Post-acute psychological effects of classical serotonergic psychedelics: A systematic review and meta-analysis

Psychedelic substances have a long history of ritual and medicinal use and re-emerged as a focus of scientific research over the past two decades after a hiatus following mid-20th century legislative restrictions. Classical serotonergic psychedelics share a common pharmacology (5-HT2A receptor agonism) and characteristic subjective effects; recent clinical and non-clinical studies have evaluated psilocybin, ayahuasca (DMT with MAO-A inhibitors), and LSD for outcomes ranging from psychiatric symptoms to well-being, spirituality and personality. Narrative reviews and a small number of meta-analyses have suggested potential therapeutic benefit and acceptable short-term safety, but a comprehensive quantitative characterisation of post-acute psychological effects across outcome domains remained lacking. Goldberg and colleagues designed this study to fill that gap. The paper reports a systematic review and meta-analysis of experimental studies (randomised controlled trials and single-group pre–post designs) that administered psilocybin, ayahuasca or LSD in controlled settings and measured psychological outcomes at least 24 hours after administration. The stated aims were to quantify post-acute effects across multiple outcome domains, test study-level moderators (psychedelic type, clinical versus non-clinical samples, presence of behavioural support, percentage female), and evaluate adverse effects and risk of bias across the literature.

The review followed PRISMA guidance and was preregistered, with some deviations specified by the authors: the analysis was restricted to post-acute outcomes (≥24 hours), moderation by specific diagnoses was avoided due to limited data, and outcomes were aggregated into coherent categories. Eligible studies administered psilocybin, ayahuasca or LSD in experimental (not naturalistic) settings, reported at least one psychological outcome assessed ≥24 hours post-administration, and provided data necessary to compute effect sizes. Both clinical and non-clinical samples and both between-group (including placebo-controlled RCTs and crossovers) and within-group pre–post designs were eligible. Studies that reported only post-treatment data without baseline or appropriate control comparisons were excluded. Six electronic databases were searched (PubMed, CINAHL, PsycINFO, Web of Science, Scopus, Cochrane) for studies from 1990 up to the search window (23–31 October 2019), supplemented by hand-searching recent reviews. Two reviewers independently screened records and extracted data; inter-rater reliability was reported as good to excellent. Extracted items included study design, psychedelic type and dose, control condition, inclusion criteria, adverse events, timing of post-treatment and follow-up assessments, presence of behavioural support (preparation/therapy), sample demographics, country, retention, and items necessary for Cochrane risk-of-bias assessment. Outcomes were grouped into 14 domains (adverse effects; targeted psychiatric symptoms; depression in samples with depression; negative affect; positive affect; social outcomes; behaviour; existential/spiritual outcomes; mindfulness; and the Big Five personality traits: openness, neuroticism, extraversion, agreeableness, conscientiousness). Risk of bias within studies was rated across five Cochrane domains (selection, performance, detection, attrition, reporting). Effect sizes were computed using standard meta-analytic approaches: within-group pre–post and pre–follow-up effects were calculated assuming a correlation of r = 0.50 between timepoints when needed; between-group effects were computed either as the difference between within-group effects or as Cohen’s d when pre–post data were not available. Summary statistics were converted to Hedges’ g to correct for small-sample bias and pooled using random-effects models. Heterogeneity was quantified with I2 (the percentage of total variability due to between-study heterogeneity). Publication bias was explored with trim-and-fill and fail-safe N calculations, although the authors treated these tests as exploratory due to small numbers in some models. Four moderators were pre-specified: psychedelic type (psilocybin versus LSD/ayahuasca), clinical sample status, presence of behavioural support, and percentage female. Sensitivity analyses excluded identified outliers using a function that flags studies whose confidence intervals do not overlap the omnibus effect CI.

Study selection yielded 14,591 citations; after deduplication and screening, 34 reports representing 24 unique samples and 549 participants were retained (studies published 2006–2020). Study designs were evenly split: 50.0% single-group pre–post, 16.7% within-subject RCTs (crossovers), and 33.3% between-group RCTs. Psilocybin was the most studied agent (58.3% of studies), with ayahuasca in 25.0% and LSD in 16.7%. The average timing of the first post-test was 5.54 weeks (SD = 6.48; range 0–26 weeks); 54.2% of studies included at least one follow-up, with the last follow-up on average 53.34 weeks post-treatment (SD = 64.25; range 3–234.9). Mean sample size was 22.88 (SD = 17.42; range 6–85), mean age 42.13 years, and samples were 51.5% female. Approximately 45.8% of studies recruited clinical samples, most commonly depression (k = 4) and life‑threatening illness with comorbid anxiety/depression (k = 3). Risk-of-bias assessments varied by design, with single-group studies lacking randomisation and blinding features. Domains most at risk were blinding of participants/personnel and blinding of outcome assessment; selective reporting was often rated unclear. Concerning adverse events, 79.2% of studies reported adverse effects; none reported serious adverse events such as death or hospitalisation. Transient effects commonly reported included headache, anxiety, nausea and increased blood pressure. Where longer-term adverse measures were available, meta-analytic within-group effects suggested reductions in adverse symptoms at post-treatment and follow-up (Hedges’ g = 0.40 and 0.50, respectively)—note that a positive g was coded to indicate improvement (reduction in adverse effects). Within-group meta-analyses showed statistically significant improvements across multiple domains at both post-treatment and follow-up. Domains with consistent beneficial effects included targeted psychiatric symptoms in clinical samples, depression in depressed samples, negative affect, positive affect, social outcomes, and existential/spiritual outcomes; within-group Hedges’ g estimates ranged approximately from 0.44 (positive affect) up to 2.06 (depression in depressed samples). Behaviour and mindfulness showed improvements at post-treatment though follow-up estimates were not always available. Most Big Five personality traits did not change, with the exception of a small increase in openness. Heterogeneity tended to be substantial (I2 > 50% for many models). Between-group (placebo-controlled) analyses at the longest follow-up likewise favoured psychedelics, with moderate-to-large effects for targeted psychiatric symptoms, negative affect, positive affect, social outcomes, behaviour, and existential/spiritual outcomes; between-group Hedges’ g values reported ranged from 0.84 to 1.16. No reliable between-group effects were observed for personality dimensions. Again, heterogeneity was generally high. Across-study bias assessments detected funnel-plot asymmetry in eight models; trim-and-fill adjustments left most statistically significant findings intact, but one model (within-group pre–post social outcomes) became non-significant after adjustment (adjusted g = 0.43). Fail-safe Ns varied from 0 to 803; applying Rosenberg’s guideline (fail-safe N > 5n + 10) indicated that some effects (within-group adverse effects, social outcomes, openness, mindfulness, and between-group behaviour) were not robust to potential publication bias. Moderator analyses were constrained by limited numbers of studies for many outcomes. Clinical samples tended to show larger improvements on some outcomes (negative affect, positive affect, adverse effects, existential/spiritual outcomes, extraversion). Psychedelic type generally did not moderate effects, except psilocybin produced larger within-group pre–post increases in mindfulness. Presence of behavioural support did not meaningfully moderate effects, and percentage female was not a consistent moderator (one exception: higher female percentage associated with smaller increases in extraversion at pre‑follow‑up). Sensitivity analyses excluding outliers did not change significance and effect-size changes were small (Δg ≤ 0.26).

Goldberg and colleagues present what they describe as the first comprehensive meta-analysis focusing on post-acute psychological effects of classical serotonergic psychedelics. Despite a modest evidence base (k = 34 reports, 24 unique samples, n = 549), the study found moderate-to-large improvements in a range of psychological domains. Most notable was a large between-group effect for targeted psychiatric symptoms in clinical samples (Hedges’ g = 1.08), which the authors contextualise as comparable to or larger than effect sizes often reported for psychotherapy versus waitlist or for antidepressants versus placebo, and robust to publication-bias adjustment and outlier exclusion. Psychedelics also showed favourable between-group effects on affective, social, behavioural, and existential/spiritual measures, while evidence for enduring changes in personality and mindfulness was weaker and more sensitive to publication-bias considerations. The authors emphasise important caveats. Risk of bias within included studies was common, especially for blinding and selective reporting, and many studies used single‑group designs without randomisation. Difficulty blinding psychoactive interventions is a prominent concern for causal attribution. Small sample sizes, heterogeneity in doses, provision of behavioural support, and other design features limited the precision and generalisability of pooled estimates. The participant pool lacked racial/ethnic diversity in many studies and selection processes may have introduced expectancy or other sampling biases. Trim-and-fill and fail-safe N analyses suggested some models were vulnerable to publication bias, and the authors note that many studies were not preregistered or did not explicitly report intention‑to‑treat analyses, leaving attrition and selective reporting as addressable sources of bias. In terms of implications, the investigators argue the results support further, larger-scale, carefully controlled randomised trials to clarify efficacy for specific clinical conditions (notably depression and anxiety) and to define the role and necessary dose of accompanying behavioural support. They recommend extended follow-up, inclusion of populations formerly excluded from trials (for safety evaluation), naturalistic and population-based designs to examine real-world safety and persistence of effects, and more consistent preregistration to reduce selective reporting. While no clear evidence of persistent adverse effects emerged in the analysed studies, the authors call for vigilant adverse-event monitoring in future work, especially when broadening inclusion criteria beyond the relatively screened populations represented in the current literature.