Depression as a state measured through its contents
Working Papers ·
written 19 September 2026 · open
Download the paper (.docx)
Dr. Toye Oyelese, MBBS, CCFP, FCFP
Clinical Associate Professor, Department of Family Practice
University of British Columbia
This is a working paper, written in September 2026. It sets out the mind as a regulated system, and it will not stand on its own.
Almost none of this is proven. It is what makes the most sense to me.
In 2014 Eiko Fried and Randolph Nesse took the STAR*D data — 3,703 depressed outpatients — and asked a simple question. How many different ways does depression actually present?
They found 1,030 unique symptom profiles.
Of those, 864 were reported by five patients or fewer. 501 were reported by exactly one person. The single commonest profile accounted for 1.8 per cent of the sample. On average there were 3.6 patients per profile.
Then they did the thing that matters. They controlled for overall severity, to see whether this was just one illness at different intensities.
It made no difference. The heterogeneity did not reduce.
Their paper is called Depression is not a consistent syndrome, and they read the result as a problem: it explains the failure to find biological markers, the poor reliability of the diagnosis, and the difficulty of treatment trials. They are right about the consequences. I think they are wrong that it is a problem.
I think it is the expected result, and I think a framework that could not produce it would be the one in trouble.
The Navigational Mind Framework starts from one premise. Uncertainty is not a problem to be solved; it is the medium. Uncertainty is constant change in every unit of change, so what I do not know will always exceed what I know. I call that the Null Hypothesis.
From it: the mind is plural. Seven Residents, each with its own question, all present from birth — Trust, Autonomy, Initiative, Industry, Identity, Intimacy, Generativity. Each has its own three states, set out below. The narrating I is not one of them; it is the Archivist, and it produces the account of what the Residents already did.
The NMF explains that the mind at rest is not resting. It is holding Trust down, all day, at a cost. Anxiety is that holding failing. Depression, I think, is the account running empty.
This paper needs only two things from all that. The Residents have states. The Archivist does the talking.
A depressive state and a depressive presentation are not the same object.
The state is what the house is doing. Two things can be wrong with it. The reassurance can fail, which is the homeostatic axis. Or the account can be empty, which is the capacity axis. They are independent — you can be exhausted without being frightened, and frightened without being exhausted.
The presentation is what the person says and what a clinician records. And that is not the state. It is the Archivist's account of whichever Residents happen to be at their floor.
Symptom checklists measure the presentation. They have never measured the state, and there is no reason to expect the two to be the same shape.
| The state | The presentation | |
|---|---|---|
| What it is | what the house is doing | the Archivist's account of it |
| Varies with | reassurance and capacity | which Residents are at their floor |
| Number of kinds | two axes, so three states | Combinatorial |
| Measured by | nothing currently | symptom checklists |
| Stable across people? | should be | cannot be |
This is why a symptom list cannot be made more reliable by being made more careful. The unreliability is not measurement error. The instrument is pointed at the output of a narrating function, and the narration varies with which parts are down.
Seven Residents. Three states each. That is 2,187 possible configurations of the house.
| Resident | States |
|---|---|
| Trust | secure → uneasy → threatened |
| Autonomy | choosing → pressured → coerced |
| Initiative | exploring → hesitant → frozen |
| Industry | productive → struggling → stalled |
| Identity | clear → questioning → lost |
| Intimacy | open → guarded → hidden |
| Generativity | creating → maintaining → depleted |
A depressive state does not require any particular Resident to be down. It requires the account to be empty, or the reassurance to have failed, or both. Which Residents end up at their floor depends on the person, the history, and what has been spent.
So the state is one thing — or, as I will argue, three. The presentations are combinatorial. And a study that samples presentations will find a very large number of them.
Here is the arithmetic, and I want to be careful about what it does and does not show. If you draw 3,703 patients from 2,187 configurations with every configuration equally likely, you would expect about 1,785 distinct profiles, of which around 38 per cent would be seen once. Configurations are obviously not equally likely — some combinations are far more common than others — and uneven likelihood lowers the number of distinct profiles and raises the proportion seen once.
Fried and Nesse observed 1,030 distinct profiles with 48.6 per cent seen once. Both numbers move in exactly the direction uneven sampling predicts, from a starting point set by the size of the house.
I will not overstate this. It is an order-of-magnitude check, not a fitted model, and I have not estimated the true distribution of configurations. What it shows is that a seven-part house with three states each produces heterogeneity of the size actually observed. A house of two parts could not. A house of thirty would produce far more.
This is the part of their result that surprised them, and it is the part I think is most diagnostic.
If depression were one condition at different intensities, severity would be the hidden variable. Control for it and the profiles should converge. They did not converge at all.
On this account they could not have. The variation is not in how far down the house is. It is in which parts are down. That is a categorical difference, not a dimensional one, and controlling a dimensional variable cannot remove it.
Two patients can be equally severe and share almost no symptoms, because severity says how much has been spent and says nothing about where.
This makes the severity-control result a test rather than a curiosity. A dimensional account predicts convergence. This account predicts none. The data show none.
If presentations are selected by which Residents are down, then no single symptom should be necessary. Including the one everybody assumes.
Sadness is one Resident's signal. It belongs to Intimacy, whose question is can I be truly seen — the same territory as separation distress. To require sadness in the definition of depression is to require that one particular Resident be among those affected.
There is no reason to expect that, and the literature says it often is not. Non-dysphoric depression is a described entity: depression whose core symptoms do not include sadness. Patients report guilt, worthlessness, pessimism and a sense of failure, with vegetative symptoms, and do not endorse sadness. In some series they do not endorse anhedonia either. It runs at 27.4 per cent of nursing home residents, and at thirteen-year follow-up carries raised risk of death, functional impairment and cognitive decline. It is not a mild variant and it is not a diagnostic error.
Loss of meaning has the same shape. Meaning is the Rhythm that reads Generativity. Requiring it requires that one Resident's readout to be low. It is commoner than sadness, because an empty account drags Generativity toward depleted, which is its own third state. It is still not necessary.
So the two symptoms most people would call definitional are contents, not the state. Requiring either of them is requiring a particular narration.
If the two axes are independent, then depression is not one condition. It is at least three, and they need different handling.
| Type | State | Presents as | What helps |
|---|---|---|---|
| Depletion | capacity spent, reassurance intact | flat, slow, anhedonic, not frightened | restore capacity; do not ask for a step yet |
| Resident-state | reassurance failed, capacity intact | one or more Residents at their floor, fuel still present | work at the Resident; a true step will move |
| Both | reassurance failed and account empty | the alarm running on an empty account | capacity first, then the step |
The literature already describes two of these as distinct entities, arrived at without any of this.
Melancholia carries the depletion picture: psychomotor retardation, profound anhedonia, unreactive mood, altered cortisol reactivity, a more recurrent and severe course, and discrimination from non-melancholic depression on imaging. And the sentence that matters most for my account — in melancholia, depressed mood tends to arise out of the blue rather than in response to antecedent stressors. Depletion with no deforming force.
Anxious distress carries the combined picture. It predicts poorer outcome, lower remission, slower response and more side effects, and it outperforms comorbid anxiety disorder diagnoses as a longitudinal predictor — which is what you would expect if it marks a second axis failing rather than a second illness present. About three-quarters of depressed patients meet it, so on this account the combined type is the common one and the pure types are the minorities.
The middle type is the one I expected least support for. One latent class analysis reports a severe form characterised by a relative lack of neurovegetative symptoms alongside high dysphoria, guilt and suicidal thoughts. Capacity intact, Residents at their floor. That is the type, described by someone with no reason to be looking for it.
Heterogeneity should not reduce under severity control. It does not.
No single symptom should be necessary, including sadness. Non-dysphoric depression exists and is serious.
Biological markers should fail for a syndrome defined this way, and should be findable for a state defined by mechanism. Markers have failed for MDD as a whole and have had more success for melancholia specifically, which is the type most cleanly defined by mechanism rather than content.
The three types should respond differently to the same intervention. Behavioural activation should move the Resident-state type and not the depletion type, because a spark does nothing in an empty tank and each failed attempt adds load. That is a testable prediction and I do not know whether it holds.
And two patients with identical severity scores should be able to share almost no symptoms. They can, and do, and the syndrome view has no account of why.
Fried and Nesse note that the heterogeneity explains why treatment studies struggle. I want to put the mechanism more sharply, because it changes what a trial is doing.
A randomised trial of an antidepressant against placebo in major depressive disorder is, on this account, a trial in a mixed population of three different states. If an agent works on one of them and not the others, and that one is a third of the sample, the effect is diluted by two thirds before anything else happens.
That is not a power problem. Adding patients does not fix it, because every additional patient is drawn from the same mixture. It is a problem of the inclusion criterion selecting on presentation when the agent acts on state.
It also predicts the specific pattern the field reports: modest average effects, large individual variation, and a stubborn placebo response. A mixture of responders and non-responders with no way to tell them apart in advance produces exactly that.
And it predicts where the exception should be. If any subtype is defined closely enough to a mechanism, trials in that subtype should behave better. Melancholia is the candidate, and melancholia is where the claim of better response to biological treatment has always lived.
This makes the three types a prediction about trial design rather than only about diagnosis. Stratify by state rather than by presentation and the effect sizes should separate. Nobody has done that, because the state is not currently measurable — which is the same wall as before.
The strongest objection is methodological, and it is not small.
When the field lets the data find the classes, it mostly finds severity. A systematic review of latent class analyses concludes that they group patients by overall severity rather than into qualitatively different symptom profiles, and that the studies fail to give adequate evidence for qualitatively distinct subtypes. If my three types were real and separable, data-driven subtyping ought to have found them.
My answer is that latent class analysis clusters symptoms, and my three types differ by mechanism. Depletion and the combined type can produce overlapping symptom profiles, because exhaustion looks like exhaustion whether the account was emptied by alarm or by nothing at all. Separating them requires measuring load and capacity, which no symptom instrument does.
But I want to state plainly what that answer costs. It means my split cannot be validated the way the field validates subtypes. It would need a different measurement, and the obvious way to measure load and capacity is to ask the person — which returns me to clinical impression, dressed as a variable. That is the sharpest objection to everything here and I do not have a solution to it.
Second. The arithmetic in section five is a consistency check and nothing more. A framework with a different number of parts would produce a different expected heterogeneity, and I have not shown that seven is the number that fits best. I have shown that seven is not the wrong order of magnitude. Those are different claims.
Third. Fried and Nesse's own conclusion is that the way forward is symptoms rather than syndromes — study the individual symptoms and their causal relations, and drop the category. That is a coherent programme and it does not need my account at all. If network models of individual symptoms succeed, the state-and-content distinction becomes unnecessary machinery.
Fourth, and this one I cannot dismiss. A general factor of psychopathology fits the data better than distinct factors in much of the work on this. If psychopathology is substantially one dimension, then Residents with structurally different failure modes is a harder position to hold than I have made it sound.
What is invariant. If no symptom is necessary, and no Resident is necessary, what makes a depressive state depressive? My candidates are the protraction itself, the failure of return, and the Loop no longer completing. None of those is a content, and none would be visible to a symptom checklist. I do not know which, or whether it is one of them.
Whether the Archivist's account hardens. A hard week produces a narration that passes. A depression produces one that does not. Whether that hardening is a separate mechanism or just duration, I cannot say.
Where suffering sits. A person can be compensated, within capacity, and in unbearable pain. A criterion built on function says that is not pathology. Forty years says otherwise, and I cannot reconcile them.
Whether the three types are three, or two, or a continuum I have cut arbitrarily.
And whether any of this can be measured without asking the person.
Fried EI, Nesse RM. Depression is not a consistent syndrome: an investigation of unique symptom patterns in the STAR*D study. Journal of Affective Disorders, 2015; 172: 96-102.
Fried EI. Problematic assumptions have slowed down depression research: why symptoms, not syndromes are the way forward. Frontiers in Psychology, 2015.
van Loo HM and colleagues. Data-driven subtypes of major depressive disorder: a systematic review. BMC Medicine, 2012.
Ulbricht CM and colleagues. The use of latent class analysis for identifying subtypes of depression: a systematic review. Psychiatry Research, 2018.
Gallo JJ and colleagues. Depression without sadness: functional outcomes of nondysphoric depression in later life. Journal of the American Geriatrics Society, 1997.
Melancholia: reviews of biological differentiation from non-melancholic depression, and the Lundby fifty-year follow-up on recurrence, chronicity and severity.
Zimmerman M and colleagues. Validity of the DSM-5 anxious distress specifier for major depressive disorder. Depression and Anxiety, 2019.
Caspi A and colleagues on the general factor of psychopathology.