- A new study in the journal Science concludes that a widely-used algorithm used by providers to determine the complexity of healthcare needs has an inherent bias against African-American patients.
- The bias in the algorithm — used for high-risk care management programs — is based on cost of care. As less money is generally spent on care for African-Americans, the algorithm concludes that African-Americans are therefore healthier than other patients, regardless of their actual health status.
- Remedying the disparity would more than double the percentage of African-Americans receiving additional care, according to researchers.
The history of U.S. medical care and African-Americans has often been fraught.
The Science study shows how racially biased assumptions can be baked into even supposedly neutral algorithms.
The particular algorithm researchers at the University of California, Brigham And Women’s Hospital, the University of Chicago and Massachusetts General Hospital focused on are known as “high-risk care management” programs. They are used to improve the health of patients with complex healthcare needs and are intended to improve coordination of healthcare services.
Researchers concluded such programs include as many as 200 million patients nationwide. Their study focused on nearly 50,000 commercially insured and Medicare patients attached to a large teaching hospital.
Prior to being admitted to such programs, an average of $1,801 less per year is spent on care for African-American patients versus whites. As a result, the algorithm for a high-risk care referral kicks in later for African-American patients compared to whites.
As a result, the study concluded African-Americans whose health indicators automatically directed them into a care management program have an average of 26.3% more chronic illnesses than whites who are also recommended for such care.
Curing for the bias would mean the percentage of African-Americans in the study sample directed into such programs would increase from 17.7% to 46.5%.
“Rather than predicting costs at all, we could simply predict a measure of health," like the number of active chronic health conditions, the study’s authors wrote. “Because the program ultimately operates to improve the management of these conditions, patients with the most encounters related to them could also be a promising group on which to deploy preventative interventions.”
A simulation based on this assumption reduced bias in the selection process by 84%, the authors concluded.
The study’s authors not only recommended addressing the bias in the algorithm, but addressing other factors, such as the ongoing distrust between African-American patients and white providers.