点击显示 收起
1 From the Fred Hutchinson Cancer Research Center, Seattle, WA
2 Presented at the conference "Multivitamin/Mineral Supplements and Chronic Disease Prevention," held at the National Institutes of Health, Bethesda, MD, May 15–17, 2006. 3 Supported by NIH award CA53996. 4 Address reprint requests to RL Prentice, Division of Public Health Sciences, Fred Hutchinson Cancer Research Center, 1100 Fairview Avenue North, M3-A410, PO Box 19024, Seattle, WA 98109-1024. E-mail: rprentic{at}whi.org.
ABSTRACT
Multivitamin-multimineral (MVM) supplements are widely used in the United States, often in the hope of reducing the risk of cancer, cardiovascular disease, or other chronic disease. This article assesses the potential of randomized controlled trials and epidemiologic cohort studies for yielding reliable information on the effects of MVMs on chronic disease. A brief review of the available literature on MVMs in relation to incidence and mortality rates from prominent cancers and cardiovascular diseases is also provided along with a discussion of needed research. Specifically, the strengths and weaknesses of epidemiologic cohort studies and randomized controlled trials are summarized and discussed in the context of single-vitamin supplements when both types of studies are available. Recent review articles that include an assessment of MVMs in relation to cancer and cardiovascular disease are updated to provide a summary of available data. Few randomized controlled trials and few cohort studies of MVMs that are directly pertinent to cancer or cardiovascular disease are available. The data are not compelling concerning a role for MVMs in preventing cancer or cardiovascular disease morbidity or mortality, although some interesting leads merit further evaluation. Investigators responsible for cohort studies that assessed MVMs should be encouraged to report available data on MVMs and chronic disease. Depending in part on the results of such additional reports, a full-scale randomized controlled trial of well-selected MVMs in women may be warranted on public health grounds.
Key Words: Cohort study • multivitamin • multimineral • randomized controlled trial • research needs • study design
INTRODUCTION
This report on clinical trials and observational studies to assess the chronic disease benefits and risks of multivitamin-multimineral (MVM) supplements begins with a description of the strengths and weaknesses of cohort studies and randomized controlled trials (RCTs). This is followed by a brief discussion of 2 important settings for which substantial data are available from both observational studies and RCTs, namely, 1) β-carotene and lung cancer and other diseases and 2) vitamin E and cardiovascular disease (CVD) and cancer. These settings illustrate the importance of the study design and permit insights into the strategies needed to yield additional reliable information on MVMs and chronic disease.
A brief summary of reports from RCTs and cohort studies is then presented for both cancer and CVD. This review highlights the potential of further reports from established cohorts to guide the MVM research agenda; some information on MVM and total supplement use in the Women's Health Initiative cohorts is provided to illustrate such potential. This report concludes with a summary of available data on MVMs in relation to cancer and CVD and with a listing of suggested additional research activities to enhance the science of this important public health area.
RESEARCH DESIGNS FOR THE STUDY OF MVM SUPPLEMENTS AND CHRONIC DISEASE
Most available data on MVM supplements and chronic disease are from cohort or case-control studies.
Observational studies
Cohort studies, which involve the identification of a roster of persons from a population of interest who are followed over time to ascertain disease events, constitute a mainstay epidemiologic approach to the study of the association between various exposures and behaviors and subsequent chronic disease incidence or mortality. Strengths of the cohort study design include the potential to assess and compare the multiple MVM preparations used by members of a study cohort and to assess both short- and long-term health effects. The latter may be possible even in a cohort that has been fairly recently assembled if a sufficient number of cohort members have used MVMs for some years before study enrollment. This feature contrasts somewhat with RCTs, for which the randomized comparisons pertain to the typically shorter periods of time since trial enrollment.
Compared with case-control studies, cohort studies can be expected to avoid recall bias that may arise if persons who experience a study disease recall MVM usage histories or confounding factors differently from persons without a disease. Cohort studies can also be expected to avoid selection bias as may arise in case-control studies if the study is unable to enroll a representative sample of cases and controls from the study population. Finally, cohort studies that are large and of substantial duration can typically accumulate large numbers of events for chronic diseases of interest and hence may yield precise contrasts between MVM users and nonusers, at least for MVM preparations used by a noteworthy fraction of the study cohort.
Perhaps the most important weakness of the cohort study design arises from the potential for residual confounding in estimated MVM and chronic disease associations. It has been well-documented that, compared with nonusers, supplement users are more likely to be white, well educated, nonsmoking, consumers of a low-fat and high-fiber diet, and regular exercisers. Such factors are difficult to measure and fully accommodate in data analysis, even if recognized. Supplement users are screened more frequently for diseases (eg, breast and prostate cancer), which may lead to differential outcome ascertainment for supplement users and nonusers. White et al (1) illustrate these issues in a cohort study of dietary supplement use in western Washington.
An additional cohort study issue concerns the reliability of MVM assessment. Observational studies typically use self-administered questionnaires to obtain information on current and past MVM use, often including attempts to collect information on frequency and duration of use. For example, Patterson et al (2) noted that supplement intake data obtained with a brief self-administered questionnaire or a telephone interview were variably correlated (correlation coefficients ranged from 0.1 for iron to 0.8 for vitamin C) with data derived from supplement bottle labels. This suggests that MVM associations of practical interest may be severely attenuated or otherwise obscured by measurement error. Also, to the extent that health effects are related to the total intake of nutrients included in MVMs, association studies may also need to address the thorny issues of dietary nutrient consumption assessment to interpret MVM versus disease associations.
RANDOMIZED CONTROLLED TRIALS
RCTs, which involve the random allocation of a suitable set of study participants to active or placebo MVM groups, with subsequent follow-up to ascertain disease events, obviate most of the cohort study limitations just mentioned. The randomized assignment of study subjects implies a statistical independence of MVM versus placebo assignment for all baseline potential confounding factors, whether or not such factors are accurately measured or are even recognized, and also independence of randomized group assignment with prerandomization disease screening practices. Furthermore, unlike the cohort study, the validity of chronic disease comparisons between randomization groups does not depend on MVM assessment, dietary nutrient assessment, or the assessment of other disease risk factors. These assessments are used in the RCT setting, but only for the interpretation of the principal randomized contrasts in relation to such factors as study power, adherence to study pills, or interactions with baseline characteristics. Also, RCTs frequently provide a clinical context that can ensure equivalent clinical outcome ascertainment between randomization groups, particularly if the trial is placebo controlled and effectively blinded.
RCTs also have their own limitations and tend to be comparatively expensive so that few full-scale trials can be afforded. Study power is also typically a concern. Because of cost, it is usual to design chronic-disease-prevention RCTs to have adequate power (eg, 80–90%) to identify an intervention effect of plausible size overall, implying limited power for important subset comparisons and a sensitivity of overall study power to departures from design assumptions concerning baseline disease rates (eg, healthy volunteer effect, especially if stringent eligibility criteria are imposed), medication adherence assumptions, average follow-up durations, and, perhaps most importantly, departure from the basic intervention effect assumption. There are typically few pertinent data to aid in the specification of a time course or magnitude of an intervention effect on chronic diseases of interest, especially for lifestyle interventions (eg, nutrient consumption modifications, physical activity patterns) for which disease risks may have been substantially influenced by decades of other lifestyle patterns and where disease processes may be underway for decades before surfacing clinically.
RCTs are usually conducted with only a few years of average follow-up (eg, 10 y), which may be short in relation to the major effects of an intervention. Even within such a time frame, it may be difficult for a trial to continue to its planned termination because the benefits and risks of an intervention may have different time courses, and RCT-monitoring committees may have limited tolerance for continuation in the face of a current adverse effect versus a potential future benefit. Finally, the randomized comparisons in RCTs are typically limited to the evaluation of a single intervention or a small number of interventions, whereas a large variety of MVMs are being used by substantial numbers of persons in the United States.
Additional considerations for both cohort studies and RCTs
Both cohort studies and RCTs rely on volunteers for a long-term research project that may require a lot of time. Hence, volunteers may not be representative of the populations from which they are drawn. This feature generally does not affect the reliability of study contrasts but rather may restrict the overall findings to a subpopulation. The presentation of results in a manner that stratifies on key baseline factors can partially overcome this limitation, but power may be limited.
An important data analysis issue is the handling of multiple testing issues. This can be a concern in RCTs if investigators filter through many clinical outcome definitions or many subsets of a study cohort and report study findings (eg, significance levels) without regard to multiplicity issues, but this usually does not happen. The study design usually involves advance specification of primary and secondary outcomes and perhaps a few subsets of principal interest; principal findings are based on comparison among randomized groups, with formal recognition of the study monitoring process.
In contrast, observational study reports usually make little or no allowance for multiplicity issues and may consider many clinical outcomes, multiple subgroup analyses, and periodic assessment of associations of interest and, importantly, may classify MVMs in multiple ways according to nutrient content, dose, duration of use, or other factors. Statistical methods are not available to adequately adjust significance levels and CIs for such an unstructured analytic approach.
In summary, RCTs and cohort studies have complementary strengths and weaknesses and share some challenges. In general, much care is needed in observational studies to minimize potential biases such as residual confounding bias and measurement error bias. RCTs, because of their expense and logistical challenges, typically require careful preliminary development that may draw on multiple sources of information, including observational studies, trials with intermediate outcomes, and basic science research. An RCT may be justified when the preliminary research is strong and the public health implications are sufficiently great. When a cohort study or RCT is underway, care must be taken to collect clinical outcome data in a manner that is unrelated to the MVM and other exposures of interest, and an analytic plan should be in place that addresses multiplicity issues and that prespecifies major aspects of data analysis and reporting.
SOME LESSONS FROM SINGLE-NUTRIENT STUDIES
Settings in which data from both a cohort study and a clinical trial are available provide an opportunity to assess the importance of the design and analysis issues mentioned above and to identify opportunities for strengthening either type of study. Many observational studies have reported a lower risk of cancer among persons whose self-reported diet includes relatively large amounts of foods rich in β-carotene and other carotenoids and relatively high intakes of fruit and vegetables. Review papers of these associations have noted that cohort and case-control study findings are particularly consistent for lung cancer (3).
At least 10 prospective studies used blood concentrations of β-carotene. These studies reported a lower risk of lung cancer, heart disease, other cancer, and all-cause mortality among persons having relatively high blood concentrations of β-carotene. For example, on the basis of an average 8.2-y cohort follow-up, Greenberg et al (4) reported an estimated total mortality relative risk of 0.62 (95% CI: 0.44, 0.87) for persons in the highest versus the lowest quartile of blood β-carotene.
The consistency and magnitude of the β-carotene relative risk trends stimulated 3 large lung cancer prevention intervention trials in the 1980s. The Alpha-Tocopherol Beta-Carotene Cancer Prevention (ATBC) Trial (5) randomly assigned 29 133 male Finnish smokers to supplemental β-carotene (20 mg/d) or placebo and to -tocopherol (vitamin E) versus placebo in a factorial design. After 5–8 y of follow-up, the estimated lung cancer relative risk for β-carotene supplementation was 1.18 (95% CI: 1.03, 1.36) based on 876 incident lung cancers during the planned follow-up. There were no significant differences for other cancers, whereas the total mortality relative risk was 1.08 (95% CI: 1.01, 1.16).
This unexpected finding of an elevated lung cancer relative risk was followed by the results from the Carotene and Retinol Efficacy Trial (CARET) (6). This US trial randomly assigned 18 314 smokers, former smokers, and asbestos-exposed persons to a combination of 30 mg β-carotene/d plus 25 000 IU retinol/d versus placebo. This trial stopped 2 y early after an average 4-y follow-up and reported β-carotene and retinol relative risks of 1.28 (95% CI: 1.04, 1.57) for lung cancer based on 388 incident cases, 1.26 (95% CI: 0.99, 1.61) for CVD mortality, and 1.17 (95% CI: 1.03, 1.33) for total mortality with no significant differences for other cancers.
In comparison, the Physicians' Health Study (7) was conducted in 22 071 male US physicians, aged 40–84 y, of whom only 11% were smokers and 39% were former smokers at baseline. Subjects were randomly assigned to receive 50 mg β-carotene on alternate days in a factorial arrangement with aspirin supplementation. After an average 12 y of follow-up, the β-carotene and placebo groups showed no significant differences in the incidence of lung cancer, other cancers, or CVD or in total mortality. For example, the lung cancer comparison involved 82 incident cases in the β-carotene group and 88 in the placebo group.
The lack of support from intervention trials for a protective effect of β-carotene on lung cancer raises questions about the reliability and specificity of the observational studies leading up to these trials. The human diet is a complex mixture of foods and nutrients with many highly correlated elements. Furthermore, measurement error in dietary self-reports may combine with these high correlations to much reduce the reliability of observational study associations. Confounding by nondietary characteristics that distinguish frequent fruit and vegetable consumers and dietary supplement users from other persons is also an important issue for the interpretation of observational studies of micronutrients, such as β-carotene, and disease risk. Some such characteristics, for example, those related to physical activity patterns, may also be quite difficult to measure and accommodate in data analysis.
An explanation for the elevation of lung cancer in persons supplemented with β-carotene in the ATBC Trial and the elevation in lung cancer in persons supplemented with β-carotene and retinol in CARET is still lacking. Many mechanistic possibilities have been proposed, including the inhibition of absorption of other nutrients in persons consuming large amounts of β-carotene and the possibility of a prooxidant effect of large doses of β-carotene in persons with lungs damaged by cigarette smoking or asbestos exposure. Mayne (8) provides an overview of related mechanistic research for lung cancer and other diseases. She also notes some marked differences in bioavailability of the β-carotene preparations used in the various trials as a contributor to differences among trial results.
The lessons from the β-carotene studies for the study of MVMs in relation to chronic disease include the possibility that nutrients in MVMs may have surprising health consequences, even if generally regarded as safe, and that study subject exposures (eg, cigarette smoking) or characteristics could be important in determining the balance of benefits and risks. Also, the dose of nutrients could be highly influential; hence, extrapolation of results from one dose to another, and presumably from one agent to a mixture of agents, should be avoided.
As a second example, consider vitamin E supplementation in relation to CVD and cancer. In presenting the results from their well-conducted intervention trial of 600 IU vitamin E on alternate days with 39 876 women aged 45 y, Lee et al (9) summarized the motivating literature by writing that "several large cohorts observed decreased CVD rates among individuals who self-selected for higher intakes of vitamin E though diet and/or supplements" and "several observational studies, particularly case-control studies, reported reduced rates of cancer among persons who self-selected for higher antioxidant intakes." Their RCT, however, found no overall benefit for major cardiovascular events or cancer for this vitamin E supplementation over an average 10.1 y of follow-up.
For the designated primary outcome, a major cardiovascular event defined as any of nonfatal myocardial infarction, nonfatal stroke, or cardiovascular death, the hazard ratio was 0.93 (95% CI: 0.82, 1.05). This result leaves some uncertainty as to whether a CVD benefit exists, but the study, even though large and with 999 primary outcome events, was not powered to detect a small (eg, 7%) reduction in risk. Note that even a difference as small as 7% could translate to a meaningful risk reduction among women who adhered to the supplement and did so for some decades.
Lee et al (9) presented active versus placebo comparisons for 16 outcome categories. One category, total cardiovascular death, had a nominal significance level of <0.05. This illustrates that multiplicity issues accompany RCTs, which if overlooked could allow a vitamin E enthusiast to claim that a benefit had been established.
Substantial observational data and RCT data exist for calcium and vitamin D supplementation in relation to fractures and colorectal cancer (10, 11), selenium supplementation in relation to skin and other cancers (12), and folate supplementation in relation to CVD (13). Space does not permit a discussion of these settings, but they add to the following summary comments: 1) For individual supplements or simple supplement combinations, RCTs provide limited evidence of benefit and some evidence of harm (eg, β-carotene and lung cancer, calcium and vitamin D and kidney stones). 2) Dose and milieu may be important. It seems unwise to extrapolate from dietary associations to supplement effects or from single-supplement effects to MVM effects. 3) The reliability of observational studies in the settings mentioned above remains uncertain.
BRIEF REVIEW OF COHORT STUDIES AND RCTS OF MVMS, CANCER, AND CVD
Neuhouser et al (14) recently provided an excellent summary of observational and clinical trial data on dietary supplements in relation to cancer. For MVM supplements, case-control studies of cancer at various sites (eg, colon, esophagus, stomach, oropharynx, breast, cervix, bladder, prostate, and skin) suggested that benefits may exceed risks. Corresponding cohort study results, however, were mostly consistent with no association; in the single RCT, a combination of 15 mg β-carotene plus higher-dose multivitamins with minerals did not affect total, esophageal, or stomach cancer incidence or mortality among persons with esophageal dysplasia in Linxian, China (15). Some detail on the cohort studies examined in this review and some recent cohort study data are provided in Table 1. In the large Cancer Prevention Study II cohort, there was no suggestion of total or prostate cancer mortality benefit associated with MVM use and some suggestion of harm. Similarly, primarily from the Harvard cohorts (Nurses' Health Study, Health Professionals Follow-Up Study, and Women's Health Study), evidence for an MVM benefit on site-specific cancer is weak, with only some modest suggestions of benefit for colon and breast cancer in the Nurses' Health Study.
View this table:
TABLE 1. Cohort studies of multivitamin-multimineral (MVM) use and cancer1
The cohort study data on MVM use and CVD were reviewed by Morris and Carson (29). As shown in Table 2, the available cohort study data provide little support for lower CVD mortality among MVM users. Data from the Nurses' Health Study, however, suggest a possible benefit for coronary heart disease incidence defined as nonfatal myocardial infarction or coronary heart disease death.
View this table:
TABLE 2. Cohort studies of multivitamin-multimineral (MVM) use and cardiovascular disease1
One additional RCT deserves mention. As shown in Table 3, the SU.VI.MAX study (Supplementation en Vitamines et Mineraux Antioxydants; 33) of a combination of vitamin C, vitamin E, β-carotene, selenium, and zinc in 13 017 women and men in France reported a hazard ratio for total cancer incidence of 0.90 (95% CI: 0.76, 1.06) over an average 7.5 y of follow-up with a hazard ratio that was lower in men than in women.
View this table:
TABLE 3. Prevention trial of combinations of MVMs (SU.VI.MAX)1
SUMMARY AND RESEARCH RECOMMENDATIONS
Results to date are not compelling concerning a role for MVMs in preventing morbidity or mortality from cancer or CVD. The SU.VI.MAX trial provides some intriguing leads concerning a specific vitamin and mineral combination in relation to total cancer incidence that, if substantiated, could have public health implications. Cohort study data on MVMs in relation to these major diseases remain fairly limited; other cohort study research groups having quality data on MVM use should be encouraged to report associations with cancer, CVD, and other clinical outcomes. Some existing cohort study reports are of interest, including possible associations of MVMs with coronary heart disease and perhaps colon and other cancers in women.
As an illustration of the potential of existing cohorts to contribute to the observational data on MVMs, consider the Women's Health Initiative clinical trial and observational study in 68 132 and 93 676 postmenopausal women, respectively, across the United States. Information on current dietary supplement use was obtained from product labels at baseline and subsequently at 1 y in the clinical trial and at 3 y in the observational study; this information was linked to nutrient content through a dietary supplement database. These cohorts have an average follow-up of 9 y at the time of this writing and carefully adjudicated outcome data are available for a broad range of clinical outcomes. Multivitamin use, with or without minerals, at baseline was somewhat higher (41.5%) in the observational study than in the clinical trial (32.0–35.6% depending on the component 34). MVM use at baseline was more common among white women than among racial and ethnic minority women and slightly more common among older than younger postmenopausal women. For example, in the observational study, use of multivitamins with or without minerals at baseline was 43.8% for white, 32.0% for Asian and Pacific Islander, 28.4% for Hispanic, 26.8% for black, and 28.0% for native/American Indian women (35). Corresponding usage rates were 28.5% for women aged 50–9 y, 42.5% for women aged 60–69 y, and 43.4% for women aged 70–79 y at baseline (35).
The Physicians' Health Study II is well along (average follow-up > 7 y) in studying a particular MVM versus placebo in 14 641 US physicians as a component of a factorial design with vitamins C and E and initially with β-carotene. This study can be expected to add substantially to the scientific data pertinent to MVMs and chronic disease.
Because of unexplained adverse effects of relatively high-dose single-agent supplementation (eg, β-carotene and lung cancer) and the likelihood of complex dose effects and interactions among vitamins and minerals in relation to chronic disease risk, among other reasons, it does not seem advisable to project MVM benefits and risks from single vitamin and mineral studies, which, at any rate, have shown few clear effects on chronic disease. Given the widespread use of MVMs in our society, which is often motivated by chronic disease prevention hopes and suggestive observational study data, it may be appropriate to soon consider the merits of a full-scale trial of a well-selected MVM in women, perhaps depending on findings from additional large-scale cohort study reports.
ACKNOWLEDGMENTS
I thank Marian Neuhouser and Emily White for helpful input.
The author had no financial or personal interest in any company or organization sponsoring the research, including advisory board affiliations.
REFERENCES