By analyzing only the Facebook posts of nearly 1,000 patients, an artificial intelligence algorithm was able to correctly detect 21 separate conditions, according to a study published June 17 in PLOS One.
Researchers from the University of Pennsylvania's Perelman School of Medicine in Philadelphia and Stony Brook University in New York used the algorithm to analyze Facebook posts from 999 patients, comprising approximately 20 million words in total. The posts were analyzed both alone and in tandem with demographic data, with the AI's findings compared to each patient's electronic medical record.
When used alone, the Facebook posts were found to be accurate predictors of all 21 conditions studied, with 10 of the conditions better predicted by Facebook posts than by demographic data. The algorithm was particularly successful at detecting diabetes and mental health conditions such as anxiety, depression and psychosis.
According to the researchers, the algorithm's findings were somewhat surprising: Using religious language like "God" and "pray" on Facebook, for example, made a patient 15 times more likely to have diabetes than the patients who used those terms the least, while words like "dumb" and some expletives were strong indicators of drug abuse and psychosis.
Because some of the words and phrases found to be linked to various diseases are not necessarily ones that a patient might use during a clinical visit, the study's authors concluded, utilizing the AI to monitor patients' social media feeds could provide valuable information to which healthcare providers might not otherwise have access.
More articles about AI:
Investors, accelerators urge AI startups to implement code of ethics
Harvard Business Review: 3 ways AI marketplaces will change radiology
'Structural change' necessary to reduce gender bias in AI, say Google, Microsoft leaders