Inside STAT: Widely used algorithm in hospitals is racially biased, study finds

Bias in algorithms is something that health technology experts worry about — especially as machine learning and other AI systems become more of a routine part of health care. And a new study finds that these experts’ concern is well-placed. Researchers looked at an algorithm that hospitals routinely use to rank which patients are likely to need follow-up care, and found that it identified white patients overall as being more ill than black patients — even when they were just as or more sick. Only 18% of patients ranked as needing more care were black, whereas an unbiased algorithm should have put that figure at around 46%. Still, the discrepancy may not be intentional. “We haven’t told algorithms yet to do things without racial bias,” Senthil Mullainathan, senior author of the new study, tells me. Read more from our conversation here.