Diagnosing bias in data-driven algorithms for healthcare
From the article:
A recent analysis highlighting the potential for algorithms to perpetuate existing racial biases in healthcare underscores the importance of thinking carefully about the labels used during algorithm development.
Many data-driven algorithms in healthcare map a set of patient characteristics to the patients’ estimated risk of experiencing a future outcome1,2,3. Such algorithms are often used to identify high-risk patients for targeted interventions. Recently, Obermeyer et al. examined one such algorithm currently used by health systems in the USA to target patients for high-risk care management4. Analyzing the algorithm’s predictions by race, where race was self-reported and extracted from hospital records, they identified racial bias in the algorithm. Specifically, black patients were less likely to be identified by the algorithm as candidates for potentially beneficial care programs than were white patients who had the same number of chronic illnesses.