- Click here for explanation of classification scheme.

To describe Bayesian meta-analysis methods for combining data from clinical trials, using recent studies that analyzed the relationship between vitamin E dosage and all-cause mortality.

• Men and non-pregnant women (19 years of age or older)
• Use of vitamin E alone or in combination with other vitamins or minerals
• Randomized control trial intervention studies and follow-up of longer than one year; 10 or more deaths in the trial
• Study dates from 1966 to 2005.

• Not within included dates (1966 to 2005)
• Men and women that did not match the age criteria established
• Trials that violated the search criteria.

• Intervention: Vitamin E doses, ranging from 15 IU to 5,000 IU
• Outcome investigated: All-cause mortality
• Populations included: Adults from multiple countries (France, China, Finland, Italy, United States, United Kingdom, Australia, Canada, Israel and Germany)
• Search procedures: Medline search from August 2004 to December 2005, using the MeSH terms vitamin E and alpha tocopherol. A previously published meta-analysis study by Miller et al was also included. Terms searched on Medline and Cochrane database included vitamin E, antioxidant vitamins, alpha tocopherol, tocopherol and clinical trials.
• Search limits: Original research, review articles and previous meta-analyses that were randomized controlled trials
• Blinding used: The majority of the studies were double-blinded
• Statistical analysis: Data collection included the number of patients and deaths, percentage of men, use of other vitamins or minerals, mean age and length of follow-up. The authors used a Bayesian hierarchical model for data synthesis and Markov chain Monte Carlo computational techniques in the analysis. Bayesian model averaging is a technique that is designed to help account for the uncertainty inherent in model selecting. By averaging over many different competing models, this technique incorporates model uncertainty into conclusions about parameters and prediction. Quality of studies was not assessed except for standard use of strict inclusion criteria.

Information Abstracted from Articles

• The study characteristics included the description of the study population, N, sex, mean age, follow-up in years, deaths in the experimental vs. control group, vitamin E dosage, experimental procedures, use of other antioxidants and the experimental design
• Bayesian hierarchical model was used for data synthesis and Markov chain Monte Carlo computational technique. The Bayesian hierarchical modeling allows for different values for different death rates and P-values, for different studies. The authors used a variety of dose-response relationships to contribute to the final conclusion by averaging the results based on the probability that each model is correct. 

Analytical Methods

Inferences about dose-response were made by considering the mortality rate to be dependent on the dose of vitamin E within the trial and on trial-specific covariates. The covariates included: vitamin E alone or in combination with other vitamins or minerals, mean age, sex, and follow-up duration. 

Timing of Measurements

All of the trials occurred between 1966 and 2005.

Dependent Variable

All-cause mortality.

Independent Variable

Vitamin E intake.

• Articles identified: 154
• Number and type of studies reviewed: 124
• Articles included in meta-analysis: 22
• Sample size of studies: 160 to 39,876 subjects.

Characteristics of Study Participants

• Age: 19 years and older
• Ethnicity: Many countries involved
• Other relevant demographics: Multiple age groups, multiple states of health, smokers and non-smokers
• Anthropometrics: Varied in age and health state
• Location: Multiple locations.

• Using the updated data and adjusting for subject age, sex, follow-up duration and vitamin E used in combination with other vitamins or minerals, the authors found the total posterior probability for all covariate-adjusted models to be less than 0.025. This indicates that there is little evidence that these covariates matter in regards to any effect of vitamin E.
• The Bayesian approach provides little additional insight into the effect of vitamin E on all-cause mortality, as was concluded in other analyses. A variety of models were used to study the effect of vitamin E dosage. The Bayesian meta-analysis of data from 22 clinical trials concludes that vitamin E is unlikely to affect all-cause mortality and that this is true regardless of dosage.

• Vitamin E intake is unlikely to affect mortality, regardless of dosage
• The Bayesian meta-analysis presented here is ideal for incorporating disparate sources of variability, including trial effect and model uncertainty.

University/Hospital:

Dept. of Biostatistics, University of Texas, MD Anderson Cancer Center Division of Internal Meedicine