Diagnosis as Welfare-Relevant Attribution

Working note. Diagnosis is treated here as a multi-valued attribution and a medical instrument whose value lies in its expected welfare consequences, not as a binary label or the detection of a primitive disease object.

The starting point is welfare. Social meaning, institutional legitimacy, medical authority, and diagnostic practice matter because they affect welfare bearers.

There are real states: pain, distress, impairment, attention difficulty, grief, low mood, cognitive variation, behavioral dysfunction, risk, incapacity, identity-related distress, and social dysfunction. Biological and psychological mechanisms can also be entirely real. But the reality of a state or mechanism does not by itself show that a diagnosis should be made.

state exists ≠ diagnosis justified
mechanism exists ≠ one diagnostic attribution required

A disease category is therefore not treated as the primitive ontological unit. It is an attributional tool for organizing variables and mechanisms so that a state becomes medically legible and actionable. A disease-like object can be understood as a state made available to medical action through a diagnostic framework.

diagnosis ≠ detection of a self-standing disease entity
diagnosis = attribution → interpretation and action → welfare consequence

This does not make diagnosis unreal or arbitrary. A diagnostic attribution can be more or less accurate about the provenance of a presentation, more or less useful at routing action, and more or less welfare-positive. The claim is only that correspondence to a category is not the final point of medical recognition.

From raw state to welfare consequence.

The basic layers are:

raw state
→ potential diagnostic object
→ diagnostic framework
→ actual diagnosis or diagnostic bundle
→ expected welfare consequence

A raw state is a real psychological, cognitive, behavioral, social, or bodily condition.

A potential diagnostic object is a state that could, in principle, be made medically actionable. This category is not limited to what current DSM or ICD systems recognize. Keeping it open preserves the possibility that present systems miss cases which future systems could recognize beneficially.

A diagnostic framework is a selection mechanism: a classification system, threshold, screening protocol, clinical assessment practice, or future method that maps presentations into diagnostic outcomes.

An actual diagnosis or diagnostic bundle is the attributional act produced by that framework. It may classify, label, route care, activate treatment or accommodation, alter prognosis, affect self-understanding, create stigma, consume resources, and change the behavior of institutions or other people.

The welfare consequence is the expected net effect generated by the diagnostic act relative to the relevant alternative.

The diagnostic attribution—not an isolated disease object—is therefore the main unit of evaluation.

Presentation and evidence.

Let:

$$s\in S_t$$

denote a baseline condition-state at time $t$. It may contain pain, distress, impairment, risk, agency-relevant difficulty, functional loss, or other states that create potential value for recognition.

Let:

$$x\in X_t$$

denote a person's VMS-mediated presentation. It may include behavior, functioning, affect, history, self-report, context, task performance, and clinical observation. These are evidential variables, not primitive disease entities.

Self-report is especially important but remains mediated by memory, interpretation, language, self-understanding, and social context. It is evidence about Self, Mechanism, state, and world—not direct identity with any one of them.

Diagnosis is multi-valued.

Diagnosis is not adequately represented by:

$$\mathrm{diagnosis}\in\{0,1\}.$$

Let:

$$a\in A\cup\{\varnothing\},$$

where $a$ is a diagnostic attribution and $\varnothing$ is no formal diagnostic attribution. $A$ is the space of possible attributional outcomes, not merely the list of current DSM or ICD labels.

An attribution may locate a presentation as:

a DSM/ICD category
a mechanism-level pattern
a temporary or recurrent state
an environmental response
a Self-related pattern
a Self/Mechanism conflict
a mixed structured profile
no formal diagnosis

This accommodates comorbidity, diagnostic substitution, subtyping, mixed causal pathways, structured profiles, and non-diagnostic outcomes.

A diagnostic system is an attribution map:

$$g:X_t\to A\cup\{\varnothing\}.$$

The central question becomes:

which attribution
at which level and resolution
for this presentation
with what downstream welfare effect

The same presentation may be assigned to disorder, temporary state, mechanism-level dysfunction, environmental response, Self-related conflict, or no diagnosis. These are not interchangeable names; they route different interpretations and actions.

Diagnostic value is marginal.

A person may be in a bad welfare state before diagnosis. That baseline can create a high potential value for recognition, but it is not itself the value of the diagnostic act.

Let:

$$o(a,x;\Omega_t^{conv})$$

denote the full counterfactual outcome-state generated when attribution $a$ is made for presentation $x$ under conversion environment $\Omega_t^{conv}$.

The outcome-state includes more than treatment: accommodation, self-understanding, family response, institutional access, stigma, prognosis, resource allocation, future options, public meaning, and any later consequences causally affected by the attribution.

Let:

$$o(\varnothing,x;\Omega_t^{conv})$$

denote the outcome-state when the evaluated diagnostic system makes no formal attribution. This does not mean nothing happens. Informal labeling, ordinary care, family interpretation, self-interpretation, social reaction, or delayed support may still occur.

Let $W^{soc}(o)$ be the social welfare level of outcome-state $o$. One representation is:

$$W^{soc}(o)=\sum_{j\in N}\lambda_j W_j(o),$$

where $N$ is the set of affected welfare bearers and $\lambda_j$ is the ethical weight assigned to each. The person diagnosed normally bears the largest and most direct effect, but not the only one:

$$W^{soc}(o) = \lambda_iW_i(o) + \sum_{j\neq i}\lambda_jW_j(o).$$

Positive and negative values enter only through comparison. The social welfare yield of attribution $a$ is:

$$\delta W^{soc}(a,x_i;\Omega_t^{conv}) = W^{soc}(o(a,x_i;\Omega_t^{conv})) - W^{soc}(o(\varnothing,x_i;\Omega_t^{conv})).$$

Thus:

$$\text{baseline welfare state}\neq\text{diagnostic value}$$ $$\text{diagnostic value}=\text{welfare effect of attribution}.$$

If $\delta W^{soc}>0$, the attribution is welfare-positive relative to that counterfactual. If $\delta W^{soc}<0$, it is welfare-negative.

The relevant alternative need not always be no diagnosis. It may be:

diagnosis A rather than diagnosis B
diagnosis now rather than delayed diagnosis
current attribution rather than a more precise attribution
diagnostic support rather than non-diagnostic support
coarse rather than refined attribution

The comparison must match the decision under evaluation.

The conversion environment.

The conversion environment at time $t$ is:

$$\Omega_t^{conv} = \{\text{treatments, therapies, accommodations, supports, institutions, social responses, resource conditions}\}.$$

It contains the downstream conditions through which an attribution can be converted into welfare. It may also include the broader social background insofar as that background is exogenous to the diagnostic system being evaluated.

It does not fix the attributional quality of that system. DSM/ICD categories, thresholds, professional attitudes, eligibility boundaries, and policy classifications belong to the attribution system itself, not to the supposedly neutral background.

background fixes conversion conditions
background does not fix attribution quality

The distinction matters because treatment matching, accommodation, blame reduction, or self-understanding can be part of diagnostic value when good attribution produces them. Conversely, a bad category or threshold cannot be protected from evaluation by being placed in the background.

Treatment is one conversion route, not the only one.

Recognition may be welfare-positive even when direct treatment is unavailable. Possible routes include:

prognosis and care planning
prevention and risk communication
avoidance of harmful intervention
institutional accommodation
research classification and coordination
future treatment development
social recognition
de-moralization or reduction of personal blame

Diagnosis can therefore sometimes be part of the intervention. A person may interpret persistent difficulty as laziness, moral failure, or a defect of character. A better attribution can alter that interpretation before medication or therapy occurs. The reverse is also possible: false attribution can damage identity, confirm a misleading self-story, stigmatize ordinary variation, or locate a social failure inside the person.

Severe baseline harm may lower the threshold at which even a nonspecific attribution is beneficial. Severe pain, for example, can make recognition valuable because it opens urgent care, protection, or relief. That still does not turn “disease” into a primitive object. It means the counterfactual cost of non-recognition is high.

Coarse and refined attribution.

A coarse attribution might be a broad DSM/ICD category or threshold-based diagnosis. A refined attribution might specify mechanism, subtype, symptom structure, clinical state, environmental pathway, Self/Mechanism relation, or a treatment-relevant profile.

This is an operational distinction, not a definition of good and bad diagnosis. Coarse attribution can be valuable through consistency, fairness, service access, research coordination, communication, and de-moralization. Refined attribution can also fail if its extra precision is spurious, illegible, costly, or actionably irrelevant.

$$\text{good diagnosis} = \text{socially welfare-positive attribution at the right resolution}.$$

More precise language is not automatically better attribution.

Expected consequences and control.

The diagnostic system chooses diagnostic acts or diagnostic bundles. It does not directly choose every patient, family, school, clinician, institutional, or social response.

For attribution $a$:

$$E[W(a)] = \sum_{p}P(p\mid a)W(p),$$

and for a bundle $B$:

$$E[W(B)] = \sum_{p}P(p\mid B)W(p),$$

where $p$ ranges over downstream paths.

If a later step is controlled by the diagnostic system, it should be included in the diagnostic policy bundle. If it is controlled by patients, families, schools, clinicians, or society, it normally appears as a probability-weighted consequence.

controlled by the evaluated system → part of the choice bundle
not controlled by it → probability-weighted downstream path

This keeps responsibility for the diagnostic choice distinct from uncertainty about what follows it.

Externalities and interaction.

Diagnostic attributions can affect people other than the diagnosed person. Externalities may include family burden, classroom conditions, waiting times, institutional trust, public fear, stigma, and resource availability.

Diagnoses also interact:

$$W(A+B)\neq W(A)+W(B).$$

They are substitutes when their combined value is less than the sum of their isolated values:

$$W(A+B)<W(A)+W(B).$$

Two labels may route to overlapping support, so the second adds little after the first.

They are complements when:

$$W(A+B)>W(A)+W(B).$$

A complex case may receive the correct support only when two attributions are jointly recognized.

Interaction does not break the welfare framework. It shows only that welfare cannot always be treated as additively separable across isolated diagnoses. The relevant object is often the whole diagnostic configuration:

$$W(B),$$

not a simple sum of isolated $W(d)$ terms.

Strong harm and scarcity are different.

A diagnosis is strongly welfare-negative if removing it without replacement increases total welfare. The attribution is net harmful even before considering a better use of the same resources.

actual bundle contains d
→ remove d without replacement
→ total welfare rises

This is a strong form of overdiagnostic harm: stigma, wrong treatment, anxiety, or identity damage may occur without compensating support.

A different case arises when a diagnosis is beneficial in isolation but not part of the optimal feasible bundle under scarcity. If diagnosis A yields +10, diagnosis B yields +50, both use the same limited resource, and only one can be selected, A is not thereby intrinsically harmful. It is non-optimal under scarcity, dominated in allocation, or diagnostically inefficient.

If opportunity cost is built directly into the definition of welfare-negative diagnosis, almost every non-best choice becomes “negative,” making the category too broad.

welfare-negative diagnosis
= harmful even without replacement

non-optimal diagnosis
= possibly beneficial, but excluded from the best feasible bundle

Opportunity cost is represented by comparison among feasible bundles under a resource constraint. It should not be subtracted a second time after the bundles have been compared.

Bundle A generates W(A)
Bundle B generates W(B)
choose the higher-welfare feasible bundle

If B is forgone when A is chosen, B is the opportunity cost. Subtracting $W(B)$ again from $W(A)$ would double-count.

The general benchmark.

Let the actual attribution system be:

$$g_t^{act}:X_t\to A\cup\{\varnothing\}.$$

Its actual social welfare yield is:

$$Y_t^{act} = \int_{X_t} \delta W^{soc}(g_t^{act}(x),x;\Omega_t^{conv}) \,dF_t(x).$$

Let $\mathcal G$ be the general set of possible attribution systems. It is not restricted to current categories, thresholds, or professional practice. The optimal attributional welfare frontier is:

$$Y_t^* = \sup_{g\in\mathcal G} \int_{X_t} \delta W^{soc}(g(x),x;\Omega_t^{conv}) \,dF_t(x).$$

This asks what the best possible attribution system could generate under the same presentation distribution and conversion environment. It is a theoretical benchmark rather than a directly estimable empirical quantity.

Attributional efficiency can then be written:

$$E_t=\frac{Y_t^{act}}{Y_t^*}, \qquad Y_t^*>0.$$

The point of the benchmark is not to claim that welfare can already be measured perfectly. It is to keep explanatory uncertainty from defining the optimum. Limited evidence may explain why an attribution system fails; it should not make failure disappear by lowering the standard to whatever the system currently knows how to recognize.

The practical conclusions are compact:

need for help ↛ need for a diagnostic label
real distress ↛ one particular source attribution
recognized mechanism ↛ one fixed diagnostic category
more precision ↛ more welfare

A good diagnosis is a welfare-positive attribution of a real presentation to the right source, level, and pathway at the resolution appropriate to the available conversion environment.

related: Sulmasy discusses disease and natural kinds, while Gale describes diagnosis as problem-identification. Garrett offers a related account of how causal attributions shape psychiatric responses. Clarke and Kissane and de Figueiredo provide the terminology and phenomenology of demoralization and its distinction from depression used in the example.