Including a gender and sex analysis in scientific research can open the door to discovery and innovation.
By Melissa De Witte
Whether it’s designing equipment or developing drugs, scientists often fail to consider how gendered preferences, biases and assumptions can lead to unintended consequences.
According to Stanford historian Londa Schiebinger, it’s time for science to catch up.
To address the lag, Schiebinger and a group of scholars have published a paper in Nature that discusses why a sex and gender perspective matters in science and how the scientific community can make research more inclusive – all while fostering growth and innovation.
“Integrating sex and gender as variables in research, where relevant, enhances excellence in science and engineering,” said Schiebinger, who is the John L. Hinds Professor of History of Science in Stanford’s School of Humanities and Sciences. “The operative question is how can we harness the creative power of sex and gender analysis for discovery and innovation? Does considering gender add a valuable dimension to research? Does it take research in new directions?”
Schiebinger and her co-authors address these questions and draw from a range of scientific areas – such as biomedicine, artificial intelligence and machine learning, social robotics and the marine sciences – to show how a gender and sex analysis can lead to new discoveries.
Opening opportunities, advancing discovery
Scientists have long excluded females from experiments because they incorrectly assumed that fluctuations in the female reproductive cycle could alter their results, Schiebinger and her co-authors point out.
In medicine, sex-specific drug dosages while uncommon do exist, the researchers wrote.
For example, when it was discovered that older women are more likely than men to experience the side effects of vasopressin – an antidiuretic hormone used to regulate water in the kidneys – the European Union and Canada recommended lower dosages for women taking the drug.
Cancer immunotherapy is another area that is tailoring treatment based on sex differences. Patients with melanoma or lung cancer who are treated with checkpoint inhibitors – a therapy that stimulates natural killer cells – respond differently based on their sex. A possible explanation for the sex-based difference is the natural killer cells’ sensitivity to estrogen and testosterone.
These examples show how a sex-based understanding of treatments can help improve patient care.
Along with medicine, artificial intelligence is another area where scientists are beginning to recognize underlying consequences of the technologies they create. James Zou, who is an assistant professor of biomedical data science and co-author on the paper, has examined how machine-learning algorithms can pick up and perpetuate existing biases and stereotypes in society.
Algorithmic bias is well documented, the researchers emphasized in the paper. They provided several examples, including a previous finding of a search engine that was more likely to display ads for high-paying executive jobs to men than women; and in another case, an algorithm that misidentified images of a man in a kitchen as a woman.
As medicine, healthcare and technologies are becoming increasingly personalized, “We need to better model and incorporate sex/gender in order for the products to be robust and reliable,” Zou said. “Understanding sex/gender differences also reveals important fundamental new science.”
In the paper, Schiebinger and her co-authors offer a variety of ways the scientific community can incorporate a sex and gendered perspective into their research.
One suggestion is for educational institutions to integrate disciplines that historically have remained separate.
At Stanford, Schiebinger runs the Gendered Innovations project that provides scientists and engineers with practical methods for sex and gender analysis. As part of the project, Schiebinger teamed up with a colleague in computer science and organized two workshops – one on Gender and Fairness in Machine Learning and another on Gendering Social Robots. These sessions put scientists and engineers in conversation with humanists and social scientists, where “we can really exchange research and talk,” Schiebinger said.
The project also publishes case studies to illustrate how sex and gender analysis can lead to new understandings of a problem or phenomenon. For example, one case study examines bias in machine learning and shows the various ways algorithms, if left uncorrected, can reinforce stereotypes and social inequalities.
Another opportunity is to improve study design from the beginning and, when applicable, include sex and gender as categories of analysis.
To help scientists and engineers decide whether a sex and gendered analysis is suitable, Schiebinger and her co-authors created a set of decision trees to help guide study design and reporting structure.
The researchers also urge scientists to be more transparent when they report their findings and share when they do or do not discover any differences and similarities between sex and gender.
“A lack of transparency in reporting sex and gender-related variables makes it difficult to reproduce experiments in which these variables affect experimental results,” they wrote.
Change is afoot, Schiebinger acknowledged.
“There is an unusual dedication to the way we have always done things,” Schiebinger said. “But things are changing quickly now. Since 2016, the National Institutes of Health requires analyzing ‘sex as a biological variable’ to enhance reproducibility in science. The National Science Foundation is holding workshops in 2020 to investigate how best to integrate the sex, gender and diversity analysis into its funding guidelines. The European Union is ahead of the U.S. in this regard – it’s time to catch up!”
Schiebinger is also the director of the Gendered Innovations in Science, Health & Medicine, Engineering, and Environment Project.
Other authors of the paper include Cara Tannenbaum from the Université de Montréal in Canada, Robert P. Ellis from the University of Exeter in the U.K. and Friederike Eyssel from Universität Bielefeld in Germany. Ellis acknowledges financial support from a NERC Industrial Innovation Fellowship and Zou is supported by a Chan-Zuckerberg Investigator Award and the NIH.