There's a thing that happens in every conversation about mental illness, where somebody reaches for a word to sort the serious from the not-so-serious, and the word they reach for is psychosis.
It's the closest thing we have to a bright line. Mood disorder on one side, psychosis on the other, and everyone nods like that settled something.
Daniel's been poking at that line. He wrote in with four questions stacked on top of each other. First, he says psychosis is one of several heuristics we use for severity, and he lines up the two sides. Mood disorders, ADHD, conditions that can be debilitating and dangerous, on one side. Psychotic illness on the other.
And he notices the obvious thing about that second side.
He does. Schizophrenia is the famous resident, but the existence of a category called psychotic illness rather implies schizophrenia isn't the only tenant. So question two is: what else lives there? Question three, he says we flatten psychosis into hallucinations specifically, and he wants the broader description. How does this actually manifest, beyond the voice in the hallway. And question four, the severity stratification itself. Is it outmoded, or is it still doing work? His guess is routing, specialists versus generalists.
Four questions, and honestly they're one question wearing four coats.
Which is why we're going to take them in order. Start with the definitional one. What are we actually talking about when we say psychosis?
Here's the paradox that makes this episode worth doing. The formal definition is narrow. Narrow. Both the American Psychiatric Association and the World Health Organization define psychosis by requiring hallucinations without insight into their pathologic nature, or delusions, or both. That's it. That's the whole formal definition.
And the colloquial version?
Broader. Much broader. People say psychosis and they mean crazy, unmoored, out of touch. The formal term is tighter than the slang, which is the opposite of what most people assume.
So the word Daniel's using as a dividing line is doing something stranger than it looks. It's precise at the symptom level and vague at the policy level.
Exactly the tension. Because on top of that narrow symptom definition we've built a severity label, serious mental illness, and that label gets applied broadly and inconsistently. There's a systematic review from a few years back, Gonzales and colleagues, seven hundred and eighty-eight empirical studies on serious mental illness. Eighty-five percent of them never defined the term. Not once. They used it as if everyone agreed what it meant.
And the fifteen percent who did define it?
Varied wildly. Some used any diagnosis. Some used specific diagnoses. Some used functional impairment, some used duration. And when you apply different definitions to the same community sample, the share that qualifies runs anywhere from four percent to eighty-eight percent. Same people, different rubric.
That's not a margin of error. That's a different population.
That's the whole episode in one number. And I want to flag the sharpest version of it early, because we'll come back to it. Eating disorders have the highest mortality rate of any diagnostic category in psychiatry. Highest. And they appeared in one percent of those serious mental illness studies.
One percent.
Suicidality appeared in exactly one study's criteria. Out of seven hundred and eighty-eight.
So the label that's supposed to sort by severity is missing the category that kills the most people.
Which tells you the sorting isn't really tracking severity. It's tracking something else, and we'll get to what. But first, the symptom question. Because Daniel's right that we flatten this, and the flattening is worth undoing properly.
The five domains.
The DSM-5-TR defines the schizophrenia spectrum and other psychotic disorders by five domains of psychopathology. One, delusions. Two, hallucinations. Three, disorganized thinking, which is formal thought disorder. Four, grossly disorganized or abnormal motor behavior, including catatonia. Five, negative symptoms.
And the first four are psychosis, the fifth is adjacent?
Negative symptoms are a related but distinct domain. Restricted affect, amotivation, poverty of speech. They're part of the disorder but they're not what we mean by psychosis proper. So four of the five domains are psychotic manifestations, and the colloquial version collapses all of them into hallucinations.
Give me the hallucination side properly, then. Because even that's more varied than people think.
Hallucinations occur across all sensory domains. Auditory is the famous one but you get visual, tactile, olfactory, gustatory. And within auditory there's a specific historical set, the Schneiderian first-rank symptoms. Voices conversing with each other. A running commentary on the patient's behavior. Thought echo, hearing your own thoughts spoken aloud.
Those were supposed to be the signature of schizophrenia.
They were treated as pathognomonic for decades. And they are not specific. They show up in mania. They show up in psychotic depression. Temporal lobe epilepsy. Dissociative identity disorder. The first-rank symptoms don't sort schizophrenia from anything. What they sort is psychosis from not-psychosis, and even that's leaky.
So the symptom that everyone agrees is psychosis turns out to be a family, not a signature.
And the other domains are stranger still. Formal thought disorder. Illogicality, tangentiality, derailment, neologisms, thought blocking. That can actually supplant hallucinations and delusions in diagnosing a psychotic disorder, provided it's severe enough to substantially impair effective communication and it comes with disorganized behavior, catatonia, or negative symptoms.
So a person can be diagnosably psychotic with no hallucination and no delusion at all.
Purely on the structure of their thinking and the way it comes out. Which almost nobody outside the field knows.
What about the delusional side?
Delusional misidentification syndromes are the ones I find most striking. Capgras, where a familiar person is believed replaced by an identical impostor. Frégoli, where a persecutor is believed to be disguising themselves as different people. Intermetamorphosis. Subjective doubles. And here's the figure that stops me cold. Twenty to forty percent of these occur in neurologic conditions, not primary psychiatric illness. Right hemisphere lesions, most commonly.
So a third of the time, roughly, this psychotic-looking presentation is a brain injury or a lesion wearing a psychiatric mask.
Which is a big deal for the routing question. Because if you route that person to a psychiatric service and nobody images the brain, you've missed the actual cause.
That's a case where the severity label gets you to the wrong specialist.
Gets you to a specialist, just not the right one. Which is a preview of the whole policy problem. But before we get there, the structural point. Because the way we model psychotic symptoms has changed, and it changes how you think about the categories.
Factors, not categories.
That's the pivot. In 1987 Liddle proposed a three-factor structure. Positive symptoms, hallucinations and delusions. Negative symptoms. Disorganization. And then the positive factor got split further, into what's called reality distortion and disorganization. Andreasen validated it in the nineties.
And that's held up?
It's held up reasonably well across twin samples and general clinical samples. Then the EU-GEI first-episode study went further, and a bifactor model fit best. One general factor plus five specific dimensions. Positive, negative, disorganization, manic, depressive.
One underlying severity dimension plus five flavors.
And then the conversion study, Grot and colleagues, took the PANSS and the SAPS and SANS, tried to convert scores between them, and found the positive symptoms don't factor as one thing either. They split into three distinct subdomains. Hallucinations, delusions, and disorganization. Separate enough that they don't convert cleanly.
So even positive symptoms, the thing everyone thinks is one bucket, is actually three.
Which lands the sub-thesis for this half of the conversation. Psychosis isn't one thing. It's a dimensional structure of factors. And once you've accepted that, the categorical boxes start to look like administrative conveniences rather than natural kinds.
And that sets up Daniel's second question. What else features it?
The list is longer than most people expect, and I want to go through it properly because the surprise is in the breadth.
Go.
Mood disorders first. Psychosis occurs in bipolar disorder during manic or depressive episodes, and in major depressive disorder during major depressive episodes. And the psychotic symptoms can be mood-congruent, so a depressive delusion of guilt fits the mood, or mood-incongruent, so a delusion that has nothing to do with the mood state. That distinction used to matter enormously for diagnosis. Less so now, but it's still recorded.
Psychotic depression is common enough to have its own treatment literature.
It does, and there's an interesting gap there. A network meta-analysis in Lancet Psychiatry in 2024 looked at pharmacological treatments for psychotic depression and noted up front that there are no recommendations based on the efficacy of specific drugs for it. So we have a common condition and a thin evidence base for what to give.
That's a recurring theme in this territory, isn't it.
It is. Schizoaffective disorder next, and this one's a mess. The definition requires psychosis for at least two weeks in the absence of mood symptoms over the lifetime, plus mood symptoms for the majority of the total illness duration. But the construct itself remains unresolved. There are competing views that it's a variant of schizophrenia, a variant of mood disorder, or just two illnesses co-occurring.
So a diagnosis that's definitionally contested.
Openly. The literature says unresolved, in those words. Delusional disorder, delusions for at least a month without prominent hallucinations. And DSM-5 broadened it to include bizarre delusions, which used to be sufficient for schizophrenia on their own.
That's a real boundary shift.
It is. Brief psychotic disorder, at least a day but less than a month, with full return to premorbid function. And it has a postpartum onset specifier. Schizophreniform disorder, one to six months. Then the substance-induced category, psychosis during intoxication or withdrawal, sometimes becoming chronic.
And the medical mimics.
Psychotic disorder due to another medical condition. DSM-5 lists neurocognitive disorders due to Alzheimer's, Parkinson's, Lewy body dementia, frontotemporal lobar degeneration, Huntington's, prion disease, cerebrovascular disease, traumatic brain injury, and HIV.
Every one of those is a condition where the psychiatric presentation is downstream of something else entirely.
Which is why the differential matters. There's no single diagnostic test that separates primary from secondary psychosis on psychopathology alone. You rely on atypical features, age of onset, functional change. A sixty-five-year-old with a first psychotic episode is a different workup from a twenty-two-year-old.
Late-onset psychosis.
First episode after forty. That includes late-onset schizophrenia, delusional disorder, psychotic depression, plus the secondary causes. Neurodegenerative, metabolic, infectious, inflammatory, nutritional, endocrine, medication toxicity. The medication one is worth sitting with. You can induce psychosis in someone with a drug interaction and the whole presentation resolves when you fix the prescription.
Postpartum psychosis next?
Incidence of 0.25 to 0.6 per thousand births. Twenty to fifty percent of those women have isolated postpartum psychosis, meaning it's not part of a bipolar spectrum picture. The rest usually do fall within the bipolar spectrum. And the treatable causes matter here. Autoimmune thyroiditis, infections, and NMDA-receptor encephalitis.
Encephalitis presenting as postpartum psychosis.
It's a known mimic and a treatable one. Which is the case where the psychosis label gets you to the right specialist fast, and the specialist finds the cause. Bergink and colleagues wrote the review on this and the framing is exactly that. Madness, mania, and melancholia in motherhood, with a differential that includes things you can actually cure.
Now the personality and dissociative end, because that's where Daniel's dividing line gets tested hardest.
It does. Borderline personality disorder includes transient stress-related paranoid ideation or severe dissociative symptoms as part of the diagnostic criteria. Not a separate condition layered on top. Part of what BPD is. Lifetime prevalence somewhere between 0.7 and 2.7 percent in the general population, around twelve percent outpatient, around twenty-two percent inpatient.
So it's common in the settings where the routing decision gets made.
And it's excluded from the UK serious mental illness register, with the exception of schizotypal. So a person with severe BPD and transient paranoid ideation is administratively not serious, while a person with a single psychotic episode that resolved is.
That's the one-off-episode problem.
Dissociative identity disorder and complex PTSD also feature auditory hallucinations. Ross put it plainly: voices are common in those conditions and are not specific to psychosis. And the voices differ in character. Psychotic voices carry more formal thought disorder, more negative symptoms, more delusions alongside. Dissociative voices tend not to.
So the voice is not the discriminator.
The voice plus everything around it is. And then OCD. When insight into obsessions is lost, the obsessions get described as delusions. The content is the same, the relationship to it changed.
And ADHD, which Daniel put on the non-psychotic side?
Correctly, mostly. ADHD is notably not typically psychotic. But it appears in serious mental illness research samples. Two percent of the SMI studies, nine percent of the serious emotional disturbance studies. Which means even the non-psychotic conditions are drifting into the psychotic bucket at the sample level.
The buckets are leaking in both directions.
Now here's the data that makes the categorical claim hardest to defend. Diagnostic stability. Fusar-Poli and colleagues, fourteen thousand four hundred and eighty-four patients, four and a half year follow-up. How often does the diagnosis you got at first episode still apply later?
Give me the numbers.
Schizophrenia, 0.90. Affective spectrum psychoses, 0.84. Schizoaffective, 0.72. Substance-induced, 0.66. Delusional disorder, 0.59. Brief psychotic disorder, 0.56. Psychosis not otherwise specified, 0.36. Schizophreniform, 0.29.
Schizophreniform at 0.29. That means fewer than a third of people diagnosed with it still have that diagnosis four years later.
And that's not clinician error. It's that the construct itself is a provisional box, one to six months, waiting to see where the person lands. But even brief psychotic disorder at 0.56 means nearly half move. The categories aren't equally stable, and the unstable ones are exactly the ones the routing decision depends on.
So now we can take Daniel's fourth question properly. Is the severity stratification outmoded?
It's not outmoded in practice, and it's not defensible as science, and both of those are true at once. Start with the science. Serious mental illness is not an official DSM or ICD diagnosis. It's a research and policy construct. And the Gonzales review's conclusion is blunt. The term has poor reliability and validity across the empirical literature, and its use should be consistently defined or suspended in the absence of clear operational definitions.
That's the review saying either define it or stop using it.
And they name the harm too. If we deem some mental illnesses serious, what does that say about the others? The labeling itself may further stratify mental illnesses. Which is a strange thing to do in a field that's supposed to be destigmatizing.
But the defense.
The strongest defense comes from Wing, quoted in that same review. The concept is fuzzy, he grants that, but it has gained substantial official and professional acceptance because it's relevant to the reality of the case-mix decisions that have to be made. In particular, crossing the invisible boundaries between primary and secondary services.
So the defense is not that it's true. It's that somebody has to decide who gets the specialist.
That's the whole argument. The UK Quality and Outcomes Framework defines serious mental illness as a diagnosis of schizophrenia, bipolar affective disorder, or an episode of non-organic psychosis. That's the register. And that definition excludes personality disorders except schizotypal, excludes substance misuse without comorbid psychosis, excludes eating disorders, and excludes recurrent depression.
And being on that register is what unlocks things.
It provides access to specialist treatment and clinical support. Personalised care plans, enhanced monitoring. It enables targeted resource allocation and commissioning of services. People with serious mental illness die fifteen to twenty years earlier than the general population, and the register is the mechanism that triggers the physical health checks that try to catch that.
So the label is doing real work even though the label is incoherent.
Now the cleanest example of psychosis-triggered routing. Early Intervention in Psychosis services. Introduced across the UK by the National Service Framework in 1999. They offer antipsychotic therapy plus psychological, social, occupational, and educational interventions, and the whole aim is to reduce duration of untreated psychosis. Get people in faster.
And the evidence?
Mixed and contested, and this is the part that should keep everyone honest. The Danish OPUS trial, five hundred and forty-seven patients, first-episode schizophrenia spectrum, showed benefits. OPUS II compared five years of specialized early intervention against two. But the BJPsych commentary on the whole program says studies trying to demonstrate improved outcomes have largely failed to show any enduring effects.
Enduring being the operative word.
Most show some benefit while the service is being delivered, but the gains largely revert once the specialist service is withdrawn. A Cochrane review found low to moderate certainty evidence that these services improve outcomes, from a small number of studies in high-income countries. And the twenty-year OPUS follow-up is the longest randomized follow-up we have.
So the honest read is that specialist routing helps while you're doing it, and the help evaporates when you stop.
Which raises a question the evidence can't yet answer. Is the benefit from the specialist expertise, or from the extra resources attached to the label? Because if it's the resources, then the psychosis criterion is doing nothing except deciding who gets funding.
And if it's the expertise, then the criterion matters enormously and we'd better get it right.
Now the dimensional shift, because the field has already moved. DSM-5 reframed schizophrenia as one of several psychotic disorders existing on a spectrum, with schizotypal personality disorder at the mild end and schizophrenia at the severe end. ICD-11, implemented from January 2022, abandoned the classical schizophrenia subtypes entirely and dropped Schneider's first-rank symptoms from the criteria. Requires two key features, at least one positive.
They retired the first-rank symptoms.
Formally. And ICD-11 added an optional dimensional category, 6A25, symptomatic manifestations of primary psychotic disorders. So you can code the category and then, optionally, code the dimensions. Cohen, Öngür and Babb put the case directly: consistent evidence from genetics, pathobiology, and clinical presentation all support an underlying structure of factors, not categories, as best characterizing psychoses.
Which is where Daniel's dividing line sits now. The science moved to factors. The administration is still running on categories.
And the open question is whether the administrative utility outlasts the scientific critique. Wing's case-mix decisions still have to be made by somebody. The mortality inversion is still sitting there, eating disorders at the top of the mortality table and one percent of the studies. The one-off-episode problem is still sitting there. None of that has been resolved by the dimensional turn.
Hold on, because I want to push on the mortality inversion, since it's the sharpest thing in the whole conversation. Why are eating disorders excluded from the register?
Not because anyone decided they're less serious. It's a definitional artifact. The register was built around psychosis and the conditions that reliably produce it, and eating disorders don't typically present with psychosis. So they fall outside the boundary by construction, not by judgment.
That's almost worse. Nobody chose to deprioritize them. The boundary just happens not to include them.
The same artifact runs the other way, which is the part I find perverse. The register includes people who had a single psychotic episode with no current symptoms. Full recovery, years ago, nothing ongoing. They're administratively serious. Meanwhile someone with lifelong severe functional impairment from a non-psychotic condition is administratively not.
The label tracks history, not current state.
It tracks whether psychosis ever appeared, not whether anything is wrong right now. Which is a strange thing to build a severity metric on.
Yet removing the label would cut the funding.
That's the trap. Pina and colleagues, the UK commentary, note the criteria are clearly medically dominated and that the register approach may narrow approaches to support. But they also document that the register is what triggers the support in the first place. You can't just delete it without replacing the mechanism.
Where does that leave the answer to Daniel's question? Outmoded, or still helpful?
Both, in different registers of the word. As a scientific construct, it's poorly defined and the evidence doesn't support it as a severity measure. As an administrative mechanism, it's the thing that gets people into specialist care and triggers physical health monitoring. The debate is whether poor reliability is a price worth paying for functional routing, and nobody has a clean answer.
The ICD-11 dimensional category is the field's bet that it will eventually be replaceable.
It's a bet, and it's optional, and it's sitting in the manual as an option rather than a requirement. Which tells you the field isn't confident either.
Hilbert: You keep saying register. The register is a form.
Go on.
Hilbert: I did a stint in a GP surgery, doing the coding. Late in it. The QOF register wasn't a clinical judgment. It was a checkbox on a template. If the code went in, the patient was on the register. If the code didn't, they weren't. And the code went in when a psychiatrist's letter landed on the desk and somebody read it and picked the right entry from a dropdown.
The whole construct reduces to which box gets ticked.
Hilbert: There were two dropdowns. One for the diagnosis, one for whether they were on the register. Sometimes they didn't match. You'd have a diagnosis of schizophrenia from nineteen ninety-six and a patient who hadn't been seen in eleven years, and they're on the register because nobody took them off. Nobody's job was to take them off.
The one-off-episode problem isn't even a decision. It's just an absence of a decision.
Hilbert: The eating disorder patients. There were two on our list, severe, both under the community team, one of them had been in hospital twice that year. Neither one on the register. Not because anyone ruled them out. Because the code for it wasn't on the template.
That's the mortality inversion as a dropdown menu.
The physical health checks.
Hilbert: Those came off the register. If you weren't on it, you didn't get the call for the annual check. That's the bit that stays with me. The check itself is twenty minutes and a set of bloods. The thing that decided who got it was a box somebody ticked in a letter eight years earlier.
I keep coming back to the fact that the person doing the ticking had no authority over any of it.
Hilbert: Nobody ever asked me whether it was right. They asked me whether the template was complete before the end of the quarter.
The construct is fuzzy in the literature and rigid in the software.
Hilbert: It's rigid in the software. That's the part that's hard to change. You can revise a definition. You've got to get somebody to rebuild the template, and nobody's paying for that.
Which is the strongest version of Wing's argument I've heard, and it isn't a scientific one at all.
If the construct has poor reliability but real administrative utility, and it's baked into the software that nobody wants to rebuild, where does that leave us?
I think it leaves the science and the plumbing running on different clocks. The dimensional turn is real and it's happening in the manuals. The routing decisions are still being made by a register built on a symptom boundary that the field has already partly abandoned.
The two live contradictions don't go away. Eating disorders sit at the top of the mortality table and outside the register. People with a single resolved episode sit inside the register with nothing currently wrong.
If the construct can't be defended scientifically, the honest question is what replaces it for the case-mix decisions Wing described. Because those decisions don't stop just because the construct is fuzzy.
If psychosis really is a structure of factors rather than a category, then the boundary between primary and secondary services is drawn on something the evidence no longer supports as a single thing.
The bet the field is making is that the dimensional category in ICD-11 eventually becomes the routing mechanism. That it becomes possible to allocate on factors instead of categories.
Or the administrative utility simply outlasts the critique, and we keep using a word that eighty-five percent of studies never bothered to define.
If you take one thing from this, take the fact that the line we use to sort mental illness by severity was never sorting by severity. It was sorting by a symptom that a third of the time belongs to a different disease entirely.
The category at the top of the mortality table isn't on the list, because the list was built by asking a different question.
This has been My Weird Prompts. Our producer is Hilbert Flumingtop.
If you want to send us something, leave us a review wherever you're listening. It helps people find the show.
We'll be back soon.
See you then.
