# Psilocybin direct-evidence pilot synthesis v0.1

**Framework:** The Shape of Harm v0.4  
**Estimand:** EST-AE-PSI-001  
**Search date:** 23 July 2026  
**Status:** Author-run pilot extraction and synthesis. **Not** a completed systematic review, duplicate extraction, independent risk-of-bias assessment, or clinical recommendation.

## Why this stage exists

The feasibility screen showed that controlled 25 mg psilocybin was the first pilot scenario with trial evidence close enough to the defined estimand to test the research machinery. This stage asks a narrower question: can published reports be transformed into estimand-compatible safety quantities without inventing denominators or silently mixing windows?

## Search and screening result

The author pilot located nine major candidate records. Two trials were included in the direct MDD stratum:

1. Raison et al. (JAMA, 2023; NCT03866174): 50 participants received 25 mg psilocybin and 54 received 100 mg niacin in the safety population.
2. Yngwe et al. (JAMA Network Open, 2026; NCT04630964): 17 received 25 mg psilocybin and 18 received 100 mg niacin.

Goodwin 2022 and Mertens 2026 were retained as **indirect TRD evidence**, not pooled with direct MDD trials. Weight-based, repeated-dose, fixed-order, or waitlist designs were excluded from the primary stratum.

The formal review still requires Embase, PsycINFO, CENTRAL, WHO ICTRP, forward/backward citation searching, information-specialist review, and two independent reviewers.

## Direct quantitative result that can be released

Across the two direct MDD trials, **0 of 67** psilocybin participants and **0 of 72** niacin participants had an investigator-attributed drug-related serious adverse event. Because each trial reported no drug-related SAE over follow-up longer than seven days, the count is also zero within the prespecified seven-day window.

- Psilocybin: **0 per 1,000**, 95% exact binomial CI **0 to 53.6 per 1,000**.
- Niacin: **0 per 1,000**, 95% exact binomial CI **0 to 49.9 per 1,000**.
- Risk difference: **0 per 1,000**, 95% Newcombe CI approximately **−50.7 to 54.2 per 1,000**.

This is not evidence that the true risk is zero. The active-arm interval remains compatible with roughly one drug-related SAE per 19 administrations at its upper limit. Investigator causality classification is also an imperfect outcome and may differ across studies.

## Outcomes that remain blocked

### All-cause serious adverse events in exactly 0–7 days

Not estimable from the reports in a common participant-level form. The Swedish trial reports procedure-related and unrelated SAEs at scheduled checkpoints; some were caused by lumbar puncture or arterial cannulation, which are not part of the intended clinical scenario. The exact first-seven-day participant count is not provided.

### Severe psychiatric or behavioral events requiring medical attention in exactly 0–7 days

Not poolable. Raison reports one participant with severe panic attack/paranoia through day 9. Yngwe reports two participants with persistent severe anxiety requiring medical attention, but exact onset is not reported. Combining those values would manufacture a common time window.

### Rescue medication and extended observation

Not estimable. Complete arm-level participant counts in the prespecified windows are absent from the direct reports.

## Study-level safety signals that must remain visible

- Raison: 4/50 participants had severe related adverse events through day 9; one participant had panic attack and paranoia. No suicidal or self-injurious behavior was reported, and recorded ideation was passive.
- Yngwe: one severe adverse event occurred on the dosing day; two participants later reported persistent severe anxiety requiring medical attention. Ten SAEs occurred across the year, none attributed to psilocybin; several were caused by research procedures.
- Indirect TRD studies contain more concerning signals, including suicidal ideation/self-injury and serious adverse reactions. Those findings are not erased by keeping strata separate.

## Preliminary bias judgment

Both direct trials are judged **some concerns** for this exact safety estimand in the author pilot. The central issues are functional unblinding, subjective causality attribution, small samples, and safety reporting that does not map cleanly to 24-hour and seven-day participant-level outcomes.

These judgments must be repeated independently using RoB 2 before publication.

## Instrument-design consequence

The pilot confirms that the model cannot rely on a single generic “bad reaction” score. The safety schema needs distinct fields for:

- all-cause versus investigator-attributed treatment-related events;
- participants with an event versus total event counts;
- dosing-day, 24-hour, seven-day, and longer follow-up windows;
- standard treatment procedures versus optional research procedures;
- serious regulatory events, severe nonserious events, psychiatric rescue, and prolonged symptoms.

A value is publishable only when numerator, denominator, event unit, causal attribution rule, and window all match the estimand.

## Next gate

1. Have two independent reviewers repeat screening and extraction.
2. Run the complete database and registry search.
3. Request missing exact-window safety counts from study authors/sponsors.
4. Adjudicate RoB 2 judgments.
5. Update the synthesis without changing the estimand or inclusion rules after seeing the direction of results.