JMIR Publications

ABSTRACT

Background:

Studies on the link between social media and mental health have primarily focused on depression. However, few review studies have been conducted, and most of them did not focus on depression, searched only a few databases, only analyzed language features, and did not apply machine learning.

Objective:

This systematic review and meta-analysis evaluated the effect of social media texts overall and of demographic, linguistic, activity, and temporal features in predicting depression through machine learning.

Methods:

We searched articles from 11 databases from inception to August 2023. Data analysis was performed using Comprehensive Meta-Analysis version 3. We used a random-effects model to pool the effect sizes with 95% CIs. Study heterogeneity was evaluated using forest plots and P values in the Cochran Q test. Moderator analysis was performed to identify the sources of heterogeneity. We used the Begg–Mazumdar rank correlation and Egger’s tests to assess publication bias, and we performed a sensitivity analysis using the leave-one-out method to determine result stability.

Results:

We included 36 studies on machine learning–based depression prediction using social media textual data. We observed a significant overall correlation between social media texts and depression, with a large effect size (r = 0.630, 95% CI: 0.565–0.686). We noted the same correlation and large effect size for demographic (largest effect size; r = 0.642, 95% CI: 0.489-0.757), activity (r = 0.552, 95% CI: 0.418-0.663), linguistic (r = 0.545, 95% CI: 0.441-0.649), and temporal features (r = 0.531, 95% CI: 0.320-0.693). The social media platform type (public or private, P < .001), machine learning approach (shallow or deep, P = .048), and use of outcome measures (yes or no, P < .001) were significant moderators. Sensitivity analysis revealed no change in the results, indicating result stability. The Begg–Mazumdar rank correlation (Kendall’s Tau b = 0.22063, P = .058) and Egger’s test (t = 1.28696, P = .207) confirmed the absence of publication bias.

Conclusions:

Social media textual content can be a useful tool for predicting depression. To maximize the accuracy of depression prediction models, it is important to consider demographic, linguistic, and social media platform activity as well as temporal features. Additionally, the effects of social media platform type, machine learning approach, and use of outcome measures in depression prediction models need attention. Clinical Trial: PROSPERO, registration number CRD42023427707