The Gender Health Gap
by Dr. Shruti Turner and Dr. Anahita Talwar
It isn’t new news that there is a gender health gap. Books, such as Invisible Women, have helped bring the discussion of gender equality and data bias to mainstream media. But what can we do to change this? In this article, the data science team at Haleon discusses how a purposeful use of data can drive better understanding of women’s health and inform equity and inclusivity to deliver better health outcomes for humanity.
What is the Gender Health Gap?
The gender health gap refers to inequalities in mental and physical health outcomes between women and men. A study by Manual demonstrated that the direction of these inequalities varies globally. In 41% of the 156 countries included, men were generally less healthy relative to other countries’ men across their lives, whilst in 58%, women suffered comparatively worse health outcomes compared to women in other countries (1% showed no health gap). The Netherlands, Sweden, Denmark, and the UK, all rank in the top 12 for the largest women’s health gap globally.
In 2021, the UK Government Census showed that the healthy life expectancy of women is actually decreasing while men’s showed no significant change. Despite having a higher life expectancy than men, the number of years a woman can expect to live in good health (free from long-term illness or disability) is lower than their male counterparts. Reproductive health, mental health, and cancers are just some of the areas in which women’s unmet needs have been highlighted.
Why does the Gender Health Gap exist?
Though the reasons for health gaps are complex and multifaceted, it is likely that a gender data gap plays a big part. Women have historically been excluded from research due to the belief that the bodily changes throughout the menstrual cycle might interfere with study results. This was exacerbated by the thalidomide scandal of the 1950s which led to guidance that women of childbearing age should be excluded from clinical trials in some parts of the world for nearly 20 years (the US FDA guidance only reversed this in 1993). As recently as 2017, the results of a clinical trial published for a drug aimed solely at women’s use showed that 23 of the 25 participants were men. The exclusion of females has even been observed at the pre-clinical level where typically male lab rats are used exclusively in research.
The relative absence of women in scientific research roles, particularly at higher levels, has meant that their perspectives have been missing and is likely to have influenced this gap. Having diversity at the table can help to reduce bias in research funding (all women teams are more likely to focus on solutions for women’s health), and in its nuanced interpretation (i.e., while some differences between men and women are important for improving healthcare outcomes, others might not be relevant, or even exist).
Combined, these factors have contributed to the reality that women’s bodies, and the conditions that affect them, are less likely to be studied. Governments and research groups have made progress towards fixing the gap, recommending representative samples are used in trials and encouraging gendered analysis, as well as expanding clinical education and training in women’s health issues.
Data has shown the legacy of these historical decisions in the form of previously under-acknowledged gender-specific symptoms (e.g. presentation of heart attacks) leading to under-diagnosis in women, and sentiments that women’s concerns are not taken as seriously. These issues are often exacerbated for women of colour.
What is the role of Data Science?
Left unchecked, machine learning and artificial intelligence algorithms will replicate and even exacerbate biases in the data that they are fed, which given the examples of unrepresentative data above, sounds like bad news for women’s health. If an algorithm sees more examples of men’s data when training, it will have a better representation of this category and won’t perform as well when making predictions about women.
Researchers from UCL found exactly that when they reproduced four published machine learning models that had been trained to predict liver disease. Because the dataset used for training (a common one within the field, by the way) has a 1:3 ratio of women to men, all four models showed higher false negative rates for women and higher false positive rates for men. As a result, if unchecked, these algorithms would lead to higher rates of missed diagnoses in women if used in clinical practice.
The requirement for balanced data when training algorithms is not a new discovery, yet data and predictive bias are rarely scrutinised in the context of demographic variables and their resulting impact on health inequalities.
BUT it’s not all bad news, the beauty of data and algorithms is that, when used well, they can efficiently fill gaps in our knowledge. They can pick up subtle differences in how healthcare symptoms and solutions might differentially affect men and women, even when they are challenging for researchers to detect.
What is Data Science at Haleon doing to address the Gender Health Gap?
As the world’s largest single-focused consumer health company, Haleon is dedicated to delivering better everyday health with humanity. The data science team at Haleon is uniquely placed to use its expertise and experience to help inform a better understanding of women’s health through a more purposeful use of data to drive better health outcomes.
Inspired by Dame Mary Lucy Cartwright (a British mathematician who contributed to important advances in function theory and differential equations), a women-led cross-disciplinary initiative combining data science, Haleon’s disruptive innovation unit and clinical team members are working together to look at how we can combat the gender data gap in healthcare.
Our first focus is osteoporosis, a bone disorder that increases the probability of life-changing fractures. Despite it disproportionately affecting women (approximately 90% of cases), most risk factors incorporated into clinical tools, e.g., FRAX, are gender-agnostic, and as such there is little understanding of how variation in women-specific features, say the number of pregnancies or duration of peri-menopause, may impact their risk of osteoporotic fractures.
As a combined team we are aiming to use data science and machine learning to investigate this data gap further. As part of our work, we will explore whether using sex-agnostic data fields results in equally accurate risk prediction between men and women, or if there is evidence for bias. We also aim to identify if female-specific features are predictive of osteoporosis and osteoporotic fractures, and whether measuring these data points can improve risk prediction in women, ultimately leading to better targeting of preventative treatment.
At Haleon, ensuring Responsible AI is paramount. In the data science team, this includes evaluating fairness, equity, and bias in algorithms we develop and deploy, and ensuring we apply ethical principles such as transparency and being grounded in trusted science. By finding solutions to overcome the impacts of algorithmic bias in healthcare, and raising awareness of the importance of responsible AI, we can ultimately inform decisions to provide better everyday health for everyone.
As highlighted above, representation matters…
In the UK, the Turing Institute’s recent study found that only about 20% of UK data scientists are female, with the ratio generally declining not improving.
At Haleon, the women that make-up 35% of our global data science team are highly valued for their skills and experience, and we strive to create an inclusive environment.
In progressing this exciting approach to enabling gender health equality the team at Haleon are taking another positive step in moving forward the inclusion of women’s health data for future developments.
