Response to “misclassification does not explain increased cardiovascular risks of calcium supplements”

Abstract

We thank Bolland and colleagues for providing some further details on the adverse event data that they claim supports their hypothesis that calcium medication may increase the risk of myocardial infarction (MI). It is unfortunate that it has taken so long to provide some extra data to allow a proper evaluation of their hypothesis. Consequent on this brief report of their most promoted findings several further questions arise. First, the original publication of the Record trial, which contributed 162 of 254 (64%) of the cases to the patient-level data, does not reveal or mention adverse events other than fracture. Bolland and colleagues state, ‘‘Another study supplied self-reports of hospital admissions and cause of death from death certificates. Each event was then independently adjudicated by two physicians blinded to the treatment allocation of the participants.. . .’’ As far as we can see, this statement suggests that the hospital discharge data was not adjudicated; rather, the patient self-report of the hospital admission was adjudicated. If this is correct then 45% of this data does not meet the accepted criteria of verification and should be excluded. Second, their reanalysis of the trial level MI data now reports an RR of calcium supplements of 1.27 (95% confidence interval [CI], 1.01–1.59), p1⁄4 0.04. However, if the same analysis is undertaken using the verified events independent of self-reported events from Prince and colleagues and Lewis and colleagues, the RR is 1.23 (0.98–1.54), p1⁄4 0.08, showing that small misclassification errors due to ascertainment bias can lead to reporting of an erroneous increased cardiovascular risk from calcium supplements. If, indeed, 45% of cases were not properly adjudicated, erroneous excess events occurring in participants taking calcium supplementation may have resulted in their finding of a significant increase in MI in the calcium group for the patient-level analysis. Third, in relation to the controversial 2011 meta-analysis, Bolland and colleagues now state, ‘‘the hazard ratio for myocardial infarction with calcium was 1.29 (95% CI, 1.10– 1.52), p1⁄4 0.002.’’ However, the pooled RR using the published data from the 2010 meta-analysis and the actual total published Women’s Health Initiative (WHI) data provides a pooled RR of 1.09 (95% CI, 0.97–1.23), p1⁄4 0.13. The reason that this reanalysis of all the WHI data is so important is that it does not use the post hoc decision to split the WHI data into two groups using data only from the patients who had low calcium intake at baseline. The premise for this separation of the data was claimed to be an interaction between calcium supplements and cardiovascular events; however, this interaction only became significant after the unusual and presumably post hoc decision by the authors to remove silent MIs from their analysis. This removal is particularly surprising given they are the same condition; we are unaware of any valid reason to exclude these events from the analysis. Furthermore, the claimed interaction is diametrically opposite to the direction of the interaction in their previous meta-analysis, which is reported to show an increased risk of MI in patients on high calcium intake prior to the start of the study. Next, the original numbers in the published WHI data differ from the numbers reported by Bolland and colleagues in their 2011 meta-analysis, which reported an additional four more MI cases in the calcium supplementation group and one fewer case in the placebo than the published WHI data. Although Bolland and colleagues are unconcerned about the ‘‘minor’’ differences between the datasets, we consider these discrepancies potentially important, especially given the clinical significance of their hypothesis and the nature of the restrictive and varying study designs used by Bolland and colleagues in their publications on this topic. In conclusion, ascertainment bias due to confounding by gastrointestinal complaints in participants on calcium supplementation is possible. This, in addition to the numerous concerns raised previously about the analytical methods used by Bolland and colleagues, serves to increase the uncertainty regarding their hypothesis that supplemental calcium increases the risk of MIs. Thus, at the current time we are of the view that this hypothesis should be discarded.