Menopause Live (05 April, 2010) from IMS

A recent population-based, case-control study conducted from 2003 to 2006 has reported a reduced risk of breast cancer among women using bisphosphonates [1]: 2936 incident cases of invasive breast cancer identified in the Wisconsin cancer registry and 2975 controls identified from drivers’ license lists were interviewed. Multivariate methods were used to control for confounding by age, obstetric history, family history, menopausal status, age at menopause, hormone use, mammography, osteoporosis, smoking, and ‘height change’. Participation rates for cases and controls were 74% and 67%, respectively. Among women who used bisphosphonates within the preceding year (current use), the odds ratio (OR) for bisphosphonate users was 0.67 (95% confidence interval 0.51–0.89); among women who last used bisphosphonates more than a year previously, the OR was 0.91 (95% confidence interval 0.50–1.64). For durations of use of 3–12, 13–24, and ≥ 25 months, the ORs were 0.78, 0.69 and 0.63, respectively (trend p = 0.01). The risk reduction was limited to women who were not obese (body mass index < 30 kg/m2); among obese women, the risk was increased, but not significantly.

The investigators concluded that their ‘confidence in these results [was] strengthened by the large size of the study, the population-based sampling, and … consideration of important confounders’.

Comment
This was a carefully designed study and the findings raise the hypothesis that the use of bisphosphonates may reduce the risk of breast cancer. However, the confidence of the investigators was overstated, the evidence was limited, and the conclusions should have been tentative. 

To begin, in this study the estimated OR of 0.67 for users of bisphosphonates means that, for non-users, the OR was the inverse of 0.67, or 1.49 (1/0.67), or small [2]. In observational research, when an OR is small, it is virtually impossible to discriminate among bias, confounding and protection as alternative explanations [2]. The investigators did not consider the possibility of bias, yet, given failure to recruit 26% of the cases and 33% of the controls, selection bias could have accounted for the seeming risk reduction if cases of bisphosphonate-exposed breast cancer tended not to participate, or vice versa for the controls. In addition, the failure to observe a reduced risk among obese women could have been due to detection bias: allowance was made for the number of mammograms received in the preceding 5 years, but the sensitivity of mammography is limited, and it is most limited among obese women. 

With regard to confounding, socioeconomic status was not controlled, even though ‘cases were more likely than controls to have a … college degree’. It is also questionable whether allowance should have been made for osteoporosis and ‘height change’, since those factors are indications for bisphosphonate use. 

If bias or residual confounding was present, confidence in the results was not ‘strengthened by the large size of the study’: given bias, the only effect of a large study, relative to a small one, is to confer statistical significance to a biased OR estimate. In addition, the duration–response effect was again modest, with the OR only declining from 0.78 (3–12 months) to 0.63 (≥ 25 months); it could readily have been due to bias or confounding, and have been significant because of the large sample size. 

Finally, under a protective hypothesis, it is biologically implausible that bisphosphonates would no longer confer a reduced risk of breast cancer within 1 year of stopping use.

Samuel Shapiro
Department of Public Health and Family Medicine, University of Cape Town, South Africa

References
1. Newcomb PA, Tretham-Dietz, Hampton JM. Bisphosphonates for osteoporosis treatment are associated with reduced breast cancer risk. Br J Cancer 2010;102:799-802. Published March 2, 2010.
http://www.ncbi.nlm.nih.gov/pubmed/20160722
2. Shapiro S. Commentary: Bias in the evaluation of low-magnitude associations: an empirical perspective. Am J Epidemiol 2000;151:939-45.
http://www.ncbi.nlm.nih.gov/pubmed/10853631