Pain care and AI share a similar challenge: neither starts from a level playing field. As the recently published National Commission into the Regulation of AI in Healthcare highlights, AI outcomes can depend heavily on the information used to develop and test a technology. Its recommendations call for age, sex, race, ethnicity and geography to be considered throughout development, evaluation and real-world use.
A UK report led by Imperial College London has similarly warned that AI and data-driven health technologies could worsen inequalities for minority ethnic groups through biased algorithms, poor data collection and insufficient diversity in research and development.
At the same time, the burden of chronic pain is not equally shared. An estimated 44% of UK adults, just under 28 million people, live with a degree of chronic pain, and it may affect up to 30% of younger working-age adults aged 18 to 39. The British Pain Society also reports chronic pain prevalence of 44% among people in the Black ethnic group, compared with 34% across ethnic groups overall.
Dr Ama Kissi has raised a related concern: pain research based predominantly on White participants can bias the knowledge generated and may not generalise to the wider population.
The Health Survey for England 2024 provides further evidence of prevalence in different groups looking at people that experienced on most days or every day during the preceding three months. This was more common among women than men, 29% compared with 22%, and almost twice as common in the most deprived areas as the least deprived, 36% compared with 19%.
Put these challenges together and there is a risk of a double disadvantage. Where people already experience inequalities in pain and pain care, introducing AI from an unequal starting point could reinforce the very gaps we hope innovation will help close.
We therefore need to demonstrate, rather than assume, that innovation works equitably for the different populations experiencing pain.
Who is represented in the data?
AI learns from data, so representation matters.
The Faculty of Pain Medicine recently highlighted AI’s potential to support more objective pain assessment, including through the analysis of facial expressions. However, it also stressed the need for “large, diverse datasets and real-world clinical validation to help ensure AI can translate effectively into healthcare”.
This raises key questions about whether datasets adequately represent differences associated with protected characteristics and populations with a higher prevalence of chronic pain. The evidence may not be equally developed for every population, making it even more important to identify gaps rather than assume that AI will work equally well for everyone.
A 2024 review explored how AI-supported diagnosis could draw on multiple sources of information, including facial expressions, neuroimaging and physiological signals but it also identified bias and ethical concerns about data as significant challenges.
Used well, better data could help us understand differences in people’s needs and make more informed decisions. But for patients, there is a fundamental question:
If I don’t know I'm adequately represented in the data used to develop AI, how do I know it will work as well for me?
Who can access AI?
Inclusive AI also depends on who can access and use it.
Technology could make self-management support available to more people, including those who may not easily access specialist services. But digital connectivity, confidence, and location could impact who gains access.
AI Pete, developed with Teesside University, offers a practical example. It uses an online bot to share pain advocate Pete Moore’s lived experience and self-management support more widely with others. We asked the real (!) Pete about its potential. He said AI “could make pain supported self-management more interesting and engaging, creating games and pain self-management in a familiar format if asked for by patients”.
The National Commission’s recommendations provide an important standard for what should happen next. They call for greater focus on whether AI works safely and effectively for different people, rather than simply whether it performs well on average, with particular attention to fair outcomes for underserved groups.
If people experiencing the greatest burden of pain are the least likely to access, use or benefit from new technology, an improvement on average could leave existing inequalities untouched or even widen them.
What do we need to bear in mind?
Avoiding a double disadvantage requires more than inclusive design. It means asking who is represented in the data and who can access new technologies, then continuing to design technologies with these groups in mind, measure who uses them, who benefits and whether the inequalities present at the outset are getting smaller.
The marker for successful innovation and technology is not simply whether AI works broadly. It is whether it works specifically and effectively for different people, reaches those who most need support and contributes to fairer outcomes in pain care.