Psychiatric Classification

In brief

Two manuals govern how mental suffering gets named: the American Psychiatric Association's Diagnostic and Statistical Manual of Mental Disorders (DSM), now DSM-5-TR (2022), and the World Health Organization's International Classification of Diseases (ICD), now ICD-11 (in effect since 2022). Both sort symptoms into categories with explicit criteria: a minimum number of symptoms, a minimum duration, and a requirement that the pattern cause real distress or impairment. The system exists because diagnosis, before it, was notoriously unreliable, clinicians in different cities named the same patient differently. Since its 1980 overhaul, the categorical approach has become the field's working language, and it remains contested: how reliably it applies, how often it yields overlapping diagnoses in one person, whether it stigmatizes as much as it helps, and whether mental disorder even comes in discrete categories.

The full treatment

The problem it answers

Before 1980, American diagnosis leaned heavily on the clinician's theoretical training, especially psychoanalytic training, rather than a shared checklist. The drift showed up in hard data: the US-UK Diagnostic Project, led by John Cooper and published in 1972, found psychiatrists in New York diagnosing schizophrenia far more often than London psychiatrists shown the same videotaped patients, while London diagnosed manic-depressive illness more often for identical cases. The disagreement tracked training, not the patient. A shared nosology was needed that clinicians in different cities and schools could apply the same way.

How it works

The DSM's architecture, set in its third edition in 1980, uses operational criteria: listed symptoms, a minimum number required, a minimum duration, exclusion rules for cases better explained by another condition. This came from the "Feighner criteria," published in 1972 by John Feighner, Eli Robins, Samuel Guze and colleagues at Washington University in St. Louis, refined into the Research Diagnostic Criteria by Robert Spitzer, Jean Endicott, and Eli Robins in 1978. Spitzer chaired the task force that built DSM-III on this logic, stripping out psychoanalytic language so the manual described symptoms without committing to why they occurred. The ICD runs in parallel, historically looser, but ICD-11 converged toward this style while diverging on specifics, replacing DSM's separate personality disorder categories with one diagnosis rated by severity and dimensional traits.

What it claims

The manuals claim to be atheoretical about cause: two people sharing a diagnosis may have arrived by different routes, and the manual does not adjudicate which is correct. Its narrower claim is that a cluster of symptoms recurs consistently enough to be worth a shared name, for treatment, billing, legal proceedings, and research comparing patients across countries. DSM-III through DSM-IV also used a five-axis system separating the main disorder from personality, medical conditions, stressors, and functioning, meant to keep diagnosis from flattening a person into one label. DSM-5 dropped the axes in 2013 as artificial in practice.

The key study or demonstration

David Rosenhan's 1973 paper "On Being Sane in Insane Places," in Science, is the sharpest demonstration of what goes wrong. Eight sane confederates, the "pseudopatients," presented at psychiatric hospitals reporting one fabricated symptom, a voice saying a word like "empty" or "thud." Once admitted, they behaved normally and reported nothing further. All eight were hospitalized anyway, for an average of nineteen days and as long as fifty two, and all but one were discharged with "schizophrenia in remission," not "never ill." The deeper finding was stickiness: once staff had a label, they interpreted ordinary behavior through it, a confirmation bias in clinical form, one pseudopatient's note-taking was charted as "patient engages in writing behavior." A related problem shows in comorbidity data: Ronald Kessler and colleagues' 2005 analysis of the National Comorbidity Survey Replication found people meeting criteria for one DSM-IV disorder often meet criteria for another at once, raising an unresolved question, whether this reflects separate illnesses or one disturbance a checklist cuts into several named pieces.

Related distinctions

Reliability, whether two clinicians assign the same diagnosis, is not validity, whether the diagnosis corresponds to a real condition with a common cause or course. DSM-III's reform mainly fixed reliability and left validity untouched. Categorical models treat a disorder as present or absent; dimensional models treat symptoms as a matter of degree, closer to blood pressure than to pregnancy. Nosology as clinical convenience, a shared vocabulary, is a different claim from nosology as metaphysics, an assertion that a category names a joint nature itself carved. Screening for a rare disorder also runs into a base-rate problem: even an accurate instrument produces mostly false positives when true cases are scarce in the population tested.

Lineage

The direct ancestor is Emil Kraepelin (1856 to 1926), whose textbook Psychiatrie, revised across editions from the 1890s, organized illness by long-term course rather than symptom picture. His key move separated dementia praecox, later renamed schizophrenia by Eugen Bleuler in 1911, from manic-depressive insanity, marked by episodes and fuller recovery between them. Kraepelin's bet that disease entities would reveal themselves through natural history rather than theory is the root of DSM-III's descriptive method. The DSM's line runs from DSM-I (1952) and DSM-II (1968), psychoanalytically shaped, through DSM-III (1980) under Spitzer, DSM-III-R (1987), DSM-IV (1994) under Allen Frances, DSM-IV-TR (2000), DSM-5 (2013) under David Kupfer and Darrel Regier, to DSM-5-TR (2022). The WHO's line runs from ICD-6 (1948), the first edition with a mental disorders chapter, through ICD-9 and ICD-10 to ICD-11, in effect since 2022.

The strongest case for it

Before 1980, the field could not compare its own findings across cities: a schizophrenia diagnosis in New York and one in London were not reliably the same patients, so treatment studies and prevalence estimates built on those labels were not comparable either. A shared, explicit system, whatever its flaws, raised agreement enough that clinicians trained in different schools could discuss a case and mean the same thing, and that researchers anywhere could recruit comparable samples. None of this required perfect natural kinds, only categories repeatable enough that the field could build knowledge instead of restarting the argument with each new clinician. That was Spitzer's wager, and why the DSM and ICD converged on the same architecture.

The strongest case against it

The reliability the reform promised has not fully arrived. The DSM-5 field trials, reported by Darrel Regier and colleagues in 2013, found some major categories with weak test-retest agreement, a kappa of roughly 0.28 for major depressive disorder and roughly 0.20 for generalized anxiety disorder, figures the field's own conventions call questionable at best. Frances, who chaired DSM-IV, made this a public critique in Saving Normal (2013), arguing DSM-5 loosened thresholds, including removing the exclusion that had kept ordinary grief from qualifying as major depression, drifting toward diagnosing normal variation as illness.

A second attack targets validity, not reliability. Thomas Insel, then director of the National Institute of Mental Health, announced in 2013 that the institute would redirect funding away from DSM categories toward Research Domain Criteria, since diagnoses built from consensus lack the biological markers other branches of medicine require. Roman Kotov and a large consortium made a related, statistical case in 2017 with the Hierarchical Taxonomy of Psychopathology: factor analysis of symptom co-occurrence found psychopathology loading onto continuous dimensions, internalizing, externalizing, thought disorder, rather than discrete boxes, exactly what Kessler's comorbidity finding would predict.

A third attack concerns what naming does to people. Homosexuality was listed as a mental disorder in DSM-II until the APA's Board of Trustees voted to remove it in December 1973, shaped by activism including a 1972 conference appearance by psychiatrist John Fryer, speaking in disguise as "Dr. Henry Anonymous" because he could not safely reveal himself as gay. The episode is widely cited as evidence that a category can encode the prejudice of an era as much as any biological fact. Thomas Szasz argued in The Myth of Mental Illness (1961) that "mental illness" is a metaphor for problems in living, so a nosology of mental disorder risks medicalizing moral conflicts.

Where it stands now

DSM-5-TR and ICD-11 remain the two systems nearly every clinician, insurer, and researcher uses, and neither abandoned categories for pure dimensions. Both hedged instead. DSM-5 placed a dimensional Alternative Model for Personality Disorders in Section III, for research and complex cases, while keeping the older categorical types in Section II for routine use. ICD-11 went further, replacing categorical personality disorder types outright with one diagnosis rated by severity and dimensional traits, a real divergence between the systems. Research Domain Criteria reshaped American funding without displacing DSM or ICD codes at the bedside, since clinicians, courts, and insurers still need a category to write down. The field's working consensus, when practitioners are asked directly, is that the categories are useful approximations, not lines nature drew in advance.

Test yourself

Think of a diagnostic label, psychiatric or otherwise, that you or someone close to you has carried. Ask what it actually changed. Did it open something, a treatment, a vocabulary for an experience that had none before? Did it also close something, a lowered expectation, a decision made about the person before they could speak for themselves? Most labels do both. Notice which one you remember first, and ask why.