We’d like to share some thoughts on the results of a famous Journal of American Medical Association (JAMA) study.
The study found that among resident physicians, symptoms of burnout appeared to be associated with greater explicit and implicit racial biases:
What’s even more disturbing, the study suggests it’s just an association, it’s causation: “Burnout, a syndrome characterized by emotional exhaustion, depersonalization, and a decreased sense of efficacy, is common among resident physicians, and negative emotional states may increase the expression of prejudices, which are associated with racial disparities in health care.”
What does it have to do with AI in healthcare? Whether human biases are intentional or not, they end up in the clinical and medical datasets. No matter how advanced AI and machine learning algorithms are, they inadvertently pick up these biases in their training and validation and incorporate them into their suggested solutions. Statisticians call this the GIGO concept – Garbage In Garbage Out.
One way to address this issue is to “de-bias” your algorithms – advocated by IBM and Microsoft:
Another way is to be selective about and attentive to your data. For example, the NIH’s MEDLINE/PubMed dataset doesn’t suffer from the abovementioned biases to the extent a hospital dataset would. While certainly not completely bias-free, medical studies are to some extent already “cleaned up” as they are vetted by editors and reviewers, as well as by clinicians and statisticians who conduct the study.
It’s time to realize that the healthcare system is so interconnected that solving one problem may mitigate other problems.
Solving the burnout problem may help solve the AI racial bias problem.
Stay healthy! Stay knowledgeable about your health.