To the Editor: We believe that the approach used in the meta-analysis of mortality in randomized trials of antioxidant supplements by Dr Bjelakovic and colleagues1 erred in several important ways, probably resulting in biased conclusions.
First, the Linxian General Population Nutrition Intervention Trial (NIT)2 was misclassified as a “trial with high risk of bias.” This double-blind placebo-controlled trial of 29 584 persons contained all the attributes described by the authors as defining trials with low risk of bias: more than 60% of the target population was enrolled and computer-randomized, and participant characteristics were virtually identical across all supplement groups (no selection bias); all pill bottles were masked throughout the trial (adequate allocation concealment and blinding); all participants were visited monthly to assess adherence and ascertain end points (no performance or detection bias); and follow-up and endpoint ascertainment were essentially complete, with only 0.2% lost to follow-up (no attrition bias). Indeed, the methods used in this trial were identical to those used in the concurrent Linxian Dysplasia NIT,3 which the authors classified as a “trial with low risk of bias.”
Taylor PR, Dawsey S. Antioxidant Supplements and Mortality. JAMA. 2007;298(4):400-403. doi:10.1001/jama.298.4.401-b