Social media is a window into the personalities, mental states, and health behaviors of users. But can social media platforms predict medical diagnoses?
A team of researchers from the University of Pennsylvania, Philadelphia’s Veteran Affairs Medical Center, Microsoft, and Stony Brook University recently analyzed 949,530 Facebook status updates and electronic medical records from 999 consenting participants. Drawing from individuals’ Facebook accounts, they analyzed language used in online posts and related it to medical conditions listed in participants’ medical records to evaluate the accuracy of diagnostic predictions across 21 diseases. The comparison was based on predictions from Facebook status language alone, demographic indicators (age, sex, and race) from users’ electronic medical records, and a combination of both Facebook status language and demographics.
As illustrated by the graph above, Facebook language predictions (blue bars) were significantly more accurate than predictions from only demographics (orange bars) for 10 disease categories. These categories included diabetes, pregnancy, anxiety, psychoses, and depression. When compared to demographics alone, a combination of Facebook and demographics (striped bars) was a stronger predictor of 18 out of 21 disease categories.
These findings highlight the potential to use social media as a health screening tool. Health practitioners could use social media platforms to assess behavioral and environmental disease risk factors at both the individual and population level.
In fact, Facebook recently launched a preventative health tool that allows users to log and track their health activities. Based on users’ entries, the platform recommends flu shots, cancer screenings, and heart health tests.
While early identification or even diagnosis could result in earlier intervention for certain health outcomes, predicting diagnoses from Facebook or social media statuses raises contentious questions about privacy, informed consent, and data ownership. Social media platforms make money from selling user data to advertisers. With access to users’ health information, will health data be sold to insurers and pharmaceutical companies?
Databyte via Evaluating the Predictability of Medical Conditions from Social Media Posts, PLOS One, June 17, 2019.
