BACKGROUND: An increasing number of online support groups are providing advice and information on topics related to mental health. This study aimed to investigate the needs that internet users meet through peer-to-peer interactions.

METHODS: A search of 4 databases was performed until August 15, 2022. Qualitative or mixed methods (ie, qualitative and quantitative) studies investigating interactions among internet users with mental disorders were included. The φ coefficient was used and machine learning techniques were applied to investigate the associations between the type of mental disorders and web-based interactions linked to seeking help or support.

RESULTS: Of the 13,098 identified records, 44 studies (analyzed in 54 study-disorder pairs) that assessed 82,091 users and 293,103 posts were included. The most frequent interactions were noted for people with eating disorders (14/54, 26%), depression (12/54, 22%), and psychoactive substance use disorders (9/54, 17%). We grouped interactions between users into 42 codes, with the empathy or compassion code being the most common (41/54, 76%). The most frequently coexisting codes were request for information and network (35 times; φ=0.5; P<.001). The algorithms that provided the best accuracy in classifying disorders by interactions were decision trees (44/54, 81%) and logistic regression (40/54, 74%). The included studies were of moderate quality.

CONCLUSIONS: People with mental disorders mostly use the internet to seek support, find answers to their questions, and chat. The results of this analysis should be interpreted as a proof of concept. More data on web-based interactions among these people might help apply machine learning methods to develop a tool that might facilitate screening or even support mental health assessment.