Challenges and future directions in menstrual cycle research

The menstrual cycle is an observable indicator of female reproductive function. Menstrual cycle length and menstrual bleeding can be symptoms of underlying pathology. Despite the utility of menstrual cycle information for guiding clinical practice, there is surprisingly little data on menstrual cycles from large population-based studies. The landmark study by Treloar (1) of 2700 women and approximately 250,000 menstrual cycles gathered over a maximum of 30 years, has never been replicated and still provides useful estimates of menstrual cycle length and variability over time. The article by Najmabadi and colleagues in this month’s Paediatric and Perinatal Epidemiology (2) adds to this literature by combining data from several cohort studies to examine patterns in menstrual cyclicity and menstrual bleeding, and most importantly, follicular and luteal phase length. An important contribution of Najmabadi’s study, which the Treloar study could not address, is the description of phase lengths and variability in over 500 women. It is interesting to note that prolonged follicular phases were more frequent in this study than prolonged menstrual cycles, indicating that prolonged cycles are not a perfect reflection of changes in ovulation. The variability in follicular and luteal phase length was substantial: within-woman follicular phase length variability was greater than 7 days in 42% of women and within-woman luteal phase length variability was more than 3 days in 59% of women.

This study is part of an expanding literature. Technological advancements and user adoption of cell phone applications (apps) for tracking menstrual cycles has created a potentially rich data source for menstrual cycle research. This technology may provide the breakthrough that has been sorely needed in this research field. Other recent studies have capitalised on app data to examine menstrual cycle characteristics. And while the size of these datasets is astounding, some including hundreds of thousands of women, some of the same issues exist in those studies, as with traditional epidemiologic studies, for example, selection bias or challenges for generalizability. The objective of this commentary is to describe methodological challenges for menstrual cycle research, whether those challenges differ for traditional epidemiologic research versus menstrual cycle app studies, and goals for future research through either study design.

Selection bias and challenges for generalizability can occur through several mechanisms. For cycle tracking apps, each may have a different level of accessibility (free vs not free), a different operating system requirement (which may exclude older phones), or a different user-base with a different prevalence of health conditions or a unique demographic profile. For example, one large app study includes a sample which is mostly White (3), while another app study does not describe the racial/ethnic distribution of the sample at all (4). The Najmabadi study, which did not employ an app, also includes a sample that is mostly White (2). Cycle length may differ by race or ethnicity (5), and associations between exposures and menstrual cycle length may differ within racial groups. Similarly, participant age may differ among research participants versus non-participants, as both apps and epidemiologic studies often exclude adolescent women. Another mechanism that may lead to selection bias or challenges to generalizability is volunteerism, women who volunteer for a study may differ from the target study population. For example, they may be more likely to have irregular cycles as they may be more interested in understanding their menstrual cycles. Associations between exposures and menstrual cycle length may differ among women with irregular cycles. Women with irregular cycles may also be more likely to use a cycle-tracking app for the same reason – they are seeking to better understand their cycles. On the other hand, women with irregular cycles may be less represented in a given research study, as some studies require “regular cycles” to enroll. Relatedly, some studies depend on fertility awareness methods (2,4,6) which may be designed specifically for women with regular, predictable menstrual cycles. Again, both app-based and traditional epidemiologic studies face the same challenge of identifying who is “volunteering” to participate in their research in order to identify the potential for selection bias or a lack of generalizability. As readers, and as reviewers, our challenge is to determine who the results of a given study apply to; who is represented by a given research study.

An additional source of potential selection bias is that menstrual cycle studies often focus on women trying to conceive a pregnancy (for example, references 2 and 3). This may be because fertility is the primary outcome of interest, or because menstrual cycles can only be studied in women who are not using hormonal birth control. A limitation of this design is that the menstrual cycle patterns that are observed are selected. The number of contributed cycles will depend upon how long it takes women to conceive. Women with fertile cycles conceive, and stop contributing, but infertile women continue trying, and the number of cycles each woman has is informative (also known as “informative cluster size”). To the extent that menstrual cycle characteristics and fertility are correlated, this will lead to selection bias in the factors that are associated with menstrual cycle characteristics. This may also lead to selection bias in the factors associated with menstrual cycle variability. Variability can only be assessed in women with more than one cycle, and the more cycles a woman has, the less fertile she is likely to be. Standard longitudinal data analysis methods do not account for informative cluster size.xThe selection of women who are attempting to become pregnant is an issue for both traditional and app-based studies, however, menstrual cycle tracking apps hold great promise for expanding the research beyond the typical volunteer population. Apps are available to all women, regardless of their pregnancy intentions, and information regarding both birth control use and pregnancy intentions could be collected within the app in each cycle in order to identify all naturally occurring menstrual cycles.

In addition to selection bias and generalizability, another challenge for interpreting menstrual cycle research is that measures of menstrual cycle endpoints differ across studies. Again, regardless of whether an app is used, differences in measurements limit comparisons across studies. For example, studies often rely on women’s self-identification of a menstrual period onset to define the beginning of a menstrual cycle (as in 2,3). While this is likely to be a valid measure for most cycles, some fraction (anywhere from 5 to 36% (7)) will have intermenstrual bleeding (bleeding that occurs between menstrual periods (8)), which may influence her ability to recognise a menses, and may lead to inaccurate cycle length measurements. On the other hand, applying a standard definition of menses that does not rely on the participant’s identification, based on the number of observed bleeding days in a daily diary for example, also has limitations. Missing data can cause a menses to not be identified, or women may simply forget to record a day of bleeding which leads to a menses not being identified, or some women may have lighter periods, and on average the intensity of bleeding does not achieve the chosen definition of menses. Differences in definitions translate to challenges in comparing menstrual and intermenstrual bleeding across studies, which then limits our overall understanding of intermenstrual bleeding.

Measurement error in determining the onset of menses, is related to another measurement issue: bleeding intensity. Studies typically capture menstrual bleeding by using subjective measures, such as “light” versus “heavy,” which is defined by the participant’s perspective (as in (2)). This is likely to be quantitatively inaccurate as 40% of women with heave menstruations consider them normal, and 14% of women with mild to moderate menstruations consider them heavy (9). Some studies include more objective measures of bleeding, based on the products used and the degree of saturation, but objective measures are unlikely to capture the impact of the bleeding on quality of life. Menstrual bleeding intensity is an important aspect of women’s health – it can interfere with activities of daily life and cause missed school or workdays. Thus, the measurement of menstrual bleeding intensity may differ across studies depending on whether a quantitative or qualitative measure was preferred. These definitions are fundamental to the interpretation of the study findings and can limit comparisons across studies.

The study by Najmabadi et al. can be interpreted in terms of the methodological considerations described here. The study provides novel, detailed data regarding follicular and luteal phase length. The study findings may not be generalizable to other racial or ethnic groups, to women with irregular cycles, to women who do not use fertility awareness-based methods or who are not intending to conceive a pregnancy. The study focused on women with regular menstrual cycles with menses identified by the participant among women using fertility awareness methods, all of which might limit misclassification of cycle length due to intermenstrual bleeding while also decreasing the observed prevalence of intermenstrual bleeding. Bleeding parameters were obtained through subjective measures and may not have captured quantitative differences in bleeding by age or parity.

In total, menstrual cycle research is starting to receive the attention it deserves, and new and innovative technologies will continue to improve the feasibility of large and detailed datasets that will facilitate the identification of menstrual cycle endpoints. While the size of app-based studies can improve the power to detect influences on the menstrual cycle, there are still methodologic issues that must be considered that are not necessarily solved by the increased sample size. However, with thoughtful design, including validation sub-studies, detailed characterization of the women and cycles collected, and widespread access, menstrual cycle tracking apps have great potential as a research tool for overcoming past study design limitations.