should consider the different biological forms as well as their potential differential health effects when supplemented.
The 49 observational studies were heterogenous, not only regarding methodology, but also in the quality and level of detail of reporting. The publications included ranged from a congress abstract to a full-text dissertation.
Five observational studies measured nutritional or supplemental selenium intake using questionnaires or interviews. Most studies, however, relied on selenium biomarkers such as toenail, serum or plasma selenium levels. Percentile borders, for example quartiles or quintiles, were usually applied as cut-off points for exposure categories. Our analyses were based on the comparison of highest versus lowest baseline exposure category. In our meta-analyses, different methods of selenium measurement and different numbers of exposure categories covering different absolute selenium levels were combined.
Assessment of total selenium intake with food-frequency questionnaires (FFQ) or interviews has proven difficult in other investigations because of the lack of food composition data which adequately reflects regional and seasonal variations in selenium concentration. The Duffield 1999 trial compared duplicate diet collections, dietary logs, FFQ and biomarkers as measurements for selenium intake and status in New Zealand men and women. The FFQ overestimated the mean selenium intake in study participants when compared with laboratory analyses of duplicate meals. The ranking in quartiles according to intake for both dietary logs and FFQ differed from the results from duplicate meals. Correlation between all three dietary measurements and selenium biomarkers (whole blood and plasma) were modest (r = 0.1 to 0.4) at the best. Also Karita 2003 found only a modest correlation between selenium intake as estimated from FFQ and from a 7-day dietary log in Japanese men and women. In the same study, a correlation between both estimates of dietary intake and serum selenium levels could not be seen.
Validity problems, possibly leading to exposure misclassification, have also been reported when questionnaires are used to assess supplement use (Murphy 2002).
Regarding biomarkers for selenium measurement, Ashton 2009 showed in a systematic review that plasma and whole-blood selenium concentrations increased with higher selenium intake in supplementation studies. Plasma, whole-blood and presumably serum selenium levels, although Ashton 2009 could not identify serum studies for their systematic review, were therefore considered by the authors to adequately reflect a short-term increase in supplemental selenium intake in healthy adults. However, authors also found significant, unexplained heterogeneity in the reaction of participants’ plasma selenium levels to selenium supplementation.
Concerning the estimation of long-term nutritional intake with biomarkers, Longnecker 1996 demonstrated a high correlation between long-term selenium intake as estimated from duplicate food portions and single measurements from whole blood, serum and toenail specimens.
These findings support the concern that the ranking of selenium exposure differs according to the instruments used to assess intake and also between intake assessment and biomarkers. Exposure misclassification may have biased the results of individual studies and a meta-analysis of observational data is likely to reflect these biases. Non-differential exposure misclassification might have occurred in all included studies due to measurement errors or as a result of the gap between the theoretical definition of selenium exposure and the measurement thereof, which served as a proxy. Non-differential misclassification might lead to an under- as well as over-estimation of an effect in the presence of more than two exposure categories. Our approach of a meta-analysis covering different methods of selenium assessment might have introduced additional heterogeneity to review results.
A concern, which we cannot clarify to date, is that biomarkers do not adequately reflect intake of both organic and inorganic selenium species. Animal studies indicate that selenium from inorganic sources is not retained so well in the body as organic selenium. Selenium from organic sources led to higher blood selenium levels and a higher activity of glutathione peroxidase than equal doses of inorganic supplements in veterinary studies (Slavik 2008; Steen 2008). However, symptoms of acute toxicity were observed in animals with a lower intake of inorganic than organic selenium species (Kim 2001; Tiwary 2006). Hall 2008 found an increased genotoxic effect in human cell lines by sodium selenite in comparison to organic selenium. When considering the possibly differential effects of selenium species on human health, an adequate interpretation of the biomarkers representing selenium exposure would require knowledge of the selenium sources of the individual.
The observation in our review that cancer risks only show an association with biomarker levels, but not with nutritional intake might therefore be a consequence of an invalid measurement of nutritional intake, which biased the results towards the null. Alternatively, it might likewise reflect that there truly is no association and that the findings from the biomarker studies were the result of chance and measurements of nutritional intake may provide better estimates of the exposure situation than do biomarkers, which may misclassify the exposure to inorganic selenium sources.
Furthermore, the comparison of risks between the highest and the lowest exposure category is most suitable to identify an effect when there is a consistent decrease or increase across absolute exposure levels. Other associations (e.g. threshold effects or U-shaped relationships) may be missed by this method of meta-analyses or the true effect might be diminished.
All included studies recruited participants pre-diagnostically and cases and control subjects stemmed from the same population. This approach decreased potential differences between both groups, which could have influenced cancer disease or death due to factors other than selenium exposure. We included the results from each study in meta-analyses which were adjusted for the highest number of additional variables.
All studies on total cancer risk identified cases by using registry links or a combination of several methods and losses to follow-up were low. Two studies on cancer incidence and two studies on cancer mortality analysed less than 80% of all identified cases (incidence: Persson 2000: 76%; Coates 1988: 79%; mortality: Kok 1987: 71%; Kornitzer 2004: 57%). The main reason for this was samples missing for selenium measurement. Not all studies that assessed mortality as a measure of cancer risk excluded participants with cancer disease at study inception. This might have led to an overestimation of a protective effect when selenium levels were lowered by the presence of cancer.
We therefore consider the results for cancer incidence to provide the more valid estimation for the relationship between selenium exposure and cancer risk than the mortality data.
All but two of the studies on prostate cancer risk identified cases by using links to cancer registries or a combination of personal follow-up interviews with PSA screening. Two studies with health professionals used self-reporting for case identification, followed by confirmation through medical records. The number of people lost to follow-up was low in all studies included. Two studies, however, included less than 80% of all identified cases in their analyses (Brooks 2001: 39%; van den Brandt 2003 in: vd Brandt 1993: 77%) because samples were not available for selenium measurement or diagnosis was not confirmed. In Brooks 2001, bias might have been introduced to the results to some extent, as the demographic variables differed between the identified and analysed cases.
Losses to follow-up were low in three studies (Michaud 2002; Nomura 1987; Zeegers 2002 in: vd Brandt 1993) and unclear in two on bladder cancer risk (Helzlsouer 1986 in: Menkes 1986; Michaud 2005). Endpoints were ascertained in elaborate ways in four studies including linkages to registries and regional and national databases; one study relied on the self-reporting of study participants (Michaud 2005). The latter investigation compared bladder cancer in the Nurses’ Health Study (women) and the Health Professionals Follow-Up Study (men) and was the only one to report gender disaggregated data. A gender-differential association between selenium exposure and bladder cancer risk was found, but the role of potential biases due to possible different self-reporting behaviour in these two distinct cohorts remained unclear.
The second study which found a statistically inverse association between selenium exposure and bladder cancer risk was Zeegers 2002, which could only analyse 70% of the identified bladder cancer cases as specimens for selenium measurement were not available for the remainder.
Most of the studies included controls for smoking and age, either by matching or by using multivariate techniques. However, only a few considered the potential effect of other factors. Possible confounding factors could be another food nutrient or a certain behaviour, which exhibits cancer protective effects and is associated with higher intake of selenium-rich foods. Furthermore, intake of heavy metals and other dietary factors may modify selenium health effects or the relationship between selenium exposure and biomarker concentration (overview in: Vinceti 2000). Metabolic interactions, for example, are known for arsenic (Zeng 2005).
Even in studies that considered the influence of a specific factor, validity of the assessment of the potential confounder can be challenging and is not commonly reported in study publications. For example, control for smoke exposure as a known risk factor for several types of cancer seems a crucial issue in epidemiologic studies on cancer risk. Cigarette smokers, for example, tend to have lower selenium biomarker levels, though cigarette smoking is a source of selenium exposure itself. Therefore an inverse association between selenium and lung cancer risk might also be the result of residual confounding and effect modification by smoking. Exposure to environmental and household smoking, which has been shown to be associated with increased risks of cancer (Gorlova 2006; Nishino 2001), might also be associated with selenium status due to differential nutritional behaviours or other mechanisms. We are not aware of any study that investigated this issue. Unknown factors may influence an observed selenium - cancer association and thus pose a challenge as to causal inferences.
Some of these factors cluster in population groups according to socio-economic position (SEP). Only a few studies attempted to control for indicators of adult SEP as potential confounders, e.g. education, occupation or income. None used a composite index of indicators or considered childhood SEP. Some studies restricted their cohorts to certain subgroups of a population, such as occupational groups, and were likely only to include people of a similar adult socio-economic background.
It has been claimed that associations between vitamins and diseases are the result of confounding by social and behavioural factors acting over the course of a lifetime (Lawlor 2004). Lawlor 2004 argued that the divergent results from epidemiological and randomised controlled studies on the prevention of cardiovascular diseases can be explained by unmeasured confounding due to SEP. Risk of most cancers is - like cardiovascular morbidity - known to decrease with higher SEP. Research also indicated a positive association between higher SEP and selenium biomarkers (Barany 2002; Niskar 2003). However, other investigations did not confirm these findings: Kant 2007, for example, did not find an association between a measure of household poverty and selenium status.
The hypothesis of possible confounding due to SEP leading to an indirect association between selenium and cancer in epidemiological research would be consistent with the results for all types of cancers in this review - including the null association with breast cancer - with the exception of prostate cancer findings. Prostate cancer has been found to be more often diagnosed in men of a higher SEP (Dalton 2008) while we saw a protective association with higher selenium exposure. However, it remains unclear whether the more frequent diagnoses of prostate cancer in men with higher SEP reflects an excess of prostate cancer incidence in this population. It might also result from differential health and screening behaviours leading to a detection of otherwise symptom-free cases while men with a lower SEP tend to be overrepresented in diagnoses of advanced stages of the disease (Rapiti 2009). More information on screening and diagnostic behaviour of the male cohort participants would be necessary to further elucidate these findings.
For prostate cancer, studies published before 2000 and especially those from the US found a larger protective effect with higher selenium levels than did later studies. We consistently observed this in the studies on lung cancer.
This might be attributable to differences in study design or populations (with the later studies being the larger studies including the general population) or changing health and screening behaviours over time in the case of prostate cancer studies. It could also reflect publication bias in earlier years favouring positive results.
An alternative explanation could be a ’threshold’ effect for a possible protective effect of selenium against prostate cancer around a certain level, which has been diminishing due to the increasing use of selenium supplements in the US. Brooks 2001 reportedly observed results consistent with a threshold effect at a level of 108 μg/l serum selenium. Conversely, a threshold effect was not seen in another study with almost the same percentile limits (Goodman 2001) in a population of asbestos workers, who may have had other sources of selenium exposure than the participants of Brooks 2001 from the general population. It has been frequently suggested that an increase in selenium intake might be beneficial only for men with lower selenium levels as glutathione peroxidase activity reaches a plateau above approximately 95 (range 89 to 114) μg/l (Rayman 2000).
We found no clear indication of a threshold effect in either lung or prostate cancer in the overview of study results. Heterogeneity between studies might therefore not reflect a consistent biological threshold effect of baseline selenium exposure levels, but a cluster of known and unknown influences of factors related to study design, population and potential biases.
Large epidemiological studies are not designed to test for a specific aetiological hypothesis, but enable research to investigate a large number of possible associations. Given the multiplicity of possible comparisons, associations between selenium exposure and cancer endpoints may have resulted from chance alone.
Factors which seemed to account partially for the inter-study heterogeneity were type of outcome measure (incidence or mortality), assessment of exposure and gender.
Considering the possible influences of bias, residual confounding and modifying factors on the selenium-cancer relationship, the summary estimates from meta-analyses should be interpreted with caution. Meta-analyses of spurious findings in observational studies increase the precision of a summary risk estimate, which does not itself get nearer to the true value and may suggest an nonexistent association (Egger 1998).
The NPCT, its sub-study and the SELECT trial were considered to have a low risk of bias with adequate sequence generation, allocation concealment, blinding and reporting of findings.
In the three trials on liver cancer prevention, quality of reporting was an issue and they were considered to have an unknown risk of bias. The individual trials were - in some cases discrepantly - reported in several papers and essential questions regarding sequence generation, allocation concealment, handling of drop-outs and withdrawals and detection of outcomes remained unanswered. This might be due to inadequate reporting, but might also hint to flaws in trial design and implementation. We were uncertain that the only trial which found positive results for selenium supplements in liver cancer prevention randomised participants individually. A cluster randomisation of participants who lived in the same area/village, which may have been the procedure in this investigation, might have introduced additional bias to the study results, e.g. due to different environmental factors contributing to liver cancer development or detection, and might have led to an overestimation of the protective efficacy of selenium. A duplication of results with a rigorous study design would be necessary to assess the effect of sodium selenite on liver cancer incidence.
The literature search included the major international databases in the English and German language and we applied a broad search strategy supplemented by hand searching for references. We assume that we identified all randomised controlled studies and prospective observational studies relevant to our review questions. As we did not search databases in other languages, e.g. Chinese or Russian, we cannot rule out that we missed smaller studies which were not published in international journals. There is also a chance that we might have missed observational studies whose results on selenium exposure and cancer were reported in the body of a paper but not mentioned in the paper’s title or abstract, even if the paper is indexed in the searched databases.
Although we tried to contact all investigators for missing or additional data on their studies, we were unable to retrieve answers to questions we had regarding methodology or outcomes in some studies. This applied particularly to earlier epidemiological studies where primary investigators may have relocated, died, or where data were not available in a current electronic format. Similarly, we could not make contact with the primary investigators of the Chinese RCTs.
The risk of bias assessment was based on the included publications. The risk of bias of studies that did not adequately describe the study design in the included publication but gave a reference to another paper, might therefore have been overestimated in this review.
Another concern, especially with the epidemiological studies, is publication bias. Cohort and nested case-control studies are not exclusively designed to test for a specific exposure-outcome association, but enable researchers to investigate a range of questions. It is conceivable that unfavourable results were less likely to be published.
We decided a priori to conduct meta-analyses only when five or more studies were available for a study outcome. As a result of this cut-off, we did not conduct meta-analyses for a number of observational study outcomes with two to four studies available (see: Table 1). Our primary intention was to facilitate the investigation of heterogeneity between studies that were included in meta-analyses, in order to avoid producing more precise, but still unexplainably biased results. However, results of this review revealed that the cut-off at five studies did not guarantee this possibility and could therefore be reconsidered in future updates.
Regarding the question of the efficacy of selenium for cancer prevention, the cut-off point led to the situation that no meta-analytic procedures were conducted for RCTs. Looking out our results, however, only liver cancer results would have pooled without the cut-off. Results for this meta-analysis of liver cancer have already been published in another systematic review (Bjelakovic 2008) and were included in the discussion here. Replicating the meta-analysis of liver cancer trials would not have changed the results of this systematic review.
The authors of this review came from different disciplines and have different focuses, e.g. epidemiology, clinical medicine and nutrition. We consider this internal variety of expertise to be a strength of this review and made use of it by applying double-checking procedures during the entire review process whenever possible.
The idea of selenium supplementation for cancer prevention received broad support following the NPCT and the publication of several large epidemiological studies which supported the hypothesis of an aetiological relationship between low selenium status and cancer development. Combs 2005 stated that “the hypothesis that Se (selenium) can affect cancer risk is supported by a remarkably consistent body of scientific evidence” (Combs 2005, p346). These ideas stimulated the largest ever cancer prevention trial SELECT, which failed to provide support for this hypothesis. Disagreement between the results of this systematic review and other publications may partly be explained by the differentiation between aetiology and efficacy in the research questions of this review.
A number of systematic reviews with and without meta-analyses have been conducted on selenium and the risk of different types of cancer. Overall, our combined risk estimates are consistent with their results and slight discrepancies in numbers are attributable to different inclusion criteria. However, some of the previous publications arrived at more favourable conclusions regarding a possible protective association of higher selenium exposure against cancer. Our meta-analyses of epidemiological studies suggested an inverse association between selenium exposure and risk of several cancers in men, which was reflected in a reduced overall cancer incidence and mortality. Associations with toenail selenium levels tended to be larger than with serum or plasma levels and in general no associations were seen with selenium intake. These findings were consistent with the secondary outcomes of the NPCT, which suggested a preventive efficacy of selenium supplements against several types of cancers in men, the strongest of which was prostate cancer. However, the large-scale SELECT trial failed to confirm any beneficial effects of supplemental selenium intake on prostate cancer risk. An earlier ecological analysis of a nationwide program to increase selenium intake with fortification in Finland also found no evidence of any protective effect against prostate cancer (Vinceti 2000).
Overall, there is little evidence for an association between selenium exposure and cancer risk in women and, if existent, it is likely to be small. Our meta-analyses do not support a protective association between higher selenium exposure and breast or colorectal cancer in women. These findings are consistent with the results of the NPCT trial, where all protective effects of selenium yeast supplementation were confined to men.
It has been argued that gender-related outcomes may reflect different exposure levels at baseline possibly related to gender-specific nutritional behaviour, which might be true for comparisons of distinct women-only and men-only cohorts (Michaud 2005). However, comparisons by gender within studies also pointed to a differential effect at similar exposure levels. We cannot rule out that sex or gender differences are observed by chance only, but laboratory and animal research have suggested sex differences in selenium metabolism and biology. Also sex-specific tumour biology and the predominance of specific cancer types may contribute to differential health outcomes in women and men. However, we cannot estimate the magnitude sex or gender differences possibly contribute to the observed differential health outcomes in men and women.
These considerations are of special interest as selenium supplements are aggressively marketed especially to women with regard to breast cancer prevention and treatment, which is not supported by data from observational or clinical investigations.
Heterogeneity between studies was not largely reduced by gender stratification in our meta-analyses. Furthermore, we expected that non-gender stratified data from observational studies would more or less reflect a combination of gender-stratified results for a specific tumour type, which was not always the case. In lung cancer meta-analysis, for example, the risk reduction by higher selenium levels seems to be larger in data for both genders combined than it was in data for women and men separately. This underlines the influence of other sources of heterogeneity on study outcomes. Reporting of gender-stratified results in mixed-gender cohort studies, which has become increasingly common over the years, might therefore reflect other factors related to study design, such as a better evaluation of possible confounders in more recently published studies. Socioeconomic position could be one such possible confounder, leading to an overestimation of a protective effect of selenium. Several studies have found selenium levels to be positively associated with adult socioeconomic position in both men and women (Gundacker 2006; Niskar 2003).
Doubts seem therefore justified that the observed associations point to a causal relationship between selenium biomarker levels and cancer risk.
The risk increase in non-melanoma skin cancer by selenium supplements in the NPCT raises concerns about the safety of selenium yeast supplementation in both men and women.
The overall lack of a protective effect of selenium yeast against non-melanoma skin cancer is consistent with the results of the French SU.VI.M.AX trial. However, an increased risk (non-statistically significant) of non-melanoma skin cancer was only seen in women in the SU.VI.M.AX trial, not in men. These discrepancies between the results of both trials might relate to differences in intervention and characteristics of the study populations. Increased risk of non-melanoma skin cancer could be more pronounced in or restricted to high-risk populations or observable only at certain selenium levels.
Bjelakovic 2008 conducted a systematic review of antioxidant supplements for the prevention of gastrointestinal (GIT) cancers. They meta-analysed RCT data for liver cancer prevention with selenium-containing supplements and reported a protective effect in both genders (RR 0.56 (95% CI (95% CI 0.42 to 0.76)). Three of the four trials in their meta-analysis were also included in this systematic review (Li 2000; Yu 1991; Yu 1997). The remaining RCT (Li 2004b) used a combination of selenium with allitridum, a synthetic garlic extract, in the intervention and therefore did not meet our inclusion criteria. Li 2004b found a preventive efficacy of high-dose allitridum/100 μg sodium selenite supplementation on total and gastric cancer incidence in men, but not in women. No effect on liver cancer was seen in either gender. Allitridum was considered the main intervention by Li and colleagues in their paper and the contribution of selenium to the overall effect remained unclear. The more recent RCT by Qu 2007 found no effect of 50 μg selenium yeast in combination with beta-carotene and alpha-tocopherol on liver cancer mortality.
We did not calculate a summary risk estimate for the included RCTs on liver cancer in this review. The minimum number of studies required for meta-analyses was set a priori to five in the protocol because we expected large heterogeneity between interventions and study populations and a high risk of bias in studies using selenium supplements. Considering the methodological constraints of the studies, the summary risk estimate of Bjelakovic 2008 may have overestimated the preventive efficacy of selenium against liver cancer.
We could not identify RCTs that investigated other GIT cancers as primary outcomes. The NPCT reported a (statistically non-significant) reduced risk of colorectal and oesophageal cancer as a secondary outcome in the selenium group. Other studies using multi-component selenium-containing supplements found divergent results, which also indicated potential sex or gender differences (Blot 1993; Hercberg 2004; see Background)
We consider that a replication of study results in adequately conducted randomised controlled trials is necessary before the preventive efficacy of selenium supplements against liver or other gastrointestinal cancers can be further evaluated or any recommendation made regarding supplements for GIT cancer prevention. However, the indication that nutritional supplements may prevent GIT cancers in borderline nutrient-deficient populations of developing countries should not be ignored in future research. Special consideration should be given to sex or gender-specific effects, selenium specification and possible interactions with other nutrients and the presence of risk factors for gastrointestinal cancers.
The SELECT trial failed to provide evidence for the preventive efficacy of oral L-selenomethionine and alpha-tocopherol either alone or in combination against prostate cancer.
The SELECT results contrasted with findings of the NPCT on prostate cancer. This might have occurred because of an overestimation of the real effect in the NPCT, where prostate cancer was not the primary outcome. It has also been argued that selenomethionine was the wrong supplement to replicate the NPCT results (Goossens 2009; El-Bayoumy 2009) and future trials should investigate selenium yeast as the active intervention.
SELECT participants had a higher selenium level at randomisation than men in the NPCT. While the mean plasma selenium concentration was 113 to 114 μg/l in the NPCT, median serum concentration was 135 to 138 μg/l in the different study arms in SELECT. Lower prostate cancer incidence in the NPCT trial was confined to men with baseline selenium levels in the lower two thirds (below 121 μg/l). Subgroup analyses of the SELECT trial are underway to investigate a possible modification by pre-intervention selenium levels.
The SU.VI.M.AX trial, which used a multivitamin and mineral supplement containing 100 μg selenised yeast, found a reduction in the rate of prostate cancers only in men with normal PSA levels at baseline (HR 0.52 (95% CI 0.29 to 0.92)), but an increased risk in men with elevated PSA (> 3 ng/ml) (HR 1.54 (95% CI 0.87 to 2.72) (Meyer 2005). There was no interaction with serum selenium levels at baseline in this study and authors hypothesised that their multi component supplement might be beneficial in healthy men, but might promote prostate cancer development in men at higher risk. However, an interaction with baseline PSA levels was not seen in the NPCT. In SELECT, only men with normal PSA levels (less than 4 ng/ml) were eligible for participation.
Data for a variety of other cancers were reported in the NPCT trial. Notably, the results for the primary outcome of the NPCT, i.e. the incidence of non-melanoma skin cancer, received less attention in the public debate than those for secondary outcomes, especially those in favour of selenium supplementation. As they were based on a post-hoc analysis, the question of confounding and bias arose. Also the role of chance was unclear because of the uncontrolled procedures for detection of secondary outcomes. Under-representation of women in the NPCT decreased the power to detect sex-/gender-specific effects (Duffield-Lillico 2002, in: NPCT 1996) and is a concern as the highest hazard ratio for a single cancer type in the selenium group was seen for breast cancer (HR 1.89 (95% CI 0.69 to 5.14), non-statistically significant). All possible beneficial effects on cancer incidence were confined to men in this study.
In the SU.VI.M.AX trial, the selenium yeast-containing supplement did not alter breast cancer risk in female participants (Hercberg 2004). Overall, the SU.VI.M.AX trial detected no effect on total cancer incidence by its multivitamin/nutrient supplement in both genders combined (RR 0.90 (95% CI 0.76 to 1.06). Gender-stratified analyses showed a protective efficacy in men (RR 0.60 (95% CI 0.53 to 0.91), but not in women (RR 1.04 (95% CI 0.85 to 1.29)), with a reduction of cases mainly seen in gastrointestinal and respiratory cancers in men. Women had the higher baseline levels of antioxidants and study authors hypothesised that differences in outcomes between men and women may be attributable to gender differences in nutritional behaviour and consequent antioxidant status.
Also the Linxian General Population Trial (Blot 1993) found no statistically significant protective effect of the selenium containing supplement on total cancer incidence in either gender(RR 0.93 (95% CI 0.83 to 1.03); gender stratified results were not reported). Cancer deaths were marginally significantly reduced (RR 0.87 (95% CI 0.75 to 1.00)) in participants receiving the selenium/beta-carotene/vitamin E supplement.
For lung cancer, the Linxian trial found no alteration in mortality by selenium-containing study supplements (RR 0.98 (95% CI 0.71 to 1.35); Kamangar 2006).
One study currently in progress investigates the efficacy of selenium for the recurrence of early stage non-small cell lung cancer after initial surgery (RCT ECOG 2002).
The SELECT trial reported a slightly elevated risk (statistically non-significant) for diabetes mellitus type II in the selenium group (RR 1.07 99% CI (95% CI 0.94 to 1.22)). Secondary analysis of the NPCT also indicated that long-term selenium supplementation may increase the risk for developing type II diabetes mellitus (Stranges 2007).
For women, there is little evidence for lower or higher nutritional intake of selenium exhibiting a major impact on cancer risk. The only RCT results that have a low risk of bias support concerns of an increased risk of non-melanoma skin cancer by selenium yeast supplements in women who had already suffered from this disease.
For men, there is evidence for an inverse association between higher selenium biomarker levels and cancer risk. However, we cannot exclude that this effect may well be the result of other factors related to higher selenium biomarker levels than caused by selenium exposure itself. Results from two randomised controlled trials (NPCT and SELECT) have failed to provide evidence that non-melanoma skin cancer or prostate cancer can be prevented by selenium supplementation in men.
Additionally, concerns have been raised about possible toxicities from long-term intake of supplemental selenium.
Currently, regular intake of selenium supplements for cancer prevention cannot be recommended to either the selenium-replete or deficient populations.
Selenium may have different effects on specific types of cancer. The results from randomised controlled trials for the prevention of liver cancer need to be replicated in studies with a rigorous design.
Potential differential effects of sex or gender and the use of selenium supplements in populations with a high burden of specific types of cancer diseases and differing selenium exposure levels, e.g. known low nutritional selenium intake, require further examination.
Future prospective epidemiological studies as well as intervention trials should be adequately designed to detect sex or gender differences on specific types of cancer. Results of gender-stratified analyses should be reported even when statistically significant differences cannot be found.
Further research should aim to clarify why biomarkers of selenium exposure failed to reliably predict the results of RCTs with low risk of bias.
We thank Connie Hui and Mina Nishimori very much for the translation of the Chinese and Japanese papers.
We would like to thank Christine Fink and Birgit Kraus for their valuable help with data and literature management.
SOURCES OF SUPPORT
|Database||Date of most recent literature search||Search strategy||Comment|
|www.cancer.gov||4 Feb 2011||medication: selenium|
||now included in Medline database|
|Clinical Contents in Medicine (CCMed)||4 Feb 2011||selen* OR organoselen* OR natriumselen*|
|CENTRAL||Issue 1 2011||selen*|
|Cochrane Library||Issue 1 2011||selen*|
|metaRegister of Controlled Trials (mRCT, www.controlled-trials.com)||4 Feb 2011||selen AND cancer|
|EMBASE||2010 week 50||
|German Cancer Study Register: www.studien.de||4 Feb 2011||selen|
|Medline (via Pubmed)||4 Feb 2011||
|SIGLE||Oct 2004||?selen?||database discontinued in 2005|
((*) means that a ’star’ was assigned to the study for the corresponding item)
((*) means that a ’star’ was assigned to the study for the corresponding item)
(validated in cohort assessment in question 2 - number of stars was copied)
|Outcome or subgroup title||No. of studies||No. of participants||Statistical method||Effect size|
|1 Breast cancer risk (women)||7||Odds Ratio (Random, 95% CI)||1.00 [0.77, 1.29]|
|1.1 Breast cancer (all)||6||Odds Ratio (Random, 95% CI)||1.01 [0.74, 1.36]|
|1.2 Breast cancer (premenopausal)||1||Odds Ratio (Random, 95% CI)||1.10 [0.46, 2.65]|
|2 Bladder cancer risk||5||Odds Ratio (Random, 95% CI)||0.67 [0.46, 0.97]|
|2.1 all (male + female)||2||Odds Ratio (Random, 95% CI)||0.65 [0.46, 0.92]|
|2.2 male||3||Odds Ratio (Random, 95% CI)||0.82 [0.41, 1.62]|
|2.3 female||1||Odds Ratio (Random, 95% CI)||0.36 [0.14, 0.92]|
|3 Lung cancer risk (gender-aggregated data)||11||Odds Ratio (Random, 95% CI)||0.76 [0.57, 1.03]|
|3.1 incidence||10||Odds Ratio (Random, 95% CI)||0.75 [0.54, 1.03]|
|3.2 mortality||1||Odds Ratio (Random, 95% CI)||0.98 [0.41, 2.35]|
|4 Lung cancer risk (gender-disaggregated data)||11||Odds Ratio (Random, 95% CI)||0.77 [0.60, 0.98]|
|4.1 all (female + male)||4||Odds Ratio (Random, 95% CI)||0.58 [0.39, 0.86]|
|4.2 female||4||Odds Ratio (Random, 95% CI)||0.83 [0.43, 1.61]|
|4.3 male||6||Odds Ratio (Random, 95% CI)||0.88 [0.61, 1.28]|
|5 Lung cancer risk (ascending order of selenium levels)||7||1756||Odds Ratio (Random, 95% CI)||0.90 [0.69, 1.16]|
|6 Lung cancer risk||11||Odds Ratio (Random, 95% CI)||0.76 [0.57, 1.03]|
|6.1 intake||1||Odds Ratio (Random, 95% CI)||0.98 [0.41, 2.35]|
|6.2 serum or plasma||8||Odds Ratio (Random, 95% CI)||0.84 [0.66, 1.06]|
|6.3 toenail||2||Odds Ratio (Random, 95% CI)||1.05 [0.11, 10.36]|
|7 Prostate cancer risk||14||Odds Ratio (Random, 95% CI)||0.78 [0.66, 0.92]|
|8 Prostate cancer risk (by selenium measurement)||14||Odds Ratio (Random, 95% CI)||0.78 [0.66, 0.92]|
|8.1 biochemical selenium level||12||Odds Ratio (Random, 95% CI)||0.74 [0.61, 0.88]|
|8.2 estimated selenium intake||2||Odds Ratio (Random, 95% CI)||1.00 [0.73, 1.36]|
|9 Prostate cancer risk (by exposure assessment)||14||Odds Ratio (Random, 95% CI)||0.78 [0.66, 0.92]|
|9.1 intake||2||Odds Ratio (Random, 95% CI)||1.00 [0.73, 1.36]|
|9.2 serum or plasma||9||Odds Ratio (Random, 95% CI)||0.81 [0.68, 0.97]|
|9.3 toenail||3||Odds Ratio (Random, 95% CI)||0.53 [0.35, 0.81]|
|10 Prostate cancer risk (by continent)||14||Odds Ratio (Random, 95% CI)||0.78 [0.66, 0.92]|
|10.1 Europe||4||Odds Ratio (Random, 95% CI)||0.91 [0.70, 1.17]|
|10.2 North America||10||Odds Ratio (Random, 95% CI)||0.71 [0.58, 0.88]|
|11 Prostate cancer risk (by country)||14||Odds Ratio (Random, 95% CI)||0.78 [0.66, 0.92]|
|11.1 Several European countries||1||Odds Ratio (Random, 95% CI)||0.96 [0.70, 1.31]|
|11.2 Finland||2||Odds Ratio (Random, 95% CI)||1.24 [0.75, 2.05]|
|11.3 The Netherlands||1||Odds Ratio (Random, 95% CI)||0.69 [0.48, 0.99]|
|11.4 U.S.A.||10||Odds Ratio (Random, 95% CI)||0.71 [0.58, 0.88]|
|12 Prostate cancer risk (ascending order of selenium levels)||9||2112||Odds Ratio (Random, 95% CI)||0.81 [0.68, 0.97]|
|13 Stomach cancer risk||5||Odds Ratio (Random, 95% CI)||0.66 [0.43, 1.01]|
|13.1 stomach||4||Odds Ratio (Random, 95% CI)||0.65 [0.35, 1.19]|
|13.2 stomach: cardia cancer||1||Odds Ratio (Random, 95% CI)||0.47 [0.33, 0.66]|
|13.3 stomach: non-cardia cancer||1||Odds Ratio (Random, 95% CI)||1.07 [0.55, 2.08]|
|14 Stomach cancer risk (by gender)||5||Odds Ratio (Random, 95% CI)||0.66 [0.42, 1.04]|
|14.1 all (female + male)||2||Odds Ratio (Random, 95% CI)||0.75 [0.41, 1.36]|
|14.2 female||2||Odds Ratio (Random, 95% CI)||0.73 [0.12, 4.35]|
|14.3 male||3||Odds Ratio (Random, 95% CI)||0.43 [0.14, 1.32]|
|15 Colorectal cancer risk||5||Odds Ratio (Random, 95% CI)||0.89 [0.65, 1.23]|
|15.1 colon and rectal cancer||2||Odds Ratio (Random, 95% CI)||1.11 [0.50, 2.46]|
|15.2 colon cancer||3||Odds Ratio (Random, 95% CI)||0.80 [0.56, 1.15]|
|16 Colorectal cancer risk (by gender)||5||Odds Ratio (Random, 95% CI)||0.89 [0.65, 1.23]|
|16.1 all (female + male)||1||Odds Ratio (Random, 95% CI)||1.22 [0.52, 2.86]|
|16.2 female||3||Odds Ratio (Random, 95% CI)||1.06 [0.57, 2.00]|
|16.3 male||3||Odds Ratio (Random, 95% CI)||0.69 [0.42, 1.12]|
|17 Total cancer incidence and mortality||13||Odds Ratio (Random, 95% CI)||Subtotals only|
|17.1 incidence||8||Odds Ratio (Random, 95% CI)||0.69 [0.53, 0.91]|
|17.2 mortality||5||Odds Ratio (Random, 95% CI)||0.55 [0.36, 0.83]|
|18 Total cancer incidence and mortality (ascending order of selenium levels)||11||Odds Ratio (Random, 95% CI)||Subtotals only|
|18.1 incidence||6||1297||Odds Ratio (Random, 95% CI)||0.69 [0.52, 0.91]|
|18.2 mortality||5||1032||Odds Ratio (Random, 95% CI)||0.55 [0.36, 0.83]|
|19 Total cancer incidence and mortality (women)||5||Odds Ratio (Random, 95% CI)||Subtotals only|
|19.1 incidence||2||Odds Ratio (Random, 95% CI)||0.90 [0.45, 1.77]|
|19.2 mortality||3||Odds Ratio (Random, 95% CI)||0.92 [0.79, 1.07]|
|20 Total cancer incidence and mortality (men)||8||Odds Ratio (Random, 95% CI)||Subtotals only|
|20.1 incidence||5||Odds Ratio (Random, 95% CI)||0.66 [0.42, 1.05]|
|20.2 mortality||3||Odds Ratio (Random, 95% CI)||0.56 [0.38, 0.81]|
DECLARATIONS OF INTEREST
GD: None known
MZw: None known
MB: None known
MV: None known
MZe: Maurice Zeegers is the first investigator of one included observational study and one ongoing randomised controlled trial. He is second author of another included observational study.
MH: None known
CONTRIBUTIONS OF AUTHORSGD is the primary author and co-ordinator of the authors’ group and was involved in all steps of realising the protocol and the review.
MZw commented on the protocol, extracted data from papers, conducted the data analyses in STATA and commented on the review and provided a methodological perspective.
MB commented on the protocol, assisted by checking the original literature search and inclusion criteria, providing a brief and early background on urological cancer and provided feedback at various stages of the review.
MV commented on the protocol, screened search results, extracted data from papers, provided an early version of the background and feedback on the review text at various stages of the review.
MZe commented on the protocol, screened search results, extracted data from papers and provided feedback on the review text
MH commented on the protocol, screened search results, extracted data from papers, designed the Access database for data management, provided support for securing funding for the review, supported data management and commented on the review text at various stages of the review.
* Indicates the major publication for the study