
Despite being the most common mental illness in the world, depression remains an underdiagnosed condition that can manifest devastating symptoms. Affecting over 300 million people globally, it is the leading cause of disability as per the World Health Organization. Though many patients are being properly treated for this mental health condition, there are a high number of undiagnosed cases of depression, particularly in younger populations and men.
Broadening the scope of current depression screening tools is essential in diagnosing and treating the condition more promptly, and emerging research has shown that monitoring social media activity may be effective in doing so.
Detecting Depression Through Facebook Statuses
A 2018 study published in the Proceedings of the National Academy of Sciences found that using artificial intelligence (AI) to analyze the language in Facebook posts can generate accurate predictions of one’s risk for depression. These researchers accessed the history of Facebook status posts from 683 consenting patients at an emergency department to accomplish this, with 114 of these patients having been diagnosed with depression. The language in Facebook statuses posted before their diagnosis of depression was analyzed in this work. Each of these depressed patients was matched with five patients without depression to test the programs accuracy.
Analyzing 524,292 posts, the researchers were able to identify specific language markers that were associated with depression. When the machine learning algorithm was primed with these markers, it was able to identify indicators of depression in participants from posts up to three months before they were diagnosed with the condition. The team noted that these predictions were most accurate when made six months prior to the formal diagnosis. In addition to the length and timing of their posts, early warning signs of depression included use of words such as “alone”, “ugh” or “tears”. Increased prevalence of first-person pronouns such as “me” and “I” was correlated to increased depression risk as well, being that these terms indicate preoccupation with oneself.
“Our results show that Facebook language-based prediction models perform similarly to screening surveys in identifying patients with depression when using diagnostic codes in the EMR (electronic medical record) to identify diagnoses of depression,” the authors concluded.
A Mental Health Classifier Built Using Twitter Data
A similar study used crowdsourcing to identify Twitter users who reported to have been diagnosed with clinical depression. Social media behaviors denoting social engagement, emotion, language, personality shifts, and mention of antidepressants were monitored for over one year prior to each patient’s formal depression diagnosis. A classifier was created from this information to estimate one’s risk for depression before the reported diagnosis, as in the prior study.
The team found that social media content contains useful information that can indicate one’s risk of developing depression, including the following factors:
- Decreased social activity
- Raised negative affect
- Stronger relational and medical concerns
- Increased expression of religious involvement
The authors note that their “methods may be useful in developing tools for identifying the onset of major depression, for use by healthcare agencies; or on behalf of individuals, enabling those suffering from depression to be more proactive about their mental health.” Their work was published in a paper titled “Predicting Depression via Social Media”.
Using AI to Screen Instagram Photos for Depression
Aside from the use of machine learning in analyzing text from social media such as Facebook and Twitter, researchers have analyzed images from Instagram to screen for depression as well. Using Instagram data from 166 participants, the researchers applied machine learning AI to identify significant indicators of depression. These features were deduced from 43,950 Instagram photos using analysis of colors, metadata, and facial recognition technology.
The screening models created from this data outperformed the primary care physicians’ average rate of successful diagnosis for depression. This trend remained even when the analysis was limited only to Instagram posts made before the individuals were diagnosed with depression. The researchers also had humans rate the photos in terms of happiness, sadness, and similar attributes, but found that these ratings were weaker than the AI technique in predicting depression. This work was published in 2017 in EPJ Data Science.