[ad_1]
Credit: CC0 Public Domain
Every year, depression affects more than 6% of the adult population in the United States – about 16 million people – but less than half of them receive the treatment they need. And if an algorithm could scan social media and report the language alarm signals of the disease before a medical diagnosis was made?
New research from the University of Pennsylvania and Stony Brook University published in the Proceedings of the National Academy of Sciences shows that this is now more plausible than ever. By analyzing data from social networks shared by consenting users in the months leading up to the diagnosis of depression, researchers found that their algorithm could accurately predict future depression. Indicators of the condition included mentions of hostility and loneliness, words such as "tears" and "feelings", as well as the use of more first-person pronouns such as "I" and "me".
"What people write in social media and online represents an aspect of life that is very difficult to access in medicine and research," said H. Andrew Schwartz, lead author of the publication and principal investigator of the World Well-Being Project (WWBP). "This is a relatively undeveloped dimension compared to the biophysical markers of the disease.For example, if you take conditions such as depression, anxiety and PTSD, you find more of signals in the way people express themselves digitally. "
For the past six years, WWBP, based at Penn's Positive Psychology Center and Stony Brook's Human Language Lab, has been studying how words used by people reflect their feelings and contentment. In 2014, Johannes Eichstaedt, WWBP's founding research scientist, began to wonder if it was possible for social media to predict outcomes for mental health, especially for depression.
"Social media data contains genome-related markers," says Eichstaedt. "With methods surprisingly similar to those used in genomics, we can combine social media data to find these markers." Depression seems to be something quite detectable in this way, it actually alters the use of media in such a way that diabetes does not. "
Eichstaedt and Schwartz collaborated with their colleagues, Robert J. Smith, Raina Merchant, David Asch and Lyle Ungar of the Penn Medicine Center for Digital Health for this study. Rather than doing what previous studies had done – recruiting participants who reported depression – the researchers identified data from people agreeing to share Facebook statuses and electronic health record information, and then analyzed the statuses. Using machine learning techniques to distinguish people with a formal diagnosis of depression.
"This is preliminary work from our Penn Medicine Center social media registry for digital health," says Merchant, "which associates social media with data from medical records. no data has been collected from their network., the data is anonymous and the strictest levels of confidentiality and security are respected. "
Nearly 1,200 people agreed to provide both digital archives. Of these, only 114 people were diagnosed with depression in their medical records. The researchers then linked each person with a diagnosis of depression to five people who did not have it, to serve as a control, for a total sample of 683 people (with the exception of one for them). insufficient words in status updates). The idea was to create a scenario as realistic as possible to train and test the algorithm of the researchers.
"It's a really difficult problem," said Eichstaedt. "If 683 people present at the hospital and 15% of them were depressed, would our algorithm be able to predict which ones? If the algorithm says that no one is depressed, its accuracy will be 85%. "
To build the algorithm, Eichstaedt, Smith and his colleagues reviewed 524,292 Facebook updates for years leading up to diagnosis for each depressed person and during the same period for control. They determined the most frequently used words and phrases, then modeled 200 subjects to explore what they called "language markers associated with depression". Finally, they compared how and how often the depressed and control participants used such phrasing.
They learned that these markers included emotional, cognitive and interpersonal processes such as hostility and loneliness, sadness and rumination, and that they could predict future depression three months before the first documentation of the illness in a medical file.
"There is a perception that using social media is not good for mental health," says Schwartz, "but this can prove to be an important tool for diagnosing, monitoring, and ultimately treating. used with clinical records, a step towards improving mental health through social media. "
Eichstaedt sees the potential for long-term use of these data as a form of discrete screening. "We hope that one day these screening systems can be integrated into the health care systems," he said. "This tool raises yellow flags and it is hoped that you will be able to directly refer identified individuals to evolving treatment modalities."
Despite some limitations of the study, including its strictly urban sample, and limitations of the domain itself (for example, all diagnoses of depression in a medical record do not meet the standard established by structured clinical interviews) The results offer a new way to discover and get help for those who suffer from depression.
Explore further:
The AI could be used to predict outcomes in people at risk for psychosis and depression
More information:
Johannes C. Eichstaedt et al., "Facebook language predicts depression in medical records" PNAS (2018). www.pnas.org/cgi/doi/10.1073/pnas.1802331115
Source link
