Throughout history and in current society, there is an overarching tendency to standardize and default to an average male – whereby male is viewed as universal and female is atypical. This is true for everything from seat belts in cars to public transportation system design to occupational health and medical research.
What this has led to is a world designed by men for men in which over half the world’s population of women and other gender classifications are being left out. That is the central thesis of a recent book; Invisible Women: Data Bias in a World Designed for Men by Caroline Criado Perez.
In many circumstances this generally means there is no consideration for the differences
in women’s needs, living patterns and behaviors leading to significant inconvenience and disadvantages for women – whether in pay equality or consideration for how travel patterns impact caring for children and/or sick relatives. Even if we flipped a switch today and decided to consider gender in all societal issues, the main challenge is that there is very little research and gender disaggregated data that can be referenced to ensure more equitable and representative approach to those issues. In addition, women comprise only 12% of machine learning researchers. As artificial intelligence gains momentum in health, filling the gender data gap and including women in a more transparent approach to the development of algorithms becomes of utmost importance.
There are increasing examples of how the lack of gender-disaggregated data is harming the health of women and other under-represented populations. This has led to lower detection of risk for heart attacks and more heart attacks in women. No one would argue that physiologically women’s bodies are different from men’s, but the approach to health research indicates little consideration for these differences. The way this manifests itself is in historical
pooling of data on men and women in health research and more men in research studies – leading to underrepresentation of women and other gender classifications in health research findings. Earlier this year, headlines were made over the lack of testing of an HIV prevention drug in cisgender women and transgender men.
Women experience health conditions differently and respond to treatment differently. A review of data gap related to sex and gender in diabetes research reporting highlighted that while diabetes studies focused on complications and treatment have included women, the analyses plans did not include sex/gender and less than 10% of studies reported all study outcomes by sex/gender. Studies have shown sex differences in diabetes complications, including cardiovascular disease and depression as well as in responsiveness to treatment, including less responsiveness to insulin treatment to lower HbA1c levels in type 2 diabetes. This is just one example. A whole range of others exist in other disease conditions as well as in occupational health where safety standards related to weight limits and chemical exposure are largely standardized to men’s bodies, increasing the risk of women to injury. In addition, women have the additional disproportionate burden of gender-based violence.
The good news is that when women are taken into account, even small adjustments have profound effects on both women’s health and the health of society overall. They have even led to significant reductions in health expenditures. Moving forward, we need to continue to assess the impact of the sex/gender on health outcomes, increase research on women, re-analyze existing data sets that include data on women, and insist that all health research findings are disaggregated by gender.
As machines mine imbalanced and biased data in all aspects of life, the risk of sexism and bias is increasing. To date, the lack of diversity in AI and opaque approach, with little to no accountability to how algorithms are developed, has the potential to exponentially project the default male into all aspects of society with the greatest risk to individual and societal health. To counter this, we need more women in AI and better representation of women’s perspectives in the development of algorithms and transparency in algorithm development.
But to truly harness the benefits of AI in health, we need more sex/gender disaggregated data and women in AI.
Patricia N. Mechael, PhD MHS is co-founder and Policy Lead at HealthEnabled