Developing Methods for Observing Awe Narration in Psilocybin-Assisted Therapy

Tarbi and colleagues frame psychedelic-assisted therapy as a field that still needs clearer methods for understanding what actually happens during treatment sessions and how the therapeutic context may shape outcomes. They argue that large-scale, direct observation will be important for identifying causal pathways and note that emerging digital recording infrastructure and machine-learning methods make this increasingly feasible. Within that broader challenge, they focus on awe as one plausible therapeutic mechanism, especially because awe experiences may be linked to improved quality of life and may overlap with mystical-type experiences that are thought to contribute to the effects of psychedelic-assisted therapy. The paper’s aim is to develop and preliminarily evaluate a direct observation coding system for narrations of awe during psilocybin-assisted therapy. The researchers set out to create explicit, human-readable criteria that could identify moments when participants verbally described awe, both to improve understanding of the phenomenon itself and to provide manually labelled examples that could later support computational detection. In this way, the study is presented as a methodological foundation for future mechanistic research in psychedelic-assisted therapy rather than as an efficacy trial.

This was a descriptive study using video and audio recordings from a recent Phase II, open-label psilocybin-assisted therapy trial in people with cancer and major depression. The sample included 8 participants, and the researchers purposefully selected participant-therapist dyads to maximise gender identity combinations among 7 therapists and 8 participants. All participants received psilocybin. Recordings covered preparation, dosing, and integration encounters across the three-week course of therapy. The researchers developed a conceptually grounded structured codebook based on a foundational definition of awe as involving perceived vastness and a need for cognitive accommodation. They initially included three candidate features of awe narration: vastness, need for accommodation, and ineffability. Vastness referred to expressions of something being larger than the self in physical, social, or conceptual terms. Need for accommodation referred to speech suggesting a disruption of ordinary understanding that would require mental adjustment. Ineffability referred to difficulty putting the experience into words, such as long pauses, restarts, or explicit statements of expressive difficulty. Two human coders independently reviewed all encounter videos and identified the start and stop times of moments that met the coding definition. After the first four participants, they paused to review coder feedback and refine the definitions. Coders rated their confidence in each identified moment as low, moderate, or high. Moments were counted as co-identified when the observed time periods overlapped, including boundary cases where one coder’s start matched the other’s end. The researchers also marked adjacent moments that felt like awe but did not meet the criteria as boundary moments, to improve transparency and help refine the codebook. They calculated the frequency and distribution of awe narration across preparation, dosing, and integration sessions, estimated inter-rater agreement with Cohen’s kappa, and used unadjusted odds ratios with 95% confidence intervals to examine whether coder confidence was associated with vastness. Mixed-effects logistic regression was used to assess whether clustering by participant experience affected the findings. The study had ethics approval from the relevant institutional review boards, and participants gave separate consent for analysis of the recorded trial sessions.

The participant sample had a median age of 59 years, with a range from 30 to 78 years. Half were female. All participants identified as white, and one also identified as Asian American/Pacific Islander. Cancer diagnoses included breast, kidney, lymphoma, and colon cancer. One quarter of the sample had an expected survival of less than two years. Across 16,760 total minutes of video, the coders identified 246 moments of awe narration. Inter-rater reliability was substantial, with Cohen’s kappa of 0.73 (95% CI: 0.68, 0.78). Of the 246 moments, 42% (104/246) were identified by only one coder and 58% (142/246) were identified by both. Among the moments coded with high confidence, 30% (31/104) were identified by one coder and 70% (73/104) by both. Very few awe narration events occurred during preparation encounters in the first half of coding, so after the first four participants the researchers focused more heavily on dosing and integration encounters. Vastness emerged as the most salient and useful feature for identifying awe narration. Coders were substantially more confident in identifying a moment as awe when vastness was present than when it was absent (OR: 4.3; 95% CI: 2.4, 7.8). This association remained similar after adjustment for clustering by participant experience (adjusted OR: 4.1; 95% CI: 2.1, 8.0) and did not differ meaningfully across dosing versus preparation-integration encounters. Ineffability was common: more than half of all awe narrations (130/237; 55%) included it, and most of these occurred during the dosing session or the initial integration encounter the following morning (75%; 97/130). The text does not present a separate numeric summary for need for accommodation, but it reports that coders refined this feature into two parts: an initial cognitive disruption and a later, variable possibility of accommodation. The researchers also describe boundary moments that showed related qualities such as beauty or metaphor but were not coded as awe because they lacked vastness.

The authors conclude that narration of awe during psychedelic-assisted therapy is directly observable when it is defined using explicit criteria. They argue that the findings support two practical recommendations for future coding systems. First, they suggest treating vastness as the key entry point for identifying candidate awe moments, with ineffability or accommodation used afterwards to confirm the judgement. They present this as both empirically supported by the coder data and consistent with earlier conceptual work in which vastness is central to awe. Second, they recommend operationalising accommodation as a two-step process: first identifying evidence of an initial cognitive disruption, and then examining whether the participant subsequently appears to accommodate that disruption. The authors suggest that this refinement may better capture the complexity of accommodation, including cases where participants do not try to make sense of the experience. They link this interpretation to broader psychedelic-assisted therapy frameworks, including ideas from neuroscience about temporary disruption followed by new neural pathways, and to the integration model in which post-dosing sessions help participants reframe and integrate insights. The authors also emphasise that the primary contribution of the study is methodological: a direct observation codebook that could support future human coding and, eventually, automated detection in large psychedelic-assisted therapy datasets. They note that machine-learning systems require accurate human-labelled training data and that human oversight remains important for quality control and ethical reasons. The main limitations they acknowledge are the small sample size, the lack of sufficient ethnic, racial, and linguistic diversity, and the fact that the study assessed narrated awe rather than the internal subjective state of awe itself. They state that awe narration often occurred at the same time as, or shortly after, the experience, making it a plausible observable proxy, but they call for further work in larger and more diverse samples to test validity and examine how narrated awe relates to internal experience and therapeutic outcomes.