Bias in Observational Studies of the Association between Menopausal Hormone Therapy and Breast Cancer

Background: During the period 1985-2000 the breast cancer incidence rates increased 50% in the age group invited to mammography screening in Norway and Sweden. Simultaneously, use of hormone replacement treatment therapy (HT) increased 5 times. Several influential observational studies showed that HT was associated with 50% to 100% increased risk of breast cancer and most for those using combined (estrogen plus progestin) hormone replacement therapy (CHT). In contrast, the randomized WHI trial reported that CHT increased the risk by 10% for those not having previously used hormones and 24% when including previous users in the analyses. In another randomized trial, estrogen use only was not associated with any increased risk at all. After the WHI trial was published in 2003, use of HT dropped 70% within 5 years in Norway and Sweden while breast cancer rates were essentially unchanged. After 2008, HT use has dropped further and breast cancer incidence rates have started increasing again. The study objective is to calculate and to explain potential bias in the observational study design. Methods and Findings: Here we use data from the randomized WHI trial and analyze these data as done in the observational studies to calculate the magnitude of the potential biases in the observational study design. Time varying effect of hormones and categorization of the follow-up time may increase the hazard ratio for long-term users from 1.10 to 1.48. Selective retrospective reporting of hormone use may further increase the hazard ratio to 1.68. Conclusions: We suggest that the mechanism causing higher hazard ratio of breast cancer (compared to the observational studies) is the time-varying effect of CHT on the breast cancer risk and selective retrospective reporting of hormone use. Other risk factors for the increase in breast cancer risk in the age group 50-69 years should be considered, for example, overdiagnosis.