The life of an adverse event

How a participant's words become a code, a table in a package insert and, years later, a label change. Why session-based treatments strain that system, and how we might finally describe a psychedelic experience well.

By Reid Robison, MD. About 30 minutes, with nine things to try.

Every prescription drug in the US comes with a long folded sheet of tiny print. Somewhere in it is a table: adverse reactions occurring in at least 2% of patients and more often than placebo. Each row in that table started as a sentence someone said to a nurse or a doctor in a clinic, often years earlier. "I feel kind of floaty." "My head hurts." "The wall was breathing."

This post follows that sentence all the way: how it gets written down, what it gets called, how it's counted, how it lands in the label, and how the label keeps changing after a drug is on the market. Then it looks at what happens when the treatment isn't a daily pill but a session, a few hours in a chair under observation, and at the hardest naming problem in the field: how to describe a psychedelic experience.

Part 1Why we write it all down

Modern drug safety rules were written in the shadow of thalidomide. Sold from 1957 as a sedative and for morning sickness, it caused severe limb malformations in an estimated ten thousand or more babies worldwide before it was withdrawn.³⁶ In the US, Frances Kelsey at FDA held up its approval, and the 1962 Kefauver–Harris amendments that followed required proof of effectiveness and much more systematic safety evidence before approval. The lesson was that nobody can predict in advance which unwanted effect will turn out to matter. So the rule became: record everything, and decide later what the drug caused.

That's why the regulatory definition is deliberately broad. An adverse event is any untoward medical occurrence in someone given a drug, whether or not it's considered related.³ A participant who breaks an ankle on the stairs a week after dosing has an adverse event. An adverse reaction is narrower: an unwanted response where a causal link to the drug is at least reasonably possible.^4,5 Trial tables usually report "treatment-emergent" adverse events, those that started or got worse after the first dose.

Each event then gets several separate judgments, and mixing them up is the most common mistake people make reading safety data:

Part 2How an AE gets its name

Investigators write down what happened in their own words, ideally close to the participant's. That free text is the verbatim. But you can't count free text across hundreds of sites in a dozen languages, so every verbatim is translated into a controlled vocabulary: MedDRA, the Medical Dictionary for Regulatory Activities. MedDRA was introduced in 1999, is owned by the International Council for Harmonisation (ICH), and is updated every March and September.²

A few features of MedDRA matter for everything that follows:

• The preferred term (PT) is the unit of analysis. Safety tables count PTs. Everything above PT is for grouping; everything below is for matching the coder's search.
• Near-synonyms can land in different body systems. "Dizziness" sits under nervous system disorders and "vertigo" under ear and labyrinth disorders. "Blood pressure increased" is an investigation result; "hypertension" is a vascular disorder. Split the same phenomenon across codes and it can vanish below a reporting threshold.
• Coders disagree. In a systematic review, two trained coders chose different preferred terms for the same event about 12% of the time.⁷
• Grouping tools help. Standardised MedDRA Queries (SMQs) bundle related terms to search for a broader safety concern, such as "hostility/aggression" or "suicide/self-injury."

Almost every psychedelic effect lands in a single class: psychiatric disorders. Hold onto that; it's the heart of Part 6.

Part 3The pathway to the label

Here is the route a single event travels, from the treatment room to the package insert. Each station has a deadline, a responsible party and a document.

How a label table is built

Under FDA's labeling format (the "Physician Labeling Rule"), the adverse reactions section has a clinical-trials table, and the front-page Highlights list the most common reactions.¹¹ Each label states its own cutoffs. Spravato's, for example, tabulates reactions in at least 2% of patients and more often than placebo, and its Highlights list those at 5% or more and at least twice placebo.¹ Cutoffs are a design choice, and they shape what readers see.

Part 4After approval: labels that change

A typical development program studies a few thousand people for months. A marketed drug may reach millions of people for years, including older, sicker and more medicated people than any trial enrolled. Rare harms and harms with long delays only show up after approval. So the label is never finished.

Companies must send FDA serious, unexpected reports within 15 days and file periodic safety reports, quarterly for the first three years after approval and annually after that.¹² Clinicians and patients can report directly. Those spontaneous reports used to flow into FAERS, the FDA Adverse Event Reporting System. In March 2026 FDA folded FAERS and several other databases into one platform, the Adverse Event Monitoring System (AEMS), which handled roughly six million reports a year across product types at launch.¹³ Spontaneous reports can't give incidence, because nobody knows how many people took the drug or how many events went unreported, but they're excellent at raising a signal that someone then investigates with other data, such as FDA's Sentinel system of health-insurance and medical records.

Since 2007, FDA can require safety labeling changes when new information emerges, and can require a Risk Evaluation and Mitigation Strategy (REMS) such as Spravato's in-clinic monitoring program.¹⁴ Psychiatry has some of the most instructive label histories:

Part 5The session era

The safety system grew up around daily pills. You take one every morning, the drug sits in your blood around the clock, and an unwanted effect like dizziness or nausea may be with you for weeks. "What percentage of people had nausea?" is a reasonable summary of that experience.

Interventional psychiatry works differently. Esketamine (Spravato) is given in a clinic, with at least two hours of monitoring under a REMS.¹ Psilocybin and LSD in trials are usually given once or a few times, with a full day of observation. Most of the drug's effects, wanted and unwanted, happen inside that window, then end.

Now look at what a standard AE table does with that. It counts the percentage of people who had an event at least once. A participant who felt detached for 40 minutes on two dosing days and a participant who felt detached every day for a month both count once. The table is accurate and nearly blind to time.

This isn't a new idea. Clinical pharmacologists Jeffrey Aronson and Robin Ferner proposed describing adverse reactions by Dose, Time course and Susceptibility (the "DoTS" framework) more than twenty years ago.²¹ Session-based treatment makes the time axis impossible to ignore. Useful additions to every table would be: time to onset, duration, the share of events that resolved on the dosing day, and whether anything persisted past a defined window such as 24 hours or 7 days.

When the effect is the treatment

There's a second problem. With a daily antidepressant, nausea is clearly a side effect. With a psychedelic, the perceptual and emotional changes are inseparable from how the drug is thought to work. Are they adverse events? Two recent regulatory moments answered that firmly.

In 2024, reviewing the MDMA-assisted therapy application for PTSD, FDA's briefing document noted that the studies "did not capture effects deemed positive, favorable, or neutral, such as 'euphoria' or 'elated mood,'" and that the lack of abuse-related terms "limits the assessment of abuse potential."¹⁹ The application was not approved. Then, in July 2026, FDA's final guidance on psychedelic trials said that expected effects such as euphoria, hallucinations, perceptual distortions and cognitive changes should still be documented as adverse events because of their relevance to abuse potential, even when participants don't describe them negatively.²⁰

That settles whether to record these effects. It leaves wide open how. If "hallucination" and "euphoric mood" are going to be counted in every psychedelic trial, the words had better mean something.

Part 6Describing a psychedelic experience

This is the hard part, and the most interesting. A psychedelic experience is hard to describe for reasons that are built into it: it's often called ineffable; memory for it is imperfect and state-dependent; the person rating it is the person whose perception is altered; and some effects don't scale smoothly with dose. (I go deeper into why the shapes repeat in Seeing Things.)

Researchers have built good questionnaires for parts of it: the 11-dimension Altered States of Consciousness scale,²⁴ the Mystical Experience Questionnaire,²⁵ the Challenging Experience Questionnaire,²⁶ the Ego-Dissolution Inventory²⁷ and the Emotional Breakthrough Inventory.²⁸ Reading across them, a psychedelic experience has roughly seven components.²⁹

The kinds of perceptual change

Perception is where safety coding is crudest, so it's worth slowing down. Visual changes on classic psychedelics run along a fairly consistent ladder, from subtle to profound.³⁰ Three distinctions do most of the work:

• Is there a real stimulus? If the wall is really there and it seems to breathe, that's an illusion or distortion. If something is seen with nothing there, that's a hallucination.
• Does the person know it isn't real? When insight is intact, many researchers prefer pseudo-hallucination. The MM120 trial grouped "illusions, pseudo-hallucinations, visual hallucinations" together as visual perceptual changes.¹⁸
• Eyes open or closed, and on top of the world or replacing it? Geometry overlaid on a wall, closed-eye patterns, and immersive scenes that replace the room are different phenomena.

Alongside vision there's synaesthesia (music seen as colour or felt as texture), changes in hearing (richer, distorted or echoing sound), and changes in time (minutes that feel like hours, or time seeming to stop).

The perception lab

The levels in that demo borrow from Alexander Shulgin's simple rating scale: + means an effect is perceptible but its nature unclear; ++ means the effect and its nature are clear but you can still function; +++ means it can't be ignored; ++++ is a rare, transcendent peak state.³³ It's crude, but it's an inventory, not just a volume knob.

Inventory versus dictionary

Here's the core problem in one picture. Below is a checklist of phenomena a participant might report after a session. Tick what applies. On the left you get an inventory, a fingerprint of what happened. On the right is what a typical AE table will show after coding.

Two very different sessions can produce identical rows. The inventory keeps the difference. None of this means abandoning MedDRA; regulators need a shared dictionary. It means recording the structure alongside the code.

Part 7What better looks like

A practical package for session-based trials, most of which needs no new validated instrument:

1. Keep the verbatim, permanently linked to the code.Coding is lossy. The participant's words are the only place lost information can be recovered.
2. Time-stamp onset and resolution.Report the share of events that resolved on the dosing day, and anything still present at 24 hours, 7 days and 30 days. That's the DoTS time axis, made routine.²¹
3. Record expected effects as AEs, as FDA now asks,but flag them as expected pharmacodynamic effects so analyses can separate them from unexpected harms.²⁰
4. Add a phenomenology inventory.Sense, form, eyes open or closed, insight, presence of a being, sense of self: present or absent. Infer intensity from the pattern, not a three-box rating.
5. Separate distress from harm.Record how the person felt at onset and at the integration visit as separate fields. A frightening hour that later feels meaningful and a frightening hour that leaves damage shouldn't share one row.³⁴
6. Ask about persistence on a schedule.Lasting perceptual changes are mostly found when someone asks. Ask at day 30 and day 90.
7. Run the established scales alongside.11D-ASC, MEQ30, CEQ and EDI give the dimensions; the inventory gives the names.^22,23

A safety table should tell you not just how many people had an effect, but what it was, how long it lasted, and whether it was still there a month later.

Part 8How to read the AE section of a label

• Start with the Highlights on the first page for the most common reactions, and check the cutoff the label used.
• In the clinical-trials table, subtract placebo. If 10% had headache on the drug and 8% on placebo, the drug's share is closer to 2 in 100.
• Rare but important harms live in Warnings and Precautions or a boxed warning, not in the common-reactions table.
• The postmarketing section can't give frequencies, because spontaneous reports come from an unknown number of users.
• For session-based drugs, look for timing language such as "transient" or "on the day of dosing." It tells you more about your day than the percentage does.
• You can report a side effect yourself through FDA's MedWatch program. Patient reports count.

This post is educational and isn't medical, legal or regulatory advice. Regulations and labels change; check current FDA sources and the current label for any specific drug.