A new study by Penn Medicine researchers reveals that health care algorithms can both improve and exacerbate racial and ethnic disparities in access to care, quality of care, and health outcomes for patients. The systematic review, published in the Annals of Internal Medicine, examined 63 studies from 2011 onwards, focusing on the impact of algorithms used in clinical care, resource allocation, and health care management.
Lead author Dr. Shazia Mehmood Siddique emphasized the need for transparency in algorithm use. “What we know is just the tip of the iceberg. There are many algorithms embedded in patient health records that are not transparent and have never been studied,” she said.
The researchers found that some algorithms reduced disparities, while others perpetuated or worsened them, regardless of whether race or ethnicity was explicitly included as a variable. To mitigate bias, the study suggests constructing algorithms with data that accurately represent diverse populations and replacing race variables with more precise factors, such as genetic data or social determinants of health.
The research is timely as policymakers consider ways to protect the public from unintended consequences of predictive models and AI in health care. Recently, the Office of the National Coordinator for Health Information Technology finalized a rule requiring transparency for algorithms used in Medicare and Medicaid electronic health records systems.
This study underscores the critical need for careful development, implementation, and monitoring of health care algorithms to ensure they promote equity rather than exacerbate existing disparities in the health care system.
See “Friend or foe: A closer look at the role of health care algorithms in racial and ethnic disparities” (March 25, 2024)
