Over the years, scientists had developed methods to identify depression by analyzing the language people use in social media posts. But a new study suggests these methods provide accurate results for white people but not for Black individuals.
Researchers at the University of Pennsylvania recruited more than 800 people – half of them white, half of them Black, some reporting depression and some not – and employed artificial-intelligence models to examine the language they used in their Facebook posts. Words and expressions used on social media that had been established as predictive of depression only proved to be so for the white participants. The model was three times less likely to accurately predict depression among Black people.
MORE: New drug protects people with weakened immune systems from COVID-19
"Our results raise concern that certain psychological processes thought to predict or maintain depression may be less relevant, or even irrelevant, to populations historically excluded from psychological research, including Black individuals," the researchers wrote in a paper published Tuesday in the Proceedings of the National Academy of Sciences.
The results highlight the need for more awareness about how "race and ethnicity influence the relationship between depression and language expression," the researchers wrote. The study also accentuates the need to increase representation of Black people and other underrepresented groups in medical research and to establish more accurate predictive models to address more diverse mental health needs, according to the study's lead author, Sunny Rai, a postdoctoral researcher in Computer and Information Science at Penn.
Results showed a strong correlation between white people using first-person pronouns, such as "I," "me" and "my" and the severity of their depression. White people who reported depression also used more self-critical and self-deprecating expressions, such as calling themselves a "wreck," a "mess" or "worthless" and communicated feelings of anxiety and isolation with words such as "terrified" or "misunderstood."
None of these results held up for the Black participants in the study.
"In this study, we found that use of the self-referential pronoun – also known as 'I-talk' – is unrelated to depression in the Black population," Rai said.
Overall, Black people – even those who did not report depression – used "I-talk" more frequently than white people in their Facebook posts. Also, using self-critical language, such as calling themselves "useless," and expressions of feeling like an outcast, such as "weirdo" and "creep," were not specifically associated with depression in Black people.
Even after training the AI model on language Black people used in their posts, it still did not perform well.
"So I think that the problem of underperformance is not limited to collecting more data," Rai said.
Other unknown factors may be involved, she said – factors in need of further study.
These results emphasized the need for caution when using AI in mental health settings, said Dr. Nora Volkow, director of the National Institute on Drug Abuse, which partially funded this study.
"As society explores the use of AI and other technologies to help deliver much-needed mental health care, we must ensure no one is left behind or misrepresented," Volkow said. "More diverse datasets are essential to ensure that health care disparities are not perpetuated by AI and that these new technologies can help tailor more effective health care interventions."
Rai said AI holds promise for assisting in health care settings, but that it is necessary to validate and test models being used to make sure they accurately reflect populations.
"It doesn't mean that AI is not useful or that it should not be used," Rai said. "It just means that we need to be more thorough and rigorous in our approach."
Researchers from Penn's Perelman School of Medicine and its School of Engineering and Applied Science conducted the analysis. They hope to expand on this research to examine how depression is expressed in cultures outside of the United States, Rai said.