Northwell, IBM study: Analysis of Facebook activity can predict mental health disorders

Machine learning algorithms can analyze Facebook users' activity to predict the development of schizophrenia spectrum disorders and mood disorders, according to research published Dec. 3 in Nature Partner Journals Schizophrenia.

Researchers at The Feinstein Institutes for Medical Research, New Hyde Park, N.Y.-based Northwell Health's research arm, partnered with IBM computer scientists to examine 3.4 million Facebook messages and 142,390 images from 223 consented participants, some of whom had been diagnosed with schizophrenia spectrum disorders and mood disorders. The data spanned 18 months prior to the first psychiatric hospitalization for participants who went on to receive a mental health diagnosis.

The research team found that machine learning algorithms could accurately predict the development of schizophrenia spectrum disorders and mood disorders more than a year prior to a patient’s first hospitalization and diagnosis. In the study, they said their findings support the adoption of social media and digital activity monitoring as a tool to facilitate early intervention for psychiatric conditions.

Below are some of the study's key findings:

1. Participants with schizophrenia spectrum disorders and mood disorders were more likely to swear than participants who did not have a mental health diagnosis.
   
   
2. Participants with schizophrenia spectrum disorders and mood disorders posted images with smaller heights and widths than participants who did not have a mental health diagnosis.
   
   
3. Participants with schizophrenia spectrum disorders used more perception words, such as "feel," "see" and "hear," than other participants.
   
   
4. Closer to hospitalization, participants with schizophrenia spectrum disorders increased their use of punctuation compared to other participants.
   
   
5. Approaching hospitalization, participants with mood disorders increased their use of negative emotion words compared to other participants.
   
   
6. Participants with mood disorders used more words related to blood, pain and other biological processes than other participants.
   
   
7. Images posted by participants with mood disorders contained more blues and less yellows than other participants.