Examining HiTOP Through Naturalistic Symptom Descriptions in Anonymous Online Mental Health Forums: A Mixed-Methods Study Protocol
AbstractBackground: The Hierarchical Taxonomy of Psychopathology (HiTOP) organises covarying symptoms and maladaptive traits into transdiagnostic dimensions to address limitations of categorical diagnosis. Naturalistic symptom descriptions may complement structured assessments. Objective: This protocol describes a reproducible mixed-methods workflow to derive symptom indicators from forum posts and examine their organisation relative to HiTOP. Methods and analysis: The target sample is 80,000 public Reddit posts. After local de-identification, a large language model (LLM) will extract symptom labels and supporting paraphrases. Two human experts will independently review outputs for up to 250 posts. Three embedding models will flag potentially equivalent symptom labels for local LLM assessment. Two experts will qualitatively review the most frequent canonical symptom labels (at least 1,000) and develop a codebook of symptom categories, continuing until a prespecified saturation criterion is met. An LLM will assign canonical symptom labels to these categories, producing binary post-level indicators. Multiverse-inspired structural analyses will compare network, component, factor, and clustering representations across discovery, internal tuning, and holdout partitions. Evaluation will assess interpretability, stability, and agreement with expert mapping to HiTOP. Supplementary analyses will include agent-assisted exploration, codebook application to DepressionEmo and the Reddit Mental Health Dataset, and pooled integrative analysis that may include additional comparable corpora.Discussion: The study will examine whether symptom co-expression within posts forms interpretable patterns corresponding to HiTOP.Ethics: The Ethics Committee of Charité – Universitätsmedizin Berlin approved the study (EA2/274/25). The study was preregistered on 8 May 2026 (German Clinical Trials Register, DRKS00040244).Keywords: Hierarchical Taxonomy of Psychopathology, natural language processing, large language models, Reddit, mixed methods, study protocol
Authors
- Drin Ferizaj (ORCID: https://orcid.org/0000-0002-7132-7778)
Publication Details
- Journal
- PsyArXiv (OSF Preprints)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.31234/osf.io/ahrts
- Primary Topic
- Mental Health via Writing
- Type
- preprint