5.55) for all cause mortality compared with a dietary calcium
intake of between 600 and 999 mg/day. The same comparison
among women with use of any type of calcium containing
supplement, yielded a multivariable adjusted hazard ratio of
1.51 (0.91 to 2.50), whereas the hazard ratio among non-users
of calcium containing supplements was 1.17 (0.97 to 1.41).
Thus, among women with a high dietary intake of calcium, the
addition of calcium supplements increased the risk of death in
a dose dependent fashion. The synergy index for the interaction
between a high dietary calcium intake and calcium tablet use
was 4.87 (95% confidence interval 1.11 to 21.32).
Vitamin D intake did not significantly modify the associations
between calcium intake and the rate of deaths from all causes,
cardiovascular disease, or ischaemic heart disease (results not
shown).
Discussion
In this study of women in the Swedish mammography cohort,
a high calcium intake (>1400 mg/day) was associated with an
increased rate of mortality, including death from cardiovascular
disease. The increase was moderate with a high dietary calcium
intake without supplement use, but the combination of a high
dietary calcium intake and calcium tablet use resulted in a more
pronounced increase in mortality. For most women with lower
intakes we observed only modest differences in risk.
Strengths and weaknesses of the study
Strengths of our study include the population based prospective
design, study size, and repeated measurements of calcium intake,
as well as a large number of potential covariates. Date and cause
of death were traced through national healthcare registries and
deterministic record linkage, permitting complete ascertainment
of the outcomes. The accuracy of classification of causes of
death in the cause of death registry and diagnoses in the national
patient registry are high.43 Furthermore, we adjusted for several
important covariates (for example, smoking, socioeconomic
status, physical activity, nutrients other than calcium, educational
level, and comorbidity), but residual confounding remains a
possible limitation. The lower age adjusted rates of death from
all causes and cardiovascular disease among women with a high
total calcium intake were largely explained by their use of
dietary supplements (table 2), a variable considered in the
multivariable models. Other health related covariates, including
a healthy diet and level of physical activity contributed to a
lesser degree. People who use dietary supplements have, on
average, a healthier lifestyle and a lower risk factor profile for
cardiovascular disease44 and not considering this might distort
the risk estimates. Moreover, the low proportion of women who
took prescription calcium tablets (6%), containing a four times
higher dose of calcium than in regular multivitamin dietary
supplements, made it difficult to detect modestly strong
associations with calcium tablet use specifically. Dietary
assessment methods are prone to several limitations, affecting
both the precision and accuracy of the measurement. In larger
studies, a food frequency questionnaire is used to assess the
habitual intake of diet, and a recent review concluded that it
was a valid method for assessing dietary mineral intake,
particularly for calcium.45 The food frequency questionnaire
may, to some extent, overestimate calcium intake,25 which was
also indicated by our validation. A further limitation in our study
is the use of age standardised portion sizes and not actual
individual portion sizes. By use of our calibrated analysis of
calcium intake, we none the less tried to avoid some
misclassification of study participants. By using repeated
measurements on dietary intake we increased the accuracy of
the measurement but may also have introduced bias using time
dependent Cox regression models. Indeed, after using only
baseline data and also after performing the marginal structural
model analyses, we no longer observed an increased mortality
for women with low calcium intakes or a high total calcium
intake. Without being causally linked to death, a low calcium
intake could therefore be viewed as a marker of frailty or a less
healthy behaviour associated with a higher mortality. There are,
however, also theoretical drawbacks of our causal inference
model. It is sensitive to correct model specifications and indeed
renders estimates with lower precision than ordinary Cox’s
regression.46 47 It is worth emphasising that traditionally obtained
estimates, such as those from Cox’s regression, would not
generally agree with estimates from marginal structural models
even when there is no confounding.48 Irrespective of analytical
approach, the observational study design precludes conclusions
about causality, and cautious interpretations of the results are
therefore recommended.