Institute of Medicine respondsBMJ 2011; 342 doi: http://dx.doi.org/10.1136/bmj.d4046 (Published 28 June 2011) Cite this as: BMJ 2011;342:d4046
- 1Bone Metabolism Unit, Creighton University Medical School, Omaha, NE, USA
- 2Tufts University School of Medicine, Boston, MA, USA
The Institute of Medicine (IOM) criticised inconsistencies and misrepresentations of data in the meta-analysis of vitamin D and fall prevention by Bischoff-Ferrari and colleagues.1 2 The authors replied to the IOM,3 and we respond to them as members of the IOM committee.
In assessing the dose-response relation between vitamin D dose, serum 25-hydroxyvitamin D concentration, and risk of a fall the authors opted to compare subgroups of studies above or below a cut off point of vitamin D dose and serum 25 hydroxyvitamin D concentration selected by “visual inspection”—that is, in an entirely data driven way. They found significant differences in two subgroups (in a meta-regression framework) and interpreted them as evidence for an inverse relation. However, analyses using nearby cut off points are not significant. The IOM’s reanalysis of the authors’ data using vitamin D dose as a continuous variable showed no significant relation. Furthermore, trying to establish a “threshold” between falls and serum 25 hydroxyvitamin D concentration is too simplistic when non-comparable assays are performed over 12 years.
Bischoff-Ferrari and colleagues acknowledge that they included a trial conducted by Broe et al in violation of their stated eligibility criteria.3 4 They maintain that it was appropriate to do so because it had a high quality fall assessment. However, the study by Graafmans et al also violated the inclusion criteria since falls was only added as an end point in a cohort study two years after randomisation to a fracture study.5 Four of the five negative results relating to a threshold on “low dose” vitamin D and falls were by Broe et al and Graafmans et al.
Figure 3 in Bischoff-Ferrari and colleagues’ meta-analysis is misleading2: it was not simply misinterpreted by the Institute of Medicine. The figure uses the layout of a typical meta-regression plot and shows a “trend line” that resembles a meta-regression summary line. Using the same data, the IOM found that the meta regression analysis on falls was not significant (figure⇓).1
The internal inconsistencies of the paper and the selectivity of the dose-response analyses justified the IOM’s reanalysis. Properly powered double blind studies of vitamin D on falls are needed but no more meta-analyses.
Cite this as: BMJ 2011;342:d4046
Competing interests: None declared.