Clinical note

When de-escalation meets psychosis

Every behavioural de-escalation framework rests on one assumption. Psychosis is the situation where that assumption stops holding — and the response has to change with it.

Cole G. Dyar, DMFT, LAADC, CEAP
Certified Nonviolent Crisis Intervention Instructor
Author of the Psychosis Safety Protocol

A behavioural health technician calls me about a shift that went wrong. She had done everything her training told her to do. She kept her distance, kept her hands visible, kept her voice low. She made eye contact so he would know she was listening. She explained, calmly and repeatedly, that there was no one in the parking lot. And the situation got worse every time she spoke.

She did not make a mistake in the sense her incident report implies. She executed a behavioural de-escalation framework correctly. The framework was built for a different problem than the one in front of her.

The assumption underneath the frameworks

Almost every de-escalation model in circulation — and I teach one of them; I hold a current Nonviolent Crisis Intervention instructor certification — is built on a model of escalation that runs roughly like this. Something happens. The person's arousal rises. Their capacity to reason drops as arousal climbs. If you can lower the arousal, reasoning comes back, and once reasoning is back you can resolve the thing that started it.

That model is sound, and for most of what happens on most units it is the right one. The person is angry, frightened, humiliated, in pain, or withdrawing. Bring the arousal down and you get a person you can talk to.

Psychosis is the case where lowering arousal does not restore the shared premise. The voice does not stop when the room gets quiet. The belief that the staff are part of it does not resolve because your tone improved. You can do everything the framework asks and still be talking to someone whose reality is not going to converge with yours in the next ninety seconds. What you need in that window is not a better version of the same move. It is a different move.

Three places the divergence is concrete

Reasoning against the belief. The instinct — and it is a kind instinct — is to correct. There is no one in the parking lot. Nobody is coming. A qualitative evidence synthesis of how people actually experience delusions¹ describes them as carrying meaning: they are not free-floating errors but explanations that are doing work for the person holding them, often work related to threat and to identity. Argue with the explanation and you are not offering relief, you are removing the only account they have of what is happening to them, which reliably raises threat rather than lowering it.

What replaces it is narrow and learnable, and it is the single hardest line in the whole protocol: validate the feeling, never the content. "That sounds terrifying" is true, sayable, and does not confirm anything. "I don't see anyone, and I can see you're frightened, and I'm staying" is true twice. Most staff have never been taught where that line sits, and they discover it under pressure or not at all.

Command hallucinations as a category, not a note. In a lot of documentation "reports hearing voices" lands in the same field as any other symptom. A 2026 systematic review of compliance, appeasement and resistance in command hallucinations² treats the person's relationship to the voice — whether they comply, negotiate with, or resist it — as the clinically meaningful variable. That is a different assessment question from whether voices are present, and it changes what you do next. A person actively resisting a command is doing work you can support. A person who has begun appeasing one is in a different situation entirely.

What the standard posture does in a persecutory presentation. Sustained eye contact reads as attention in most of medicine. To someone who believes they are being watched and assessed, sustained eye contact is confirmation. Squared shoulders and a direct face-on stance read as openness in a classroom and as confrontation in a corridor. The components of the standard posture are not wrong; they need to be re-derived for a person whose threat-detection is running hot. Angle the stance. Let eye contact be intermittent. Keep hands visible for a reason you could articulate.

What the evidence does and does not say

I would rather state this plainly than let a prospect discover it later. The evidence that de-escalation training reduces violent incidents is genuinely mixed. A systematic review of training in forensic psychiatric settings⁴ found the literature unable to support a confident conclusion. The EDITION programme³ reached the point of a feasibility trial, which is where a field is when it is still establishing whether an intervention can be delivered and measured well, not whether it works at scale. A cluster randomised study in acute psychiatric units⁵ is among the more encouraging pieces, and a systematic review of interventions against workplace violence toward nurses⁷ reports the same unevenness across settings.

Two things follow, and I hold both. The first is that anyone selling you a training product on a promise of measurable incident reduction is ahead of the evidence, and that includes anyone selling psychosis training. PSP has no published outcome data and I will not imply otherwise. The second is that the more recent qualitative work⁶ points at why the picture is uneven: de-escalation is not one behaviour, it is a set of them, applied to populations that differ, by staff whose situations differ. Aggregating all of it and asking whether "de-escalation training" works is close to asking whether "medication" works. The useful question is narrower — which technique, for which presentation, at which moment. Research on seclusion-reduction protocols⁸ tends to find the same thing: the specific protocol matters more than the general commitment.

That is the argument for a psychosis-specific layer. It is not an argument against the general frameworks; it is that "de-escalation" is too coarse a unit to be trained as one thing, and psychosis is the presentation where the coarse version most visibly stops fitting.

Two layers, and what each one is for

I am not describing a replacement, and I want to be exact about this because the distinction is the whole point.

A physical-skills and behavioural programme teaches what to do with your body — disengagement, team safety, the physical techniques a facility must have available and must have trained. PSP does not teach hands-on holds and is not built to. What PSP teaches is the clinical layer above that: which presentation you are looking at, what the risk actually is, which language line to hold, when the sequence has reached its limit and a clean handoff is the correct move rather than a failure.

Facilities that run both have a staff member who knows what to say in the first ninety seconds and what to do if those ninety seconds do not work. Facilities that have only the physical layer have people improvising the clinical part, on shift, at three in the morning. That is not a criticism of the programme they bought. It is a description of what it was scoped to do.

If you are holding a training mandate and trying to work out which parts of it you have actually covered, the crosswalks are the honest version of that answer, including the two elements PSP marks Partial because it does not deliver them.

This article describes clinical reasoning and training content. It is not clinical advice, not a legal opinion, and not a claim about any other training programme. PSP™ has no published outcome data; nothing here asserts that it reduces incidents, injuries or restraint use.