Menopausal hormone therapy and national time trends in mortality: cautions regarding causal inference

    loading  Checking for direct PDF access through Ovid

Excerpt

National time trend data are often used to infer cause-and-effect relationships when changes in exposures, behavioral factors, or medications coincide with changes in health outcomes. However, before acceptance, such inferences should be subjected to scrutiny by, at a minimum, assessing long-term trajectories in outcomes and comparing with an appropriate referent group. We present an example—based on recent dramatic changes in the use of menopausal hormone therapy (HT)—of the pitfalls of using time trends for this purpose. Many researchers have speculated that the precipitous drop in use of HT by US women after the 2002 publication of unfavorable findings from the Women's Health Initiative (WHI) was responsible for an almost immediate nationwide decline in breast cancer incidence. A debate regarding causality was sparked by a 2007 report1 and continues today. Although an increase in osteoporotic fracture incidence has also been attributed to the declining use of HT,2 less attention has been paid to post-WHI trends in other outcomes linked to HT use, including coronary heart disease, stroke, and other types of cancer.
The WHI assessed oral conjugated estrogens taken with or without progestin for the prevention of chronic disease in postmenopausal women aged 50 to 79 years. Initial results of the WHI estrogen-progestin and the estrogen-alone trials were published in 2002 and 2004, respectively, establishing that such therapy neither prevents coronary heart disease nor yields a favorable balance of chronic disease risks and benefits in postmenopausal women with a mean age of 63. In 2000, nearly 40% of women in their 50s and 30% of women in their 60s used oral HT. By 2010, only 7% of women in these age groups did so.3
Using narrow time trend data to assert cause-and-effect relationships at the population level can be misleading. A more considered approach in assessing whether population-level change in an exposure could be causally related to population-level change in a disease is to examine trends over a longer period of time, and, if possible, to compare the trends in an exposed population with those in a similar but unexposed population. Thus, in the case of HT, we assessed trends in mortality among midlife US women during the 34-year period from 1980-2014 (ie, from 22 y before to 12 y after initial publication of the WHI results and the start of the nationwide plunge in HT use) and compared these trends with those in similarly aged US men (to address background changes in prevention and treatment).4,5 For breast cancer, we compared trends by age group in women. We focused on mortality rather than incidence rates to minimize the influence of changes in screening and diagnostic testing (especially for breast cancer) that accompany changes in HT prescribing.
As shown in Figure 1, declines in all-cause and cause-specific mortality in women aged 55 to 64 years began many years before the 2002 publication of the WHI findings and the precipitous drop in HT use, with pre- and post-2002 trends generally similar to those among comparably aged men. These declines likely resulted from improvements in prevention and treatment of cardiovascular disease, cancer, and other conditions. Specifically, total mortality rates in both women and men steadily declined until about 2010 and then flattened or even slightly increased (panel A). For heart disease mortality, steeper declines occurred in men than in women in the years prior to as well as the years after 2002 (panel B), and for cerebrovascular mortality, a steady and parallel decline in women and men was observed (panel C).

Related Topics

    loading  Loading Related Articles