The Million Women Study: potential biases do not allow uncritical acceptance of the data

Abstract

Background The findings in the Million Women Study (MWS) have been interpreted by some as providing final and definitive evidence that hormone therapy (HT) (estrogen alone or estrogen plus a progestin) increases the risk of breast cancer.

Methods A critical review of the MWS evidence.

Critique It is established that women who attend for routine mammography are not representative of the population at large: HT is more common among attenders, as is the occurrence of breast cancer. The MWS cohort comprised women who were invited to enroll when they were scheduled for mammography. Especially in that setting, HT users could more commonly have enrolled because of enhanced anxiety about breast cancer risk, women already aware of breast lumps could also selectively have enrolled – and women who both used HT, and who were aware of breast lumps, could have been the most motivated to participate. Thus, it was possible that the selective detection of as yet undiagnosed breast cancer among HT users could have accounted for the small associations between HT and breast cancer observed.

The data supported that possibility. The incidence of breast cancer at or soon after recruitment was higher than in the population at large; HT was more common than in the population at large; the time intervals from recruitment to the diagnosis of breast cancer, or death, were implausibly short; and, in clinical and pathological terms, discordance in the risk of breast cancer among current users of HT, and among those who had recently stopped could most plausibly be explained by detection bias.

Additional defects in the MWS, pointed in correspondence, included failure to properly specify zero time; and failure to record discontinued HT use or cross-overs, with resultant misclassification of the time, type, and duration of HT. The MWS report also contained multiple errors – sufficient to raise doubts about the care with which the study was carried out, and reviewed. Finally, findings in the MWS were discordant with those reported in other studies.

Conclusions Based on the MWS findings, it is not possible to distinguish between bias and causation as alternative explanations for the observed associations, and the conclusion that it has been established that HT increases the risk of breast cancer is not justified.