A reanalysis too far?BMJ 2011; 342 doi: https://doi.org/10.1136/bmj.d3538 (Published 07 June 2011) Cite this as: BMJ 2011;342:d3538
- 1School of Mathematical Sciences, North Terrace Campus, University of Adelaide, Adelaide, SA 5005, Australia
- 2Endocrine and Metabolic Unit, Royal Adelaide Hospital, Adelaide SA 5000, Australia
One of the remarkable features of the paper by Bolland and colleagues on the risk from calcium supplementation and vitamin D is its failure to draw attention to the hazard ratio of 0.84 in favour of calcium supplements for deaths from all causes in the “any personal use of calcium” subgroup; this is the most significant of all comparisons in table 2 (P=0.01).1 To suppose that high doses of calcium reduce the death rate is no more fanciful than to imagine that personal use of calcium protects against the alleged adverse cardiovascular effect of prescribed calcium. A more prosaic explanation is that analysis of subgroups for many different end points is likely to unearth a few small P values.
To include the results from the “no use of personal calcium” subgroup in any sort of follow-up meta-analysis also seems inappropriate. Conversely, any meta-analysis of this subject should surely take account of the study by Lewis et al of 1460 women with a mean age of 75.1 years at baseline who were treated with calcium or placebo for five years and followed up for four and a half years.2 The hazard ratio (calcium supplementation v placebo) for atherosclerotic vascular disease was 0.92 (95% confidence interval 0.74 to 1.15; NS). For adjudicated myocardial infarction in particular, the hazard ratio was 1.0 (0.54 to 1.84).3
Cite this as: BMJ 2011;342:d3538
Competing interests: None declared.