点击显示 收起
Department of Epidemiology
University of North Texas Health Science Center
3500 Camp Bowie Boulevard
CBH Suite 355
Fort Worth, TX 76107
E-mail: rojha{at}hsc.unt.edu
Dear Sir:
Lappe et al (1) recently conducted a randomized clinical trial (RCT) that evaluated the effect of calcium or calcium + vitamin D supplementation on cancer incidence as a secondary outcome. The RCT was originally designed to compare the effect of calcium or calcium + vitamin D supplementation on skeletal status and calcium economy in postmenopausal women. The authors concluded that calcium + vitamin D supplementation was superior to placebo in overall cancer prevention in postmenopausal women. However, insufficient information and limitations of the study may require a more subdued interpretation of the results than originally purported.
The authors did not provide key information, such as demographic and baseline risk factor distributions, on the randomized groups. Such information would have allowed for further evaluation of the randomization procedures and confirmed equivalent distributions between comparison groups. Furthermore, it would have been useful to have provided detailed information regarding loss to follow-up in the treatment group, compliance characteristics by treatment group, and details regarding the per protocol analysis for both treatment groups. Treatment-specific compliance is of particular interest because of potential differential misclassification of the exposure (intervention). For example, even 10% noncompliance by the calcium + vitamin D participants who did not develop cancer may overestimate (bias away from the null) the protective effect of calcium + vitamin D supplementation by 21% [reported crude risk ratio (RR) for study period, assuming complete follow-up: 0.42; 95% CI: 0.21, 0.83; crude RR after correction for 10% misclassification: 0.53; 95% CI: 0.27, 1.06). However, we offer only theoretical speculation that cannot be evaluated without additional information. Therefore, we recommend that future investigations adhere to the standardized CONSORT guidelines for reporting RCT results to avoid the exclusion of such necessary information (2).
The authors used logistic regression to calculate effect estimates after citing the potential for violating the proportional hazard assumption required for standard time-to-event analysis. However, the logistic regression model assumed complete follow-up despite evidence to the contrary and presented effect estimates based on cumulative incidence. Logistic regression models in this scenario may present biased estimates because of a failure to account for various forms of loss to follow-up (3). The authors did not explicitly show a violation of the assumptions of a proportional hazards model, and it is unclear why a time-dependent Cox proportional hazard model with relevant product terms was not used to verify the assumption in question (3). Model selection (whether time-dependent or a standard proportional hazard) could have proceeded accordingly if the assumption were indeed violated. Furthermore, even a standard Cox proportional hazard model would have provided more precise estimates than a logistic regression model, regardless of the proportional hazard assumption, because it would still allow for censoring (3), but could be interpreted as an average rather than a time-dependent hazard (risk) ratio.
Insufficient duration of follow-up may also contribute to the uncertainty of the findings. Results from an investigation of vitamin D and breast cancer risk in postmenopausal women from the Iowa Women's Health Study were also recently released (4). Robien et al (4) reported a lower breast cancer risk for the first 5 y of follow-up in postmenopausal women with a total vitamin D intake >800 IU/d than in a reference group with a total vitamin D intake <400 IU/d (RR: 0.66; 95% CI: 0.46, 0.94). However, the risk estimate converged to null after 10 y of follow-up, and the higher vitamin D intake category eventually exceeded the risk of the reference category after 15 y of follow-up (RR: 1.23; 95% CI: 0.86, 1.75) (4). The findings by Robien et al indicate the importance of the duration of follow-up and raise a legitimate concern regarding the potential for a similar trend if Lappe et al continued follow-up for a similar duration. Furthermore, the well-established phenomenon of enhanced calcium absorption in the presence of vitamin D (5) raises uncertainty regarding the suggested independent action of vitamin D as a causal mechanism for cancer prevention (1). The results provided by Lappe et al are not capable of differentiating between molecular synergism and independent nutrient action. Future studies of calcium and vitamin D should account for biological interaction to better address causality.
The findings by Lappe et al warrant considerable speculation and require further evaluation to confirm their validity. Insufficient information regarding baseline characteristics, loss to follow-up, and compliance limit potential inferences. Furthermore, the inefficiency of logistic regression compared with Cox proportional hazards regression for effect estimation from data with loss to follow-up necessitates conservative interpretations. It may be argued that the pervasive consumer availability of nutritional interventions and the profound public health interest generated by such investigations warrant an even greater degree of scrutiny than do pharmaceutical interventions dispensed by licensed professionals.
ACKNOWLEDGMENTS
No conflicts of interest were declared.
REFERENCES