Self-reported negative outcomes of psilocybin users: A quantitative textual analysis

Psychedelic use has increased in recent years, with data indicating rising prevalence of "magic mushroom" consumption and growing scientific interest in psilocybin for conditions such as treatment‑resistant depression and end‑of‑life anxiety. Despite evidence of relatively low physiological toxicity, the literature remains mixed about adverse outcomes: some epidemiological studies have linked hallucinogen use to various mental‑health problems, while others have found no relationship or even lower rates of mental‑health issues among psychedelic users. At the same time, anecdotal and clinical reports describe acute ‘‘bad trips’’ and occasional medical emergencies, and the context of use is thought to shape risk, but systematic analysis of user narratives is limited. This study set out to characterise negative outcomes of psilocybin use from the users' own perspectives by quantitatively analysing self‑reported experience narratives accessed from an online repository. Bienemann and colleagues aimed to identify common semantic clusters in reports of adverse experiences, explore associations between those clusters and contextual or demographic variables (for example dose, concurrent substances, and gender), and to derive findings relevant to harm reduction and further research on psilocybin safety.

The researchers sourced textual reports from the public EROWID repository, selecting narratives filed under the Mushrooms category and specifically the subcategories "health problems", "bad trips", and "train wreck & trip disasters". Reports involving mushrooms where psilocybin was not the main active ingredient (for example Amanita muscaria) were excluded. In total 346 reports were transcribed manually, cleaned for typographical issues and standardised to permit automated text analysis. The average report length was 1,319.5 words. Reports were coded for several contextual variables where available: subcategory (the three EROWID subcategories above), presence of other substances, dosage (dichotomised into below versus above 5 g to reflect a commonly cited threshold for very high doses), route of administration, form of consumption (dried, tea, fresh/pure) and user gender. Inter‑rater coding reliability for items not provided by the site (presence of other substances, dose, route, form and gender) showed total agreement of 92.3% and kappa = 0.85 (p < .001). Quantitative textual analysis employed IRaMuTeQ and R, using text segments (TS) of roughly 40 words as the unit of analysis. The analytic procedures included Descending Hierarchical Analysis (DHA, Reinert method) to derive vocabulary‑based clusters, Correspondence Factor Analysis (CFA) to visualise proximities and oppositions among clusters and categories, and Specificities Analysis to identify word co‑occurrence patterns. Words entered into the analysis were active lexical forms (nouns, verbs, adjectives, adverbs, pronouns and out‑of‑dictionary forms). Inclusion criteria for cluster membership required word frequency above the corpus mean and an associated chi‑square > 3.84. Associations between words, clusters and coded categories were evaluated with chi‑square tests and Cramer's V for effect size, with α set at 0.01 to limit type I error. Ethical considerations noted that the data were public and anonymised, so no ethics committee approval was sought according to national regulation as reported in the extracted text.

The DHA retained 98.4% of the corpus and yielded four distinct clusters of text segments. The extracted text reports the corpus was divided into approximately 12,215 text segments and that most were retained for analysis; however, the counts in the extraction contain a small inconsistency and are not fully clear. Overall, the analysis related 15,788 unique words occurring 453,711 times (mean occurrences per TS ≈ 33.55). Active lexical forms totalled 11,239, of which 2,637 had frequency greater than six. DHA grouped the corpus into four clusters with the following approximate shares of classified text segments: Cluster 1 (30.5%), Cluster 2 (21.4%), Cluster 3 (20.1%) and Cluster 4 (28.1%). CFA showed three large areas with a strong relationship between Clusters 1 and 2. Statistical associations between clusters and report categories or coded variables were as follows: Cluster 1 was significantly associated with the "bad trips" subcategory (χ2(2) = 53.81; p < .001; V = .39) and more frequently reported by female users (χ2(2) = 13.38; p < .001; V = .20). Cluster 2 associated with the "train wrecks & trip disasters" subcategory (χ2(2) = 155.92; p < .001; V = .67), reports involving only mushrooms (χ2(1) = 15.56; p < .001; V = .21) and single doses (χ2(4) = 11.72; p < .001; V = .18). Cluster 4 associated with "health problems" (χ2(2) = 67.61; p < .001; V = .46), use of multiple doses in the same session (χ2(3) = 12.39; p < .001; V not clearly reported in the extraction) and concurrent consumption of substances other than mushrooms (χ2(1) = 17.22; p < .001; V = .22). Cluster 3 did not show significant associations with the coded categories. Thematic inspection characterised the clusters as follows. Cluster 1 encompassed vocabulary related to thought and affect (words such as "feel", "think", "insane", "mind‑race", "death", "fear") and reflected short‑term negative experiences dominated by paranoia and anxiety—hence its link to bad trips and a higher proportion of female reporters. Cluster 2 centred on action and emergency responses (words such as "walk", "call", "ambulance", "police", "hospital") and was interpreted as reflecting episodes requiring external intervention; the authors noted a suggestive but not significant trend linking this cluster to doses above 5 g when using the stricter α = 0.01. Cluster 3 captured perceptual and visual distortions ("eye", "color", "pattern", "light", "vision") that are common psychedelic effects and were not specifically tied to adverse‑outcome categories. Cluster 4 grouped words about administration and context (preparation, dose, form, timing, and other substances such as "weed" or "LSD") and was associated with longer‑term health problems and polysubstance or repeated‑dose use.

Bienemann and colleagues interpret the four clusters as reflecting distinct domains of negative psilocybin‑related experiences reported by users. They argue that bad trips are chiefly characterised by distortions of thought and co‑occurring anxiety or paranoia, suggesting that failures to manage anxiety—whether by contextual measures such as setting or by pharmacological means—are central to acute adverse psychological reactions. The observed association between Cluster 1 and female reporters led the authors to propose possible biological (for example enzymatic or hormonal) or social explanations for sex differences, although they do not claim causal proof. The proximity of Clusters 1 and 2 in the CFA prompted the investigators to note that subjective bad trips and medical or emergency outcomes are closely linked, while acknowledging that the direction of that relationship is unclear: emergencies may follow from intense anxiety, or emergency circumstances may exacerbate or precipitate a bad trip. Cluster 2 was specifically connected to single‑dose events and to reports mentioning transport and emergency services, and the authors highlight an apparent association with high single doses (above 5 g) that did not meet the stricter α threshold. Cluster 3 was framed as representing sensory‑perceptual effects that are common across psychedelic experiences and therefore not specifically indicative of harm. Cluster 4 tied contextual and administration factors to longer‑term health problems, with concurrent use of other substances (for example cannabis or LSD) and multiple doses in a session implicated. The authors suggest that polysubstance use and repeated dosing may underlie some reports of persistent problems and could help explain inconsistent findings across epidemiological studies about psychedelics and mental‑health outcomes. Several limitations are emphasised. Missing values and incomplete contextual detail reduced the ability to analyse setting‑specific influences. Sampling was non‑probabilistic and self‑selected, so reports on EROWID may not represent the broader population of psilocybin users; furthermore, anonymous online reports cannot verify the actual substance used or its purity, and illegality and black‑market factors could confound adverse outcomes. The study is exploratory and hypothesis‑generating rather than confirmatory. The investigators propose that their findings can inform harm‑reduction messaging and design of experimental studies to clarify causal relationships between context, dose, polysubstance use and short‑ or long‑term adverse effects.

Using quantitative textual analysis of 346 self‑reports from an online repository, the study identified four semantic clusters of negative psilocybin experiences: thought‑ and anxiety‑centred bad trips, action‑oriented emergency episodes, sensory‑perceptual distortions, and contextual/administrative aspects linked to longer‑term problems. The authors conclude that distortions in thought and unmanaged anxiety are central to bad trips, that high single doses were associated with emergency‑type reports, and that multiple dosing and concurrent substance use related to longer‑term health complaints. They stress the exploratory nature of these results, note important limitations (missing data, self‑selection, unverifiable substances, and legal‑market confounds), and recommend that the findings be used to guide harm‑reduction efforts and generate hypotheses for controlled research on psilocybin safety.