AI’s Potential Impact on Diagnostic Errors and Health Disparities

More and more, the use of artificial intelligence (AI) is helping physicians overcome many of our most common Health Disparities.  An article published in JAMA Health Forum does a deep dive into this game-changing tool.

In one example, the use of knee replacement surgery for the management of knee osteoarthritis remains one of the largest racial disparities in US health care today, mostly because pain, while highly subjective, is a key clinical indication for surgery.  Applying machine learning markedly improved the predictive performance of the diagnostic process. Machine learning was used to develop a new algorithm to measure the severity of knee osteoarthritis based on existing radiologic data and showed that Black patients had more severe osteoarthritis than White patients.

In another example, the study indicated that Black patients are more likely than white patients to be under-diagnosed with depression during their first primary care visit for mental health needs and that Black children with ear infections are less likely to be diagnosed compared with White children with similar symptom presentations.  With the use of AI, physicians can help eliminate any implicit bias at the diagnostic level.  AI relies on algorithmic data to predict health needs.

Technological transformation with AI in healthcare has the potential to address diagnostic errors, health care disparities, and more. However, AI and machine learning algorithms are only as good as the data they are trained in. There is increasing evidence that machine learning algorithms may cause unintended harm, resulting in the under-diagnosis and delayed treatment of chronic diseases.

Overall, AI in health care has the potential to address the twin challenges of diagnostic errors and health care disparities. However, it is important that diverse perspectives inform AI’s development, that AI models are trained with data from and about diverse populations, and that any unintended negative consequences of the models are identified and addressed.