JMIR Publications

ABSTRACT

Background:

TikTok has become one of the leading sources of mental health information among children, adolescents, and young adults. Although the platform has increased access to psychoeducational content, concerns remain regarding the prevalence of misinformation and disinformation related to child and adolescent mental health and their potential influence on public understanding and health-related decision-making.

Objective:

This study aimed to characterize child and adolescent mental health misinformation and disinformation on TikTok, identify the mental health topics most frequently associated with misleading content, determine predictors of misinformation using machine learning, and develop evidence-informed recommendations for improving the quality of psychoeducational videos.

Methods:

We conducted a cross-sectional content analysis of 122 publicly available TikTok videos addressing child and adolescent mental health, selected from an existing dataset of 1,000 videos. Videos were coded using a structured framework assessing content characteristics, intent, authenticity, and clinical variables. Descriptive statistics, chi-square tests, and one-way analyses of variance were used to explore associations between video characteristics and information quality. A multilayer perceptron classifier was trained using an 80:20 train-test split with five-fold cross-validation to identify predictors of misinformation and disinformation through permutation feature importance. Video transcripts were also analyzed using Braun and Clarke's reflexive thematic analysis.

Results:

Of the 122 videos analyzed, 17.2% (n=21) contained misinformation and 1.6% (n=2) contained disinformation, whereas 49.2% (n=60) were classified as primarily fact-based. Autism spectrum disorder represented the topic with the highest proportion of misinformation (40.8%), followed by impulse control disorders (33.3%) and mood disorders (30.7%). More than half of presenters (51.6%, n=63) did not disclose professional credentials, and only 7.4% (n=9) of videos cited external scientific sources. Machine learning identified fact-based content (relative importance=0.027) as the strongest protective factor against misinformation, whereas opinion-based content (0.015), hoax-related framing (0.018), and greater engagement metrics, particularly shares and favorites (0.027), were the strongest predictors of misinformation. The classifier achieved an accuracy of 0.83 and an F1 score of 0.74 for misinformation detection and an accuracy of 0.91 and an F1 score of 0.54 for disinformation detection. Qualitative analysis identified five major themes in viewer comments: personal experience and validation, diagnosis and misdiagnosis, criticism of mental health care, support and encouragement, and treatment experiences.

Conclusions:

Mental health misinformation is common within child and adolescent mental health content on TikTok, particularly for neurodevelopmental conditions. Content characteristics, rather than professional credentials alone, appear to be the strongest predictors of misinformation. These findings provide evidence to inform recommendations for creating higher-quality psychoeducational content and support the development of digital health literacy interventions aimed at young social media users. Clinical Trial: Not applicable. This study was a secondary analysis of publicly available social media content and did not involve a clinical trial.