Patient level pooled analysis of 68 500 patients from seven major vitamin D fracture trials in US and EuropeBMJ 2010; 340 doi: https://doi.org/10.1136/bmj.b5463 (Published 13 January 2010) Cite this as: BMJ 2010;340:b5463
All rapid responses
Erratum - Reduction in hip fracture risk was statistically significant, but underestimated in the CaD trials subanalysis due to coding error
To the editor of BMJ
Dear Dr Godlee,
Regarding our paper "Patient level pooled analysis of 68 500 patients
from seven major vitamin D fracture trials in US and Europe", published in
BMJ 2010 (BMJ. 2010 Jan 12;340:b5463.).
In subsequent work using the DIPART dataset we have become aware of a
coding error that led to a modest underestimation in the BMJ paper of the
reduction in hip fracture risk for vitamin D supplements given with
For one of the studies (the Porthouse study) there was a coding error
when importing the treatment allocation variable, so that participants
allocated to no treatment were considered in the analysis as being on
calcium and vitamin D supplements and vice versa. The results have been
strengthened slightly by correcting this error and the reduction in hip
fracture risk is now statistically significant. All results pertaining to
vitamin D given without calcium are unaffected.
In figure 3 we reported a hazard ratio, HR for the any fracture
outcome of 0.95 (95% confidence interval, CI 0.90-1.01) for all trials,
0.92 (0.86-0.99) for CaD trials and 1.01 (0.92-1.12) for D trials. These
results were all unaffected by correcting the treatment status coding for
the Porthouse trial.
In figure 4 we reported a HR for the hip fracture outcome of 0.97
(0.86-1.10) for all trials, 0.84 (0.70-1.01) for CaD trials and 1.09 (0.92
-1.29) for D trials. After correcting the coding error, the HR for all
trials combined was 0.96 (0.85-1.09) while the effect in the CaD trials
became statistically significant with an HR of 0.83 (0.69-0.99). Again, D
trials were not affected by the miscoding.
In table 2 we reported a subanalysis that found a HR for the 20ug
vitamin D dose given with calcium of HR 0.95 (0.80 to 1.14) for any
fracture, HR 1.30 (0.88 to 1.92) for hip fracture and HR 0.97 (0.48 to
1.98) for vertebral fracture. The correct risk hazard ratios are HR 0.99
(0.84 to 1.19), HR 1.19 (0.81 to 1.75) and HR 0.99 (0.49 to 2.04). All
other hazard ratios in the table are correct as stated and there are no
alterations to the conclusions of the subanalysis.
On behalf of the DIPART authors, I wanted to make you aware of this
coding error. The change in effect size is very small and the conclusions
still hold as reported. However, we understated the treatment effect on
hip fractures, which is in fact statistically significant. We would be
grateful if these corrections could be published as an erratum to the
paper as soon as practicable. If desired, we can supply revised tables or
the erratum could simply consist of the text in bold above
Competing interests: Grant / Research support from Novartis, Nycomed, Amgen, Merck Speakers Bureau with Nycomed, Merck, Eli Lilly.
Thank you for raising the question of adverse effects of calcium and
vitamin D. I do not believe it would make scientific or clinical sense to
pool endpoints between studies, unless these endpoints were collected in a
reasonably comparable fashion.
As can be seen from the original publications, the frequency of
visits to the clinic varied considerably between the studies as did the
use of laboratory monitoring.
Because of the differences between studies, I would not recommend
inferences regarding endpoints other than fractures or death, irrespective
of how much we as clinicians would like to answer these important
questions. I agree that future clinical trials should be performed in such
a way that data of high quality on adverse events can be collected and
adjudicated for subsequent pooling with other studies.
Author of the paper in question; please see paper for full disclosures.
Competing interests: No competing interests
The DIPART group (1) claims in the papers’ strengths and limitations section that studying adverse events was beyond the scope of the analysis. As we have already stated in response to the recent meta-analysis on vitamin D and falls by Bischoff-Ferrari et al. (2, 3) negligence of a medications’ adverse effects within a meta-analysis is inacceptable. The results are meaningless unless a balanced assessment of benefits and harms is presented helping clinicians and patients to make informed treatment decisions. Balanced reporting is especially important in preventive options demonstrating only moderate clinical effects like vitamin D and in patients with polypharmacy as the population in question.
Considering the huge effort that has been undertaken, we do not believe that it should not be possible to perform at least an exploratory analysis on adverse event data in such a large sample as summarised by the DIPART group.
Again and hoping that constant dripping will wear away the stone, we insist that meta-analyses with a restricted view on beneficial outcomes should be abandoned. The BMJ is requested to set the benchmark publishing only balanced reviews taking into account the methodological suggestions on assessment and reporting of harms in effectiveness reviews (4).
1)The DIPART (vitamin D Individual Patient Analysis of Randomized Trials) Group. Patient level pooled analysis of 68 500 patients from seven major vitamin D fracture trials in US and Europe. BMJ 2010; 340: b5463
2)Bischoff-Ferrari HA, Dawson-Hughes B, Staehelin HB, Orav JE, Stuck AE, Theiler R, Wong JB, Egli A, Kiel DP, Henschkowski J. Fall prevention with supplemental and active forms of vitamin D: a meta-analysis of randomised controlled trials. BMJ 2009; 339: b3692
3)Meyer G, Köpke S. Vitamin D and falls. Information on harm is missing. BMJ 2009; 339: b4395
4)Chou R, Aronson N, Atkins D, Ismaila AS, Santaguida P, Smith DH, Whitlock E, Wilt TJ, Moher D. Assessing harms when comparing medical interventions: AHRQ and the Effective Health-Care Program. J Clin Epidemiol 2008; Sep 25 [Epub ahead of print]
Competing interests: No competing interests