JMIR Publications

ABSTRACT

Background:

Depression is one of the most common mental disorders, being a leading cause of disability worldwide and selective serotonin reuptake inhibitors (SSRIs) are the most commonly prescribed antidepressants. Some people share information about their experiences with prescribed antidepressants in social media platforms such as Twitter. The analysis of the messages posted by Twitter users under SSRI treatment can yield useful information on how these antidepressants affect users’ behavior.

Objective:

This study aims to compare the behavioral and linguistic characteristics of the tweets posted while users were taking SSRIs, in comparison to the tweets posted by the same users when they were not taking this medication.

Methods:

In the first step, the timelines of Twitter users mentioning SSRIs in their tweets were selected using a list of 128 generic and brand names of SSRIs. In the second step, two datasets of tweets were created, the in-treatment dataset (made up of the tweets posted throughout the 30 days after mentioning one SSRI) and the unknown-treatment dataset (made up of tweets posted more than 90 days before and more than 90 days after any tweet mentioning one SSRI). For each user, the changes in behavioral and linguistic features between the tweets classified in these two datasets were analyzed. 186 users and their timelines with 668,842 tweets were finally included in the study.

Results:

The number of tweets generated per day by the users when they were in-treatment was higher than when they were in the unknown-treatment period (P = .001). When the users were in-treatment, the mean percentage of tweets posted during the day-time (from 8:00 am to midnight) increased with respect to unknown-treatment periods (P = .002). The number of characters and words per tweet was higher when the users were in-treatment (P = .03 and P = .02 respectively). Regarding linguistic features, the percentages of the first and second-person singular pronouns were higher when users were in treatment (P = .008 and P = .004, respectively).

Conclusions:

The detection of behavioral and linguistic changes when users with depression are taking medication, can provide interesting insights for monitoring the evolution of this disease as well as offering additional information related to the treatment adherence. This information may be especially useful in patients who are receiving long-term treatments such as people suffering from depression.