Analyzing Sex in Biomedicine

Sex should be included in each step of the research process (Tannenbaum et al., 2016; Clayton, 2016; IOM, 2012; Beery et al., 2011; Wizemann et al., 2001).

Identify Problem and formulate hypotheses

• The first step toward excellent research is analyzing sex as a biological variable. Sex has historically been ignored as an explanatory variable; hence, literature searches may underplay its importance. Nevertheless, the absence of pilot data should encourage, rather than discourage, analyzing sex. Sex has already been linked to numerous biological and physiological sex differences and to specific health outcomes. Some resources for obtaining sex-specific literature are available online (Montgomery & Sherif, 2000; Moerman et al., 2009; Oertelt-Prigione et al., 2010; Stewart et al., 2014).
• Sex is an essential variable to consider in developing scientific hypotheses. Questions to address when developing research hypotheses include: should sex be considered a co-variate, a confounder, or an explanatory variable? How will it be possible to identify potential sex differences without overstating them? What is the biological plausibility of sex emerging as an important variable in this research question? Which intersecting factors, such as age or hormonal status, should be considered?

• Designing research requires considering the study type, study protocol, and statistical requirements, as well as practical and logistical factors.
• Study type: What type of study is needed to address the specified research hypothesis? Options include cross-sectional or longitudinal studies, observational and interventional studies, or studies with controls or without. How might potential sex differences influence these choices? One aspect to consider in longitudinal studies is how reproductive history potentially influences the cohort and if this is appropriately controlled for. If women get pregnant during the study, does this impact the potential data acquisition? If the population is of perimenopausal age, is this being taken into consideration? Will this be documented to allow for analysis? If the study is cross-sectional, do hormonal profiles play a role? When specific interventions are considered, will the populations be stratified by age and sex, and will equal distribution be ensured? Is oversampling of one group potentially warranted? If controls are being used, how are they selected? What are controls matched on?
• Study protocol: When developing the study protocol, potential sex-dependent aspects should be explicitly addressed. If hormones are going to be measured, will circadian rhythms be taken into account? Are the facilities set up to accommodate this? If metabolic differences are to be measured, will physiological sex differences be considered? Is there a contingency plan if sex differences make an intervention difficult? For example, if women’s generally smaller vessels preclude some female participants from a vascular intervention, how will this affect data collection? Statistical requirements: The fundamental requirement for statistical analysis is an adequately large sample and an appropriate number of outcomes to guarantee power for the subsequent analysis. For sex-specific research this means including adequate numbers of female and male subjects. Including both sexes does not necessarily mean doubling the group size overall, as factorial designs (progressing from a T-test to an ANOVA) reduce the sample size increase from doubling to about a 50% increase (Miller et al., 2017; Buch et al., 2019). In some cases, single sex studies might be sufficient, e.g. if a condition affects only one sex (such as prostate cancer) or if a significant amount of data has been collected in patients of one sex. In this case collecting data from the other sex using the same protocol might be the most cost-effective option.
Practical and logistical factors: Are samples of both sexes accessible? Commercially available fibroblasts tend to be male cells and many tumor cell lines are available only from the sex most affected. If a researcher wants to compare molecular patterns in breast cancer in females and males, it might be challenging to acquire samples for the latter. If patients are involved, will equal access to the study be guaranteed? Could potential differences in disease incidence limit participation or confound results?

• Sex-specific data should be collected for cells, animals, and humans. Sex definition should be pre-specified (see Term “sex”) and approaches to the inclusion of non-binary individuals should be considered (and specified). Ascertaining the genetic makeup of all cells. Analyzing sex in cell samples requires obtaining reliable information about the cell’s genetics. When the cell donor is known, this should be obvious. However, phenotype (at least in humans) does not automatically translate to XX or XY genetics (see Term: Sex) and might have to be confirmed. In addition, cell cultures frequently change their karyotype after years in culture, especially cancer cells (Duesberg et al., 1998). This affects autosomes as well as sex chromosomes. Furthermore, techniques such as the reprogramming of cells into pluripotency affects the methylation of the chromosomes and can potentially alter the inactivation patterns of the X chromosome (Lessing et al., 2013). This can impact transcriptional profiles and should be considered.

• Recording the sex of research participants is a prerequisite for sex analysis. Some granting agencies and peer-reviewed journals require reporting sex—for human, animal, and (where appropriate) organ, tissue and cell research (see Policy Recommendations). Reporting the sex of the research subject is important even in single-sex studies to allow meta-analysis, in order to prevent over-generalizing findings beyond the sex studied and to identify research gaps. When including human subjects, attention should be paid to how researchers will ask about sex (see Method: Asking about Gender & Sex in Surveys). Will sex be conceptualized as binary? When working with very large datasets, the likelihood of including individuals with differences in sex development/intersex is very high, given that these individuals range from 1:100 to 1:4500 in humans, depending on the criteria used (Arboleda et al., 2014; Huges et al., 2006) (see Term: Sex). In practice this means that some data will not fit into a pre-specified binary sex classification. Researchers should define how they want to address this issue before data collection. Data from these individuals should not be simply labelled as “outliers” and disregarded. They should be handled according to pre-planned protocols.

• Analyzing results by sex. Sex-specific analyses should be conducted and the findings reported. Providing sex-disaggregated data facilitates future meta-analysis. Women and men, for example, may require different dosages of a drug to produce a given effect (see Case Study: Prescription Drugs). Adjusting the data for baseline differences and factors that intersect with sex is a crucial step in understanding the sex differences observed. Researchers who analyzed sex in studies of cardiovascular disease, for example, identified sex differences in arterial plaque formation: women tend to develop diffuse plaques, whereas men more often develop localized plaques (von Mering et al., 2004). This difference has ramifications for the design of stents (see Case Study: Heart Disease in Diverse Populations). Nevertheless, researchers should also be aware that differences that exist within groups of females and males (or women and men) might be larger than differences between the groups. Both biological and sociocultural factors differ substantially between individuals over their lifetime and will shape the expression of many traits. These include profound changes associated with reproductive biology (such as at puberty and, in women, throughout the menstrual cycle, during pregnancy and at menopause), with transitioning from one sex to another and with aging. Take, for example, height. On the whole, men are taller than women. Large within-group variation and differences between countries, however, complicate this sex difference. Many women are taller than many men. Researchers need to consider that limiting analysis to a comparison of means will potentially limit the real-world applicability of their results.

  Collecting and reporting factors intersecting with sex. Women, men and gender-diverse individuals differ by age, lifestyle (e.g. diet, physical activity, use of tobacco, alcohol, other drugs, etc.), socioeconomic status, and other gendered behaviors and variables. It is important to consider which of these might be relevant to the proposed research.

  Sex as an explanatory variable or a confounder. In many cases, sex will play a role in both explaining the results and acting as a potential modulator for other causal pathways. Nevertheless, most studies will allow only for the description of correlation rather than causality. This should be taken into account when conducting the analysis and reporting results. Drawing a causal diagram could make the underlying assumptions explicit and sharpen the analysis.