Insights gleaned from posts on Facebook could reveal whether people have conditions including diabetes, according to an American study.
Researchers examined the language used on the platform by almost 1,000 people and found that some key words could be better at identifying certain health conditions compared with sociodemographic information.
Diabetes is one of the conditions that could be predicted (the researchers did not specify whether this was type 1 diabetes or type 2 diabetes), as well as anxiety, psychosis and depression.
The team from Penn Medicine and Stony Brook University believe its findings could eventually lead to new applications for medicine-harnessing artificial intelligence.
They say language used on the social media platform could one day be monitored, if consent was gained, just like physical conditions.
“This work is early, but our hope is that the insights gleaned from these posts could be used to better inform patients and providers about their health,” said Dr Raina Merchant, who is the director of Penn Medicine’s Center for Digital Health and led the study.
“As social media posts are often about someone’s lifestyle choices and experiences or how they’re feeling, this information could provide additional information about disease management and exacerbation.”
A special automated data collection technique was used, enabling the team to examine every Facebook post of each participants whose health records were also analysed.
A total of 21 conditions were looked into, with the results indicating that every condition was predicted by Facebook activity. The findings also revealed Facebook data was better at predicting 10 of the conditions compared to demographic information.
According to the results, some of the language was obvious, with ‘drink’ and ‘bottle’ linked to alcohol abuse, but other words were not so obviously linked to a health condition. An example being the fact that people using the words ‘God’ and ‘pray’ were 15 times more likely to have diabetes, compared with users who used these terms the least.
The research was published in the journal PLOS ONE.
