Psychiatric risks for worsened mental health after psychedelic use

Recent years have seen a renaissance in research into psychedelic-assisted interventions, with prior studies indicating potential safety and therapeutic benefit across conditions such as major depressive disorder, addiction, obsessive-compulsive disorder and end-of-life anxiety. However, the extracted text notes lingering uncertainties about rare but clinically meaningful adverse psychological responses, especially because modern clinical trials typically exclude people with histories of psychosis, bipolar disorder or personality disorders and apply careful screening and therapeutic support. The authors frame concerns that limited sample sizes, brief follow-ups, selective recruitment and measurement approaches in existing research may have under-sampled or missed low-prevalence but important negative outcomes following psychedelic use. Marrocu and colleagues set out to identify and characterise a ‘‘bottom margin’’ of individuals who experienced a clinically meaningful decline in psychological well‑being after an intentioned psychedelic experience in naturalistic contexts. Using pooled prospective survey data from three independent studies, they ask whether self‑reported prior psychiatric diagnoses predict membership in this negative‑response subset, operationalised as a decline on the Warwick‑Edinburgh Mental Well‑Being Scale (WEMWBS) at 4 weeks post‑use greater than the instrument’s standard error of measurement (SEM). The stated aim is to identify potentially vulnerable populations in order to inform risk‑mitigation for psychedelic use and therapy.

The investigators pooled data from three separate prospective, primarily online cohorts in which participants completed at least five surveys across retrospective and prospective timepoints. The extracted text contains an inconsistency in the reported analytic sample size: the Introduction states a pooled sample of 806 respondents, whereas a later Demographics section reports N=807; the baseline, 2‑week and 4‑week completion counts are reported as N=2008, N=1049 and N=881 respectively, reflecting substantial attrition. Surveys provided baseline demographic information (approximately 2 weeks prior to a planned psychedelic experience), a one‑day post‑session assessment of the subjective experience, and well‑being measures at baseline, 2 weeks and the key endpoint of 4 weeks after use. Psychological well‑being was measured with the 14‑item Warwick‑Edinburgh Mental Well‑Being Scale (WEMWBS), which yields scores from 14–70 and captures positive affect, functioning and social aspects of well‑being. Change scores were calculated as the difference between the 4‑week score and baseline. To classify individual meaningful change, the authors used a standard error of measurement (SEM) approach previously applied to WEMWBS: SEM = SD × √(1 − reliability). They computed Cronbach’s α within their pooled sample (α = 0.91), yielding an SEM threshold of ±2.82 points; decreases beyond −2.82 were labelled ‘‘negative response’’ and increases beyond +2.82 labelled ‘‘positive response’’. Values within ±2.82 were coded as no response. Prediction of negative responses was tested with logistic regression. The dependent variable was binary (negative responder = 1; positive/no response = 0). The main predictor of interest was self‑reported prior diagnosis of a personality disorder (PD). Additional models included covariates for other self‑reported psychiatric diagnoses (psychotic disorder, ADHD, major depressive disorder, bipolar disorder, anxiety disorder, eating disorder, OCD) and baseline WEMWBS to account for potential ceiling/floor effects. Odds ratios (ORs) were derived from logistic coefficients; significance was set at p ≤ 0.05. Data processing used Matlab (2019b) and RStudio (2020).

Using the SEM criterion (±2.82 points), the authors classified 55.5% (448) of the analysed cohort as positive responders, 28.1% (227) as non‑responders and 16.4% (132) as negative responders at 4 weeks post‑psychedelic use; the 16.4% constitutes their operational ‘‘bottom margin’’. Cohort mean WEMWBS scores were reported as 47.0 (±9.2) at baseline, 52.0 (±8.75) at 2 weeks and 51.6 (±8.86) at 4 weeks; the paper notes a recent English adult population mean of 49.9 for context. When stratifying by self‑reported psychiatric diagnostic history, the highest proportion of negative responders was observed among those reporting a prior personality disorder diagnosis: 31.2% (reported as 5 cases, implying a PD subgroup of 16 individuals). Psychotic disorders showed 25% negative responders (N=1). The authors also report 12.9% negative responders among those with any history of psychiatric illness and 18.11% among those reporting no psychiatric history, with the overall 16.4% negative responder rate in the total cohort. Trajectories in the PD subgroup differed from the rest of the sample. Baseline WEMWBS in the PD group averaged 43.31 (±12.47) compared with 47.08 (±9.12) in the overall cohort. Both the PD subgroup and the overall cohort showed significant increases at 2 weeks after use (PD: +5.67, p = 0.01; overall: +4.91, p < 0.001). By 4 weeks, the overall cohort was relatively stable (−0.34, p = 0.10) while the PD group trended downward (−3.6, p = 0.08); the between‑group difference at 4 weeks was not statistically significant by Welch’s t‑test (p > 0.05). Within the PD subgroup, five individuals met the negative‑responder criterion and showed significant declines from baseline (≤ −2.82 points). Logistic regression analyses yielded progressively larger PD‑associated effects as additional covariates were included. The univariate model (Model 1) produced a non‑significant coefficient (b = 0.866, p = 0.114) corresponding to OR = 2.38. A multivariate model controlling for comorbid psychiatric diagnoses (Model 2) gave b = 1.208, p = 0.065, OR = 3.35 (non‑significant at conventional thresholds). The most complete model (Model 3), which also adjusted for baseline well‑being, reported b = 1.425, p = 0.043, OR = 4.16, indicating a statistically significant, greater‑than‑four‑fold relative likelihood of a long‑term WEMWBS decline among those self‑reporting a PD diagnosis. The authors caution that the absolute probability of negative response remained low, as reflected by a significantly negative model constant (constant = −4.314, p < 0.001). Exploratory analyses of personality traits using the Ten‑Item Personality Inventory (TIPI) found negative correlations between the self‑reported PD diagnosis and items indexing emotional stability (inverse neuroticism) and agreeableness; the authors interpret this pattern as consistent with a borderline‑like personality profile.

Marrocu and colleagues interpret their findings as preliminary evidence that self‑reported personality disorder may be associated with an elevated relative risk of clinically meaningful worsening in psychological well‑being following naturalistic psychedelic use. They emphasise that 16.4% of the pooled cohort met the study’s operational criterion for negative response at 4 weeks, and that the highest proportion of such cases occurred in the small subgroup reporting a personality disorder (31%, 5 of 16). The authors position this work as filling a gap in research on baseline predictors of iatrogenic psychological outcomes after psychedelic experiences, particularly for groups commonly excluded from clinical trials. The paper highlights a distinctive temporal pattern in the PD subgroup: both PD and non‑PD participants tended to improve at 2 weeks, but some in the PD group showed marked declines by 4 weeks. The authors suggest this could reflect fragility in short‑term gains among people with borderline‑like profiles and note analogues in other studies where brief improvements returned to baseline or where acute distress emerged after study termination. They argue that psychological support and longer‑term care may be particularly important when working with people with PD, and caution that brief, single‑dose psychedelic protocols may not provide sufficient aftercare for this population. Several important limitations are acknowledged. The study relies on online, self‑report observational data with notable attrition: the extracted text reports a 56% drop‑out between baseline and the 4‑week endpoint and similar attrition within the PD subgroup, which reduces statistical power and raises concerns about attrition bias. Self‑reported diagnostic history may be inaccurate or outdated, and the PD item did not capture specific personality disorder subtypes. The authors also acknowledge potential analytic issues, including collider bias from including multiple psychiatric diagnoses as covariates, and that WEMWBS is a generic well‑being measure rather than a clinical instrument for severe adverse events. They note that their ‘‘bottom margin’’ or extreme‑value focus intentionally selects outliers and thus emphasises potential harms at the expense of average effects; for balance they point out that approximately half of participants reporting PD experienced clinically meaningful improvement. In terms of implications, the investigators recommend caution in extrapolating naturalistic findings to controlled clinical settings, but they argue that their results justify further rigorous research into predictors of adverse responses, improved screening and risk‑mitigation strategies, and the collection of observational data alongside policy changes that increase access to psychedelics. They propose that predictive modelling could support personalised approaches to psychedelic medicine and stress the importance of considering both potential benefit and harm when evaluating suitability for psychedelic interventions.