This post is also available in .
For a long time, medicine and neuroscience operated under an implicit assumption: the male body was treated as the “standard” human body. For decades, women were excluded from clinical studies because of concerns surrounding hormonal fluctuations and pregnancy-related risks, while sex differences were often dismissed as statistical “noise” rather than meaningful biology. The consequence is a scientific literature that, despite enormous progress, still contains major blind spots regarding women’s health. This is not simply a matter of representation. It is also a fundamental data problem.
Participation Alone Is Not Enough
Within medicine, women are not only underrepresented in research populations but also in the questions researchers ask and, therefore, in the information they collect. Participation itself is not necessarily the issue: women are generally included in studies, yet sex-specific effects are often barely analyzed. This is especially visible in neuroscience, where only a small proportion of published studies explicitly focus on women’s health. As artificial intelligence begins to play a larger role in biomedical research, this representational gap is becoming even more consequential.
A recent article in The Transmitter described how this problem emerges in large-scale neuroscience datasets, which form the foundation of many modern AI models. These models analyze enormous amounts of data to identify relationships between brain function, health, and behavior. However, many of the world’s largest neuroscience datasets contain little to no information specifically relevant to women. Variables such as menstrual cycle phase, hormonal contraceptive use, pregnancy history, and menopausal status are often entirely absent.
This is problematic because these are not irrelevant variables. Hormonal fluctuations influence cognition, mood, and brain function, and recent research has shown that pregnancy is accompanied by substantial measurable changes in the brain. If these factors are absent from the data, they are also absent from the models built upon those data.
In other words, we are constructing increasingly sophisticated AI systems on foundations with major gaps.
Catching Up
At the same time, there are clear signs that this is beginning to change. Organizations such as the National Institutes of Health now require researchers to consider sex as a biological variable in study design and analysis. There is also growing recognition that hormonal fluctuations should not be treated as “noise” to be removed, but as biologically meaningful information that must be understood. In parallel, an increasing number of large-scale initiatives are emerging that specifically focus on women’s health and on standardized reproductive health data within neuroscience.
Autism and ADHD
The consequences of historical bias become especially visible in research on autism and ADHD. Diagnostic frameworks for both conditions were largely developed using predominantly male research populations. As a result, women and girls often differ from the typical “classical” profile described in diagnostic criteria and clinical manuals.
In autism, women are more likely to mask social difficulties, which can contribute to delayed or missed diagnoses. In ADHD, attentional difficulties have historically received less attention than hyperactivity, despite being relatively more common in girls. This creates a self-reinforcing cycle: male-centered datasets lead to male-centered models, which in turn reinforce male-centered diagnostic frameworks and the treatments adapted to them. Consequently, we still know remarkably little about how the efficacy and side effects of ADHD medication vary across different phases of the menstrual cycle in women, despite growing evidence that these effects may be important.
The broader shift currently taking place in women’s health research is therefore about far more than representation alone. Better inclusion leads to better data, and better data ultimately leads to better science. If neuroscience and artificial intelligence aim to model humans more accurately, they cannot continue relying on only part of the picture.
Author: Lucas Geelen
Buddy & Editor: Wieger Scheurer
Vertaler: Xuanwei Li
Afbeelding by Jairo Gonzalez via Unsplash
About The Author
