Response to “calcium supplements and cardiovascular risk”
Article first published online: 23 MAR 2011
Copyright © 2011 American Society for Bone and Mineral Research
Journal of Bone and Mineral Research
Volume 26, Issue 4, pages 900–901, April 2011
How to Cite
Lewis, J. R., Calver, J., Zhu, K., Flicker, L. and Prince, R. L. (2011), Response to “calcium supplements and cardiovascular risk”. J Bone Miner Res, 26: 900–901. doi: 10.1002/jbmr.329
- Issue published online: 23 MAR 2011
- Article first published online: 23 MAR 2011
- Accepted manuscript online: 4 JAN 2011 12:31PM EST
We thank Bolland and colleagues for engaging in a dialog on an important issue. In essence, their letter raises two important and contentious issues in clinical trialing, the selection of safety endpoints and the power of the study.
Endpoint selection: As noted by a recent BMJ review of the topic of endpoint selection, the approach adopted by Bolland and colleagues risks the accusation of “cherry picking.”1 Their response was as follows. “In fact, no cherry picking took place. In 1996–1997 the protocol was finalized. . . . The specified composite endpoint was the combination of angina, chest pain, myocardial infarction, or sudden death. . . . In 2005 the trial statistician reported an adverse effect of calcium supplements on self-reported cardiovascular events. . . . We adopted the composite endpoint of myocardial infarction, stroke, or sudden death for the adjudicated events.”2 Interestingly, in their 2008 article,3 when they applied the original composite criteria, there was no adverse effect [relative risk (RR) = 0.94, 95% confidence interval (CI) 0.72–1.24], and even after application of the second set of criteria, there was no effect (RR = 1.47, 95% CI 0.97–2.23). However, that article contained the highlighted statement, “What this study adds: Healthy older women randomised to calcium supplementation showed increased rates of myocardial infarction” based on a reported RR of 2.12 (95% CI 1.01–4.47, p = .047). In our view, in light of the four separate endpoints eventually tested, myocardial infarction, stroke, sudden death, and the composite, the p value for statistical significance should have been .00625. We leave it to the reader's judgment as to whether the removal of angina and chest pain, the inclusion of stroke including hemorrhagic stroke, and the final highlighted selection of myocardial infarction without adjustment of the p value constitute cherry picking.
Next, the question arises as to why angina and chest pain were excluded from the 2005 analysis when in such epidemiologic studies it is usual to explore the effects of supplementary endpoints related to the main effect. The authors state that these endpoints were excluded because “adjudication of episodes of angina or chest pain was not possible.” However, in other studies, including our own, there was no particular difficulty in comparing the patient's self-report with coded discharge data for angina and chest pain because these are recognized ICD classification terms. Interestingly, in the Bolland and colleagues 2008 article, the self-reported RR of angina in the calcium-treated group was 0.71 (95% CI 0.50–1.01) and for chest pain was 1.08 (95% CI 0.54–2.16).
Bolland and colleagues question our choice of atherosclerotic vascular disease (ASVD) rather than their chosen endpoints.3, 4 The selection of ASVD was designed to capture all events associated with atherosclerosis, which we understand to be the putative pathologic processes that calcium therapy may exacerbate. ASVD was selected as the primary endpoint because, unlike Bolland and colleagues, we remain concerned by the problem of multiple comparisons that require an adjustment of the p value using the Bonferroni correction. We went on to consider subsets of ASVD in Table 2. The results for adjudicated myocardial infarction separately were 5-year RR = 1.00 (95% CI 0.54–1.84) and cerebrovascular disease excluding haemorrhagic stroke RR = 1.10 (95% CI 0.68–1.78) with a combined RR = 1.04 (95% CI 0.70–1.55), which is very similar to the ASVD endpoint used in our article (unadjusted RR = 1.01, 95% CI 0.78–1.30), validating our prespecified approach.
Power: Bolland and colleagues question the power of our study. This is an important issue and is in part the reason for selecting ASVD rather than the more selective endpoints used by Bolland and colleagues. They go on to criticise our methodology by suggesting that the time-to-first-event approach may miss a second event that occurs later. This betrays a serious misunderstanding of standard statistical methodology for event rates that count individuals not events. For each separate endpoint analysis, all other events are ignored. Thus, for example, in the subset analyses of ischemic heart disease, any previous events affecting other diagnostic categories were ignored. Bolland and colleagues have previously demonstrated some confusion on this point because in their 2008 article they included event data as well as per-person data.
Bolland and colleagues then go on to question event rates in the Perth study. According to our calculations, if events are restricted to myocardial infarction, stroke, and sudden death, then the overall 5-year adjudicated event rate in the Auckland study was 7.5% and 6.3% in the Perth study, which did not include hemorrhagic stroke and sudden death. For reasons that are not clear to us, despite these very similar event rates, Bolland and colleagues calculate that if vascular events are extrapolated from 2001–2002 Australian statistics,5 the Auckland study would have 30% to 35% of participants with vascular events. The use of these data raise concerns about the authors' understanding of clinical trialing methodology in that it fails to take into account a number of important factors. First, the endpoints used in the 2001–2002 Australian data include all vascular diseases (I00–I99), which encompass hemorrhagic stroke, hypertensive diseases, pulmonary heart diseases, and many other categories not considered to be due to ASVD. Second the authors again confuse events rather than participants with an event in their calculation. Third, the 2001–2002 Australian study was population-based, whereas clinical trials participants benefit from the well-known Hawthorne effect.
In conclusion, the data in our study should be considered in conjunction with the data of Hsia and colleagues,6 who reviewed the risks of calcium supplementation with vitamin D using the findings of the Women's Health Initiative. They concluded that there was no increase in cardiovascular risk. The Hsia and colleagues study had cardiovascular events as a prespecified endpoint, detailed subgroup analyses, and adjudicated events. Thus we reiterate our previous statement: “This trial provides compelling evidence that calcium supplementation of 1200 mg daily does not significantly increase the risk of atherosclerotic vascular disease in elderly women.”7
- 5Australian Institute of Health and Welfare. Heart, Stroke and Vascular Diseases: Australian Facts 2004. AIHW Cat. No. CVD 27. Canberra: AIHW and National Heart Foundation of Australia (Cardiovascular Disease Series No. 22), 2004.