Prediction of psilocybin response in healthy volunteers

Responses to classical hallucinogens such as psilocybin vary substantially between and within individuals even at fixed doses, a phenomenon commonly attributed to non-pharmacological factors summarised as "set" (personality, preparation, current mood) and "setting" (physical, social, cultural context). Earlier literature, much of it older and methodologically limited, has implicated personality, pre-session mood, social support, expectations, age, body morphology, group size, and prior drug experience as potential determinants of hallucinogen responses, but most studies examined few predictors at a time and relied on simple correlations rather than multivariable approaches that could adjust for confounding and compare effect sizes across predictors. This paper pools raw data from 23 controlled experimental studies to investigate the relative importance of 24 predictor variables (demographic, personality, recent psychological state, prior drug use, session environment, and drug dose) for acute subjective responses to psilocybin in healthy volunteers. The primary goal was to identify which non-pharmacological factors meaningfully predict different dimensions of altered states of consciousness, and to compare their effect sizes with that of drug dose, using modern statistical methods (multiple imputation, linear mixed models, bootstrap variable selection). This work is intended to inform safety screening, standardisation of experimental designs, covariate adjustment in trials, and mechanistic hypotheses linking individual differences to neurobiology.

Data were pooled from 23 experimental studies conducted between 1992 and 2011 at a single research facility; all used placebo-controlled within-subject designs. Depending on the original study, each subject received placebo and between one and four different doses of psilocybin across 2–5 sessions, with sessions separated by at least two weeks. For the present analysis only sessions in which psilocybin was administered alone and at doses of at least 115 mg/kg po were included, yielding 409 psilocybin administrations in 261 subjects; the administered dose ranged from 115 to 315 mg/kg (mean + SD: 214 + 63 mg/kg) as reported in the extracted text. Most studies (n = 16) were randomised and double-blind, while seven earlier or pilot studies were open-label. Some original studies also tested receptor blockers, but those sessions were excluded from this pooled analysis. Participants were screened and excluded for personal or family history of schizophrenia, major mood or borderline personality disorders, substance abuse, neurological disorders, or abnormal laboratory findings; many studies additionally excluded subjects scoring highly on Emotional Lability. Predictor variables were divided into subject-level (constant across sessions: age, sex, BMI, years of education, personality measures, lifetime drug use, psychological problems) and session-level (varying across sessions: current mood state by adjective lists, drug dose, environment such as PET vs non-PET, time of assessment). Personality instruments included the Zuckerman-Kuhlman Personality Questionnaire (ZKPQ), Tellegen Absorption Scale (TAS), Freiburg Personality Inventory (FPI) used as auxiliary data for imputation, PASI (a subscale of a hallucination-prediction inventory) and the SCL-90-R Global Severity Index (GSI). Acute outcomes were measured with the OAV and 5D-ASC questionnaires, yielding the primary dimensions Oceanic Boundlessness (OBN), Dread of Ego Dissolution (DED), Visionary Restructuralization (VRS), and the global Altered States of Consciousness (G-ASC), plus eleven validated lower-order subscales. Because several predictors had substantial missingness due to differing designs across studies, the investigators used multiple imputation by chained equations (MICE) to generate 20 multiply imputed data sets, with variable-specific imputation methods (predictive mean matching for continuous, proportional odds for ordered categorical, logistic for binary). Continuous predictors were rescaled by dividing by two times their standard deviation so coefficients for continuous and binary predictors would be comparable. Response variables were transformed (e.g., square root, log, or negative inverse square root) to meet linear mixed model assumptions. To account for non-independent observations (multiple sessions per subject), linear mixed effects models with random intercepts for subjects were fitted; model comparisons indicated this was sufficient. Variable selection combined backward elimination with a two-step bootstrap procedure: 200 bootstrap samples per imputed data set in step one to identify predictors selected in at least 50% of bootstrap samples, and a second bootstrap-based selection starting from those predictors to find the most stable models. Interaction terms between dose and other predictors were explored. Final parameter estimates and uncertainty were obtained using Markov Chain Monte Carlo (MCMC) sampling pooled across imputed data sets. Model fit was summarised using an R2 statistic appropriate for mixed models.

The pooled sample comprised 409 psilocybin administrations in 261 subjects, with psilocybin doses ranging from 115 to 315 mg/kg (mean 214 + 63 mg/kg as reported). Drug dose was the single strongest and most consistent predictor: it reached 100% selection frequency across all 15 response variables and had the largest effect size in 12 of the 15 final models. Explained variance (R2) in simplified final models ranged across outcomes; the OAV total (G-ASC) had the highest explained variance (R2 = 0.31) and Disembodiment the lowest (R2 = 0.163). The number of non-dose predictors retained in final models varied by outcome. Among non-pharmacological predictors, Tellegen Absorption emerged as the second most important predictor overall. High Absorption strongly predicted larger overall consciousness alteration and specifically mystical-type (OBN) experiences and visual effects, with the strongest association observed for Audio-Visual Synesthesiae. Current mood state immediately before dosing also influenced outcomes: Performance-Related Activity (items such as go-getting, active, energetic) was positively associated with G-ASC, OBN, and VRS, while Emotional Excitability before intake was positively associated with Spiritual Experience and Anxiety and moderately with other main scales and some subscales. Psychological distress in the previous four weeks (SCL-90-R GSI) was negatively associated with OBN, Blissful State and Complex Imagery, indicating fewer of those positive or imagery-related effects in subjects reporting more recent problems. Setting and demographic effects: sessions involving PET measurements were strongly associated with higher Anxiety; PET was the single strongest predictor of Anxiety and had an effect size more than twice that of drug dose for that outcome. Age showed limited associations: older participants reported less Impaired Control and Cognition and trended toward more Blissful State. Gender, years of education, and BMI were not significantly associated with any response variable. Prior drug use had modest effects: hallucinogen‑naïve subjects tended to report stronger Disembodiment, VRS, and Changed Meaning of Percepts; subjects who sometimes used cannabis (≥ once per month) reported more Blissful State and a trend toward less Anxiety; alcohol consumption was positively associated with Audio-Visual Synesthesiae and Complex Imagery. The investigators tested dose-by-predictor interactions; one interaction (PET × dose predicting Anxiety) was statistically significant at p < 0.01 but was judged questionable because dose variability within PET sessions was small, and it was not retained in final models. Statistical notes: several predictors had high missing-data fractions and were imputed; Absorption had the largest relative increase in variance due to missingness (average RIV ≈ 0.94), reflecting greater uncertainty in estimates for that predictor. Variable selection produced more stable models for main OAV scales than for some subscales; for instance, the most frequently selected model for OBN was chosen in 32.1% of bootstrap samples, whereas the Anxiety model was selected in only 0.3% of cases, indicating model instability for some outcomes. Overall, substantial residual variance remained for several outcomes (e.g., more than 80% of variance in Anxiety was unexplained), indicating imperfect predictability even under controlled conditions.

Studerus and colleagues interpret their findings as confirming that drug dose is the dominant determinant of acute psilocybin effects in healthy volunteers, but that non-pharmacological variables—personality, pre-session mood, recent psychological state, prior drug experience, demographics and environment—also make meaningful contributions. Absorption was highlighted as the most important personality trait predictor, consistently associated with stronger overall consciousness alteration, mystical-type experiences and visual synaesthesia-like phenomena; this aligns with prior reports linking Absorption to responsivity in other altered-state induction procedures and suggests a possible biological link via 5-HT2A receptor variability. Current mood and short-term psychological state were generally more predictive than stable personality traits: Emotional Excitability and Performance-Related Activity prior to dosing were associated with both pleasant/mystical and anxious outcomes depending on the scale. The strong association of a PET experimental environment with increased Anxiety is discussed in practical terms: the clinical, technically equipped PET setting may provoke more anxious reactions than more aesthetically comfortable laboratory rooms; however, the authors emphasise that anxious reactions in PET experiments remained relatively infrequent and manageable with interpersonal support. Contrary to expectations and some earlier, smaller studies, Neuroticism-Anxiety (as measured here) did not predict negative reactions—an observation the authors attribute in part to screening exclusions of very high scorers and reduced neuroticism variance in their sample, while acknowledging that high levels of Neuroticism might still pose elevated risk in the population tail. Demographic effects were limited: gender did not moderate effects, and age showed modest associations with reduced cognitive impairment and somewhat more blissful states in older subjects. The authors are careful to note important limitations. Except for dose, predictor–outcome associations are observational because predictors were not experimentally manipulated across studies; substantial residual variance remained for several outcomes, particularly Anxiety. Sample composition (young, highly educated, screened, with greater prior hallucinogen and cannabis exposure and lower Neuroticism than population norms) and the controlled research environment constrain generalisability and may have reduced the incidence of severe adverse reactions, limiting power to detect risk factors for such events. Key predictors known from the literature, notably expectancy measures, were not available in the pooled data. Methodological choices also limit inference: some predictors had high missingness, albeit addressed with multiple imputation under a plausible missing-at-random assumption; some final models showed instability across bootstrap samples, introducing uncertainty. The authors suggest that future studies might employ designs such as the balanced placebo design to disentangle pharmacological effects from expectancy, use behavioural or externally rated outcome measures to validate self-report anchors, and consider categorical clustering of response patterns as an alternative analytic strategy for detecting determinants of qualitatively distinct experiences.

Although psilocybin dose was the strongest determinant of acute subjective effects, the study concludes that a substantial portion of inter‑ and intraindividual variation in psilocybin responses is attributable to non-pharmacological factors. Important predictors were found across multiple domains—personality (notably Absorption), momentary mood and activation, recent psychological distress, prior hallucinogen and substance use, age, and the session environment (e.g., PET vs non-PET). These findings underscore the relevance of set and setting in controlled research with classical hallucinogens and have practical implications for participant screening, experimental design, and safety procedures, while recognising the observational nature of most predictor associations and limits to generalisability.