Healthcare AI has potential not only for neutralizing its inherent algorithmic bias but also for personalizing its outputs to help humans address health inequities.
All it will take is incorporating an ingredient missing from most algorithm training: patient input.
In fact, early testing of this hypothesis has already demonstrated the feasibility of patient-inclusive AI development.
Specifically, some of this research “has produced an algorithm trained to predict the knee pain reported by the patient, rather than the X-ray interpretation of the doctor,” the authors of a May 29 Lancet opinion piece write.
“This approach explained more of all patients’ pain compared with standard measures of radiographic severity,” they report, “and its explanatory power for pain was particularly useful for underserved groups of patients, such as Black patients or patients with low income and low education.”
The authors are Ziad Obermeyer, MD, of UC-Berkeley and the renowned book author and healthcare influencer Eric Topol, MD, of Scripps Research.
To the observation above they add that, compared with conventional means of understanding patients’ pain levels, the experimental algorithm outdid radiologists at this aim largely because its training included severity scores self-reported by a racially and ethnically diverse population.
The authors of the study under discussion, whose number included Obermeyer, further found that incorporating the algorithm’s patient-centric severity score while leaving out the radiologist’s pain prediction doubled the proportion of Black patients’ knees that were deemed eligible for replacement.
If replicable, Obermeyer and Topol suggest, the patient-inclusive AI training methodology might not only widen access to knee replacement surgeries but also help advance “a new kind of clinical science.”
For example, by grounding patient reports of pain in objective radiographic features, we might develop a more comprehensive understanding of what causes pain. By not being doctor-centric and incorporating the patient's perspective, machine learning has added potential for unravelling important mysteries of medicine.
Access the piece here.