Racial Bias Found in a Major Health Care Risk Algorithm
In this article, a group of researchers have found out that an algorithmic system that is being used in many hospitals isn’t predicting accurate “high risk care management” scores between white patients and black patients. This program will help identify patients’ special needs based on the amount of spending for medical issues. According to their statistics, 26.3% of black patients turned out to have more chronic issues than white patients. However, the algorithm has calculated a lower risk score for black individuals, which might result in further harms for clients’ health in general.
An explanation of what caused this algorithmic bias in this case:
- Inaccurate input data is one the main reasons that has led to unrepresentative outputs. Information source about white patients have completely outnumbered black patients' information, with the ratio of 7:1. This uneven information has caused difficulties in reflecting reality.
- Existing bias in reality has transferred into patients’ willingness to spend: Many black patients have been experiencing discrimination in real life, which decreased trust between doctors and patients. It explains the reason why black patients are less likely to spend, which does not reflect their actual their health.
Relevant stakeholders:
- Doctors: this population should be aware that the algorithm isn’t always accurate to make the right decisions while diagnosing. Blindly relying on this technology would increase the chances of making mistakes and decrease work effectiveness. They can use the algorithm for guidance and make critical decisions based on the patients’ previous health records and symptoms to maximize the use of this technology.
- Patients: inaccurate results from the algorithm might prevent patients from getting proper treatment and care, which is extremely dangerous for clients cancer and other illnesses. In the worst case, the falsity in predicting patients’ health levels can lead to many death cases.
- Algorithm developers: should constantly update their input data to avoid temporal bias and keep their information relevant to recent time, as well as being aware of their confirmation bias might also help eliminate them while coding
new algorithms. If not, this product will be excluded from the competitive market where all kinds of AI technology are enhancing everyday