Mental Health Consultations on College Campuses: Examining the Predictive Ability of Social Media

Background: The mental health of college students is a growing concern, and gauging the mental health needs of college students is a critical problem that is difficult to assess in real-time and in scale. To address this gap, researchers and practitioners have encouraged the use of passive technologies. Given the ubiquity and widespread use of social media among the college student demographic, social media has also been considered to be a viable “passive sensor.” However, the construct validity and in-practice reliability of human-centric computational assessments of mental health constructs with social media data remain largely unexplored. Towards this goal, we study how assessing the mental health of college students using college student-specific social media data correspond with ground-truth data of on-campus mental health consultations.   Method: We conducted an observational study on the data of a large U.S. public university. We collected over 66,000 posts by 18,401 users from the university’s Reddit community, and we examined if human-centric assessments of mental health on social media data corresponded with the ground-truth data of mental health consultations as obtained from the university’s health center over a period of over five years between 2011 and 2016. First, we adopted machine learning and natural language analysis methodologies to measure symptomatic mental health expressions of depression, anxiety, stress, suicidal ideation, and psychosis on the social media data, and compared the prevalence of these expressions with our ground-truth. We built seasonal autoregressive integrated moving average (SARIMA) models of forecasting on-campus mental health consultations. We also examined the language of social media expressions using unsupervised language modeling and psycholinguistic characterization to help explain the validity of social media data in predicting the mental wellbeing of university students.   Findings: We found that mental health expressions on social media show a statistically significant relationship with ground-truth college mental health data. Further, incorporating social media data leads to time series predictions with r=0.86 and SMAPE=13.30, which outperforms forecasting models without social media data by 41%. Our language analysis reveals that social media discussions during high mental health utilization months consisted of discussions related to academics, career, and other stressful events on campus, whereas months of low mental health consultations saliently corresponded to expressions of positive affect, collective identity, socialization, and better mental wellbeing. Interpretation: Our study reveals that social media data can improve our understanding of college students’ mental health, particularly their mental health treatment and support seeking needs. Funding: KS and MDC were partly supported through NIH grant #R01MH117172 to MDC. Declaration of Interest: None to declare.