What Psychosis Classification Can Learn from Epilepsy Classification
The Future of “Schizophrenia”
“Why have psychiatrists failed to reach agreement about the diagnosis of schizophrenia? In a nutshell, it is because schizophrenia is an idea without a single defining principle… The problem with schizophrenia is that its intention is ambiguous, and therefore the attempt to reach a consensus on its extension is doomed to frustration. The clarity of Kraepelin’s original idea has been lost, and it now has, not one, but two or even three totally different defining principles.”
I Brockington, Schizophrenia: Yesterday’s Concept (1992)
A little over a century after Bleuler gave us the word, psychiatry cannot agree on what, if anything, to do with “schizophrenia” as a diagnostic construct.
In 2022, Schizophrenia Research devoted a special issue — Re-Inventing Schizophrenia: Updating the Construct — to this question, and the forty or so editorials published in it make for a peculiar reading experience. There is broad agreement on the ways in which the construct of schizophrenia is limited and imperfect, but there is little agreement on what to do about it.
The complaints are by now familiar to most professionals in the mental health field. The category is heterogeneous, in terms of the diversity of clinical profiles and outcomes as well as in terms of likely bundling together many different etiological processes. While there are promising developments in discovering biotypes of psychosis (see B-SNIP biotypes), no biomarker, or even a set of biomarkers, corresponds specifically to “schizophrenia.” There are no points of rarity that separate schizophrenia from its neighbor psychotic disorders; schizophrenia shades into schizoaffective disorder, into bipolar disorder, into psychotic depression, without any natural discontinuity. The risk genes for schizophrenia are shared with bipolar disorder, schizoaffective disorder, and other psychiatric disorders. Its diagnostic thresholds are semi-arbitrary. Pat McGorry argues that the six-month duration criterion in the DSM has “institutionalized the basic flaw” of the concept by hard-coding chronicity into it. The diagnosis has clinical utility, but it is thin, and dimensional approaches to psychosis have plausibly similar or superior practical utility. Interestingly, there is still no agreement on what is even central to the concept. DSM and ICD have enshrined positive symptoms (such as delusions and hallucinations) as central, but the originators of this concept, Kraepelin and Bleuler, had emphasized chronicity, prognosis, cognitive impairment, thought disorder, and negative symptoms. It is well-demonstrated that negative symptoms and cognitive impairment are stronger determinants of long-term outcomes in those diagnosed with schizophrenia than positive symptoms.
Confronted with this list of problems, the field has divided into preservationists, reformers, and discarders when it comes to the “schizophrenia” construct. Do we say that we have nothing demonstrably better yet, so we keep it and iteratively improve? Or do we reform the DSM concept, perhaps to emphasize cognition/prognosis as central? Or do we discard the concept in favor of a psychosis spectrum?
The Epilepsy Comparison
Among the preservationists/reformers, Sophia Frangou offers a defense using an analogy with epilepsy.
The term “epilepsy” is archaic, like “schizophrenia.” It is a Greek term from the fourth century BCE meaning ‘to seize’ or ‘to attack,’ freighted originally with the belief that a seizure was the work of a demon or a god. Epilepsy has historically been subject to tremendous stigma and discrimination and still is today to some extent. And yet there is no movement to rename it; the epilepsy community has put its energy into public education instead.
Both conditions are defined by an obligatory core feature (seizures/psychosis) that persists across highly variable presentations with various companion features. Epilepsy is a disorder of the brain characterized by an enduring disposition toward recurrent seizures, together with the cognitive, psychological, and social sequelae; schizophrenia, on the same model, Frangou suggests, would be an enduring disposition toward recurrent psychotic symptoms with associated dysfunctions.
Just as a single seizure isn’t epilepsy, first-episode psychosis isn’t necessarily schizophrenia. Frangou’s observation is that in medicine these boundaries are set by consensus and revised as understanding improves. Epilepsy draws its line chiefly by recurrence, or by a high risk of recurrence signaled by EEG epileptiform activity or a neuroimaging abnormality. The parallel task in psychosis, on this reading, would be demonstration of recurrence or the presence of robust predictors of psychotic recurrence. (DSM’s 6-month duration requirement for schizophrenia is one way of enriching for chronicity.)
Extending the Epilepsy Comparison
I have also been thinking about a comparison between schizophrenia and epilepsy (and had started doing so before reading the Frangou piece), but my comparison is more along the lines of how classifications of epilepsy can inform classifications of psychosis.
The establishment framework of epilepsy classification comes from the International League Against Epilepsy. The current ILAE system is a multilevel diagnostic scheme: classify the seizure type, then the epilepsy type, then, where possible, the epilepsy syndrome, with a parallel axis for etiology. It is an integrative characterization that relies on the clinical picture, EEG findings, neuroimaging, and genetics, and it is built to be usable by the general neurologist and to guide treatment.
The rival tradition comes from Hans Lüders and his colleagues, first as the Semiological Seizure Classification in the late 1990s and later as a broader four-dimensional scheme (4D-EC). Lüders insists that the description of a seizure (that is, its semiology, what is actually observed to happen) should be kept separate from what we infer about it. This is because the two don’t correspond perfectly; the same seizure semiology can arise from different localization patterns. The semiological classification is deliberately agnostic about the EEG, about localization, about cause; it describes an “automotor” or a “dialeptic” seizure purely as an observable phenomenon. Around that descriptive core the four-dimensional system arranges three further, explicitly independent dimensions: the epileptogenic zone, the etiology, and the associated comorbidities.
Where ILAE would call “focal impaired-awareness seizure with automatisms” (the old complex, partial seizures), 4D-EC would call “automotor” on semiology, without reference to “focal.” Lüders’ objection to the ILAE approach is that its terminology smuggles inference into observation. To call something a “focal motor seizure” based on clinical observation is already to have made a claim about how/where in the brain it originates. The “epileptogenic zone” in 4D-EC is a theoretical, provisional construct. It is operationally the region whose removal renders the patient seizure-free. It can be expected to be revised as work-up proceeds without disturbing the semiology at all. The clinical description is stable; the inferences are revisable.
Epilepsy has an objective anchor that psychosis currently does not. There is an objective correlate of the seizure (the ictal discharge on EEG) and nothing equivalent for psychosis. The lesson for psychiatry, IMO, is in the general philosophy. We need to distinguish what we observe from what we infer; keep the dimensions independent (e.g. not conflate symptomatology, prognosis, and cognition), so that reassessing and revising one does not destabilize the others; and treat the inferential layers as provisional constructs rather than as clinical facts.
Enduring Disposition Towards Recurrent Psychosis
Let’s come back to the idea of schizophrenia, akin to epilepsy, as an enduring disposition toward recurrent psychosis. The current DSM does not define schizophrenia this way. The DSM construct of schizophrenia nor the old Kraepelinian and Bleularian concepts are “a tendency toward recurrent psychosis.” The DSM concept requires a stipulated period of functional decline, symptom duration, and a specified mix of clinical features, the validity of which is all quite unclear. An epilepsy-style definition sweeps those thresholds aside in favor of something closer to recurrent psychosis.
Aside from the recurrence, the second aspect is unprovoked. Isolated seizures with low risk of recurrence are excluded from the diagnosis of epilepsy. A seizure provoked by an acute insult such as a metabolic derangement, a toxin, an acute brain injury is an acute symptomatic seizure, not epilepsy. “Unprovoked psychosis” would be psychosis that is not an acute symptomatic phenomenon: not the psychosis of intoxication, of delirium, of the acute phase of an autoimmune encephalitis, etc. This is basically the primary versus secondary psychosis distinction.
Schizophrenia in this sense would be conceptualized as a primary psychotic disorder defined as an enduring disposition toward recurrent, unprovoked psychosis with associated cognitive, psychological, and social dysfunctions.
Psychosis as Multidimensional Space
What would a Luders-style characterization of psychosis look like? What would the dimensions be? I believe it would look something like below:
1. Symptom dimensions
a. Positive
b. Negative
c. Disorganization
d. Affective admixture
e. Catatonia
2. Course and staging
a. Acute/chronic
b. First-episode, recurrent, persistent
c. Onset character (abrupt vs insidious)
d. Premorbid functioning and prodromal decline
3. Cognitive functioning
Cognitive profile (processing speed, working memory, verbal learning, executive, social cognition), severity, and trajectory (static vs declining)
4. Functional impairment and disability
5. Etiology
Primary vs secondary (e.g. secondary to high-penetrance CNVs, 22q11.2, anti-NMDAR and other autoimmune disorders, substance-induced, neurodegenerative, etc, etc).
Candidate biomarkers for primary psychoses, e.g. B-SNIP biotypes and polygenic risk scores
6. Treatment response
Responsive/resistant to standard antipsychotics, clozapine-responsive/ultra-resistant
Any particular case of psychosis occupies a particular position in this multidimensional space, and different DSM diagnoses consist of different regions of this space.
Where “Schizophrenia” Sits
The prototypical case of schizophrenia in a textbook carves out a describable region of this space: presence of positive symptoms (hallucinations, delusions, disorganization); a substantial burden of negative and cognitive features; primary rather than secondary etiology; prodromal decline preceding onset; chronic or recurrent course; and high baseline functional impairment.
But move along any one of the axes, away from the center of that region, and there is no sharp boundary to be found. Shorten the course and we enter “schizophreniform” and “brief psychotic disorder” territory without ever stepping over a natural discontinuity. Turn up the affective dial and we slide through schizoaffective disorder into psychotic mood disorder along a gradient. Follow cognition and we will find cases of DSM schizophrenia without cognitive impairment and bipolar psychosis with cognitive impairment. Follow prognosis and we will find cases of good prognosis DSM schizophrenia. Etc.
Schizophrenia exists as a region of multidimensional space but without any natural boundaries separating it from other psychoses.
Agreeing on the Contours
Rajiv Tandon, in his editorial in the issue, is in favor of holding the center, that is, retaining the construct of schizophrenia pragmatically while working collectively toward one shared, clearly bounded definition. The first order of business, in his view, is agreeing on “the contours of schizophrenia.”
I am also in favor of agreeing on the contours, though not quite as he intends. Before we relitigate validity, he says, we should ensure we are all pointing at the same diagnostic construct. I believe that instead of a pragmatic DSM-style schizophrenia category, a multidimensional space characterization of psychosis is a better way for the field to agree on the contours of the problem. A common set of axes is a far more stable shared reference than a diagnostic label that different people operationalize differently or where operationalization relies on arbitrary cut-offs. If we want everyone talking about the same thing, we have to agree on a coordinate system. Furthermore, it’s not just that we want people to agree on “schizophrenia,” we want them to agree on psychosis broadly. For too long we have neglected the spectrum of psychotic experiences and devoted our attention to schizophrenia.
The contours we can come to agree on are the contours of a dimensional space, and “schizophrenia” becomes a fuzzy region within it. That is much closer to William Carpenter’s position of schizophrenia as a clinical syndrome without an essence. We are better off reorganizing the schizophrenia spectrum as primary psychoses and treating the specific psychopathologies present in each individual.
An illustrative precedent of retaining a syndrome within a dimensional classification comes from an unlikely corner… the ICD-11 classification of personality disorders. ICD-11 abolished the categories of personality disorders altogether, and created a rating of severity and a set of trait-domain qualifiers (Negative Affectivity, Detachment, Dissociality, Disinhibition, Anankastia). But the ICD-11 architects kept one of the old categories as an optional “borderline pattern” specifier (added, by most accounts, reluctantly, under pressure from clinicians who weren’t ready to give up their beloved borderline personality disorder category). The borderline pattern specifier is not a categorical diagnosis in the old sense. It is simply a way of pointing towards a fuzzy, heterogenous cluster of dimensional profiles to record that these traits have cohered into a recognizable, clinically familiar shape.
That may very well be the template for future schizophrenia in a classification of primary psychoses. Build a multidimensional space for characterizing psychoses and retain “schizophrenia” as an optional pattern specifier for those who cannot bear to give it up, the ones who think it offers a clinical utility that cannot be replaced by a dimensional characterization.
DSM-5 schizophrenia, however, is an awkward construct that cannot give up chronicity and cognitive impairment but also cannot commit to it. It is arbitrary and unprincipled. Even if we retain a schizophrenia specifier, I do not believe it is wise to adopt the DSM-5 definition.
I think there are two reasonable moves. Either we reconceptualize schizophrenia, epilepsy style, as “recurrent, unprovoked psychosis,” in which case it dissolves into the broader dimensional characterization of primary psychoses and there is no need for schizophrenia specifier at all because schizophrenia simply is primary, recurrent psychosis.
Or, we embrace the textbook schizophrenia prototype and approach schizophrenia as a region of multidimensional space characterized by a prodrome, by cognitive impairment, with a substantial burden of negative symptoms, and with a chronic or recurrent course, in addition to positive symptoms. That is, unlike DSM-5, we incorporate these features into the definition itself.
It is still reasonable for a skeptic to ask, however: why bother? Why hold on to this particular region of psychotic space as something special and distinctive? Is it because we cannot give up the hope yet of finding a unitary disease process behind this profile of clinical features? There is no specific treatment for schizophrenia, compared to other psychotic disorders, no specific biology, and no single defining principle. So what is so special about this yesterday’s concept?
See also:
Awais, good work, I like how you keep the description observing separate from what clinically we "feel" about it. In psychiatry have needed this discipline for a long time, and the multi-spatial dimension you build around it is a good, stable reference that a clinician can actually use operationally, in a very different way from the way we are using it today. I like it, and I'm with you on the whole program. I also like the coordinated system because it anchors the field on the center of the fact that schizophrenia is a category, and you make a good case for that. Bravo!
Now allow me offer constructive feedback on a piece that I think is missing. I look at the six axes, and I see that five of them are entirely inside the person. The sixth, called Functional Impairment, is a deficit that the patient carries around, but not something that lives in the space that is in between the person and their social construct, their housing, their relationships, their work, their families, the people they love. There's no access for a social context anywhere in this system.
If with epilepsy all we envy about is the EEG the pull towards finding a physical correlate keeps bringing the social off the table. But I think you would agree that the social is where a lot of the outcome actually lives. We see it in the Trieste tradition, in the Clubhouse model, how they are able to move people towards real function, even though the biology is a mystery, and the social determinants literature is now pointing strongly the same way. In your axis the thing that predicts long-term function is staying outside this map.
Here's my ask: could you consider social context and social recovery as a dimension in its own right, putting it alongside symptoms and cognition instead of folding it into one category called Functional Impairment? If the psychosis lives in a multi-dimensional space, we're missing the coordinate of the person's world. Putting it back in the picture makes the whole model stronger and not softer. I would love to see you take that dimension as one axis further. To be fair, I know you're one of the few people in this debate who will give the social its due, and I think this is what the framework is asking for.
Thank you. A lot of worthwhile observations that bear thought. Here are two reactions, brief.
1. Many of the ‘symptoms’ of SCZ are essentially first-personal in nature. Real progress dealing with SCZ (or better, whatever classifications succeed it) will need to involve a serious attempt to investigate the phenomenology of the symptoms. (Some attempts exist. Not nearly enough, and many of them are methodologically suspect IMO.) For example, ‘hallucination’ is not just one thing. I think that right now nobody really knows how best to characterize the relevant types of hallucination. Same for delusion. Until we have a better handle on the phenomenology of these things, we don’t really even know what we’re talking about.
2. Most attempts to understand the ‘symptoms’ of SCZ begin with diagnosed patients. That seems totally fine until we remember that (as you say) diagnosis is currently a mess. For example, diagnosis requires substantial dysfunction. That makes sense for clinical reasons—we don’t need to be treating people who don’t need treatment—but for understanding the nature of the symptoms, restricting attention to the cases where ‘symptoms’ are associated with dysfunction could be blinding us to a lot of evidence. What might we learn, for example, from people involved in the Hearing Voices Network, many of whom hallucinate frequently but do not report clinically significant dysfunction? If we already knew that DSM-styled SCZ is a thing, a natural kind, then it would make sense to set aside persons who are not diagnosed. But (as you also say) we don’t know this. And yet, study after study after study begins by selecting only DSM-diagnosed persons as the subjects.
Thanks again. Lots to consider, here.