AI to Reduce Racial Disparties in Healthcare

I enjoyed this article about the development of a new algorithm to read knee X-rays. Typically, such algorithms are developed by comparing their performance against doctors’ diagnoses. However, these researchers decided to evaluate the algorithm’s ability to predict patients’ self-reported pain. This was important because the standard diagnostic criteria for osteoarthritis performs much better at diagnosing White than Black patients’ pain.

1 Like

Fascinating. Thinking about self-reported pain in an African context, @helenelizabeth and I both very much appreciate the book Improvising Medicine. It’s an ethnographic text, so won’t directly/obviously translate into ideas for health system and software design, but Chapter 5 (on pain and laughter) puts some of these considerations in a rich cultural context.

1 Like

I agree, @isaacholeman. I think that notions of pain and even how pain is quantified and expressed are so steeped in cultural contexts that a number of questions are raised when thinking about self-reporting pain. I also think notions of disability and debility are really tied to different ways of knowing, so thinking about how Western biomedical ideas and categories interplay with culturally specific understandings of illness and abilities is a nuance worth exploring further.

Julie Livingston’s work, as you mentioned above, really speaks to these questions – not just Improvising Medicine but her earlier work Debility and the Moral Imagination in Botswana.

A critical question in our work is how best to capture experiences and meanings of pain or disability in ways that translate across different ways of being in the world – speaking to both Western biomedical forms of quantification and intervention, as well as different local contexts of care, community, and embodied experience.

A slight tangent but a related one below:

It’s funny, I often think about how ‘disability weights’ in the Global Burden of Disease study, which are used to calculate metrics that capture nonfatal health loss (ie. YLDs, DALYs) are created. They use surveys administered in an online format across a subset of countries and are based not on the feedback of people experiencing a given condition or cause of disability but, rather, the perceived health loss from the subset of respondents within the survey group. This, to me, is such a problematic way of thinking about and quantifying disability that both discounts lived experience AND cultural context for the opinions of a subset of people with access to computers and invitations to participate in the survey.

What might disability weighting look like if CHWs asked members of their community about how a given condition, or set of co-morbid conditions, impacted their daily life? What duration? What longterm effects? I think similar questions could be asked of pain and the communal experience of pain.

I don’t have any answers but I think these are important questions for us to grapple with together.

1 Like