“Recently the results of a meta-analysis on antioxidant supplements generated some worrying headlines in the public press all over Europe: ‘Meta-analysis calls antioxidants in question’, ‘Caution: Vitamins!’, ‘Supplements may shorten your life’. Some months later, a balanced analysis of the study is appropriate.
The review of Bjelakovic and co-workers (1), recently published in the Cochrane Database of Systematic Reviews 2008, is largely identical to the publication by the same authors in the Journal of the American Medical Association from 2007 (2). This is demonstrated by the case numbers: in 2007, 232,606 patients from 68 studies were included in the meta-analysis, and in 2008 232,550 patients from 67 studies were included.
The study has serious methodological defects and therefore holds no statistical significance. Some of the main flaws are as follows:
1) It is a recognized methodological standard in carrying out meta-analyses that the selection of the studies must be performed blind, i.e., it must be guaranteed that the results of the study may not influence the decision as to whether the study is included in the meta-analysis or not. The requirement applies equally to the classification of the studies into studies of good and poor quality made by Bjelakovic during the analysis. At any rate, according to the authors’ statements, 679 studies were excluded but Bjelakovic and co-workers did their work unblinded. This methodological defect alone degrades this meta-analysis to a mere statement of the authors’ opinion.
2) The overwhelming majority of the meta-analyzed studies did not have the objective of reducing mortality, meaning that the problem was formulated retrospectively. The analysis of such post hoc hypotheses could in the best case (which in any event does not exist here) generate a signal that a risk may exist which ought to be investigated in a study specifically designed for such a purpose.
3) The results of the meta-analysis by Bjelakovic and co-workers are based on publications of studies. Anyone who has experience here knows how study publications often come about, and that important information is often lacking, e.g., whether in fact all the patients originally included in the study were adopted into the assessment. Hence the gold standard of meta-analyses is still that the original data of the studies are acquired and the analyses are performed in a consistent manner using these original data. This is how, for example, the greatly renowned research group of Professor Sir Richard Peto at Oxford, which introduced meta-analysis into medicine about 20 years ago, proceeds to this day. The gathering of original data is costly and time-consuming, but worthwhile in the interest of scientific research and not misleading pseudo-science and as with the authors of the Bjelakovic meta-analysis.
4) The results of the meta-analysis show very small risk increases – the greatest is just 16% (for vitamin A), and for the whole antioxidants group they lie between 2% and 4%. These relative risks cannot be reliably interpreted, particularly also in view of the defects described above. Many researchers assume that risk increases below 50% in publication-based meta-analyses are not scientifically meaningful.
5) The medieval physician Paracelsus already knew it and today it is common knowledge that the dose or quantity determines whether something is a poison or not. However, certain meta-analysts such as Bjelakovic have not taken this into account and thrown all the dosages and treatment periods together. Thus, for example, dosages of vitamin A varied between 1,333 and 200,000, and dosages of vitamin E between 10 and 5,000 International Units. The treatment periods lay between one day (!) and 12 years. The least that should have expected from the authors would be a subdivision of the studies on the basis of the dose and the treatment period. It is incomprehensible how such nonsense could leave anyone’s desk in the form of a scientific manuscript at all.”
Munich, June 2008