How to develop trans people-inclusive medical AI: the objective of a study by UPF, the BSC, the URV and PRISMA

In the field of health, the development of artificial intelligence (AI) may mean major breakthroughs in personalizing diagnoses and treatments, but there is a need to promote bias-free AI with a more diverse and inclusive vision so that everyone can benefit from it. With this study, Pompeu Fabra University (UPF), the Barcelona Supercomputing Center - Centro Nacional de Supercomputación (BSC-CNS) and Rovira i Virgili University (URV) have moved towards the development of medical AI applications that overcome binary logics and include the needs of trans people. The study was carried out in collaboration with the PRISMA association, which protects the rights of LGBTQIA+ people in the field of science and innovation.

Inclusive medical AI could benefit trans people in several ways, such as the greater personalization of masculinizing or feminizing hormonal treatments according to the physiological characteristics of each individual. Another example is the design of voice apps that take the needs of trans people into account. Many of them use them to help with their process of voice change and it causes them emotional discomfort when the app recognizes them as being of a gender with which they do not identify.

This pioneering study in Spain was recently published in an article in the scientific Journal of Medical Internet Research. The main researcher of the study is Nataly Buslon Valdez, of the Social and Responsible Computing research  group of the UPF Department of Engineering. The study co-authors are Davide Cirillo, head of the Machine Learning for Biomedical Research unit at the BSC and co-leader of the research line on sex and gender bias in AI of the BSC’s Bioinfo4Women initiative; Oriol Rios (URV Department of Pedagogy and member of PRISMA); and Simón Perera del Rosario (UPF Department of Medicine and Life Sciences and PRISMA member).

Buslon had spent several years researching health and gender examining the medical and technological biases that harm women. “We saw the need to overcome the binary vision and see the problems faced by communities such as trans people in order to achieve a more inclusive, diverse and equitable approach”, she explains. On the basis of this idea, which arose from the Bioinfo4Women initiative of the BSC-CNS, she contacted the PRISMA association, which for years has denounced the gaps suffered by the LGBTQIA+ collective in relation to science and innovation for years. PRISMA helped the research team to get in contact with trans people. In addition, it enjoyed the support of representatives of TRÀNSIT, the Health Care and Promotion Service for Trans and Non-Binary People of the Catalan Health Institute (ICS).

The research is grounded in communicative methodology, which requires the active participation of the study population and of different social groups involved in the field of research. A representative cohort of 18 trans people participated in three telematic focus groups to express their views on the topic and were involved in designing the research. PRISMA representatives collaborated in overseeing the focus groups to ensure that the research was ethical and respectful towards trans people. The study sought to analyse two issues, principally: on the one hand the challenges that must be overcome to adapt the design of AI applications in health to the needs of trans people and, on the other, the potential solutions. “The key is to promote more diverse and inclusive AI both in the field of health and globally”, Buslon concludes.

The focus groups have enabled finding “that in many instances, AI replicates the biases of its creators and can increase the invisibility of trans people”, explains Simón Perera del Rosario (UPF). He exemplifies this with the existing apps to help with the voice-changing process of trans people. “These digital systems are designed in a binary way. If someone uses an app that helps them change their voice and identifies them as a man, when they are a trans woman, this will cause them discomfort”. For Buslon (UPF), the constant discrimination of this group, also by technology, “can affect their mental health, self-esteem and quality of life”.

How can trans people benefit from the medical potential of AI?

“The study recommends developing inclusive AI applications that enable personalized treatments tailored to the specific needs of each trans person”, Davide Cirillo (BSC) explains. This would allow, for example, adjusting the doses of masculinizing or feminizing hormone therapies to the physiological characteristics of each individual. In addition, AI tools could help detect and correct incompatibilities between hormone therapy and other drugs. 

In the focus groups, the trans people also expressed the need to make appropriate use of their personal data for health purposes. For example, only for medical purposes, health professionals should know if they are dealing with a trans or a cis woman, because, for the treatment of certain diseases, doses could differ in each case. In fact, in the focus groups, the trans people warned that binary models of data management and new technological developments in the medical field may lead to errors in diagnoses and treatments.

Strengthening trans people’s trust in the health system

In addition, the study stresses the need to reverse trans people’s distrust towards health institutions, due to their historical discrimination in this field. It should be remembered that the WHO did not remove transsexuality from the International Classification of Diseases until June 2019. To do so, “it is advisable to improve health professionals’ training and awareness of the needs of the trans community”, Oriol Rios (URV) adds. In addition, the study aims to carry out more scientific studies on the subject, which are very scarce to date, focusing on data management and AI with a diverse and inclusive vision. Finally, it recommends promoting solidarity networks and spaces for the exchange of knowledge between trans people and health professionals.