The DSM still rules the clinic, but psychiatry's leading research frameworks now treat its categories as useful fictions rather than natural kinds of illness.
In April 2013, a few weeks before the American Psychiatric Association published the fifth edition of its diagnostic manual, the head of one of the largest public funders of mental-health research in the world explained why he would stop treating it as the standard for science. Thomas Insel, director of the U.S. National Institute of Mental Health, wrote that the manual’s categories rested on agreement about clusters of symptoms rather than on any measurable biological fact. “The strength of each of the editions of DSM has been ‘reliability’,” he wrote; “the weakness is its lack of validity.” Clinicians using the manual agree with one another about what to call a given patient, but whether those labels correspond to real, distinct diseases is a separate question, and Insel’s answer was that DSM categories had not shown the kind of validity expected of medical diagnoses. NIMH, he announced, would begin reorienting its research away from DSM categories.
The statement carried weight because it came from inside the field’s own research establishment, from the person who directed how a large share of federal mental-health research money was spent, and it said plainly that the book governing everyday diagnosis was a poor foundation for science. NIMH said it would favor studies that looked across or subdivided the manual’s categories over studies that enrolled patients strictly by DSM criteria. In the decade since, two large research programs have grown up around that judgment, both built on the same premise: the manual’s disorders are convenient administrative units, not natural kinds. A natural kind is a category whose boundaries come from the world rather than from a committee, the way “gold” or “influenza” picks out something with a stable underlying nature. Robert Kendell and Assen Jablensky had drawn the relevant distinction sharply a decade earlier, between the validity of a diagnosis, which asks whether it names a real thing, and its utility, which asks whether it is useful to apply. A label can score high on one and low on the other.
To see why the natural-kinds assumption came under strain, it helps to be clear about what the Diagnostic and Statistical Manual of Mental Disorders does and where it came from. The DSM is categorical: for each disorder it lists criteria, and a person either meets the threshold or does not. In simplified form, five of nine symptoms over two weeks, including depressed mood or a loss of interest, together with real distress or impairment, and the diagnosis is major depressive disorder; fall short of the count, and it does not apply. Those yes-or-no labels then run through the whole apparatus of care, determining which code a clinician submits, which treatments an insurer reimburses, which patients a drug trial can enroll, and which diagnosis a chart carries for years. This structure was consolidated in 1980, when DSM-III, led by the psychiatrist Robert Spitzer, replaced psychoanalytic theories of illness with explicit symptom checklists. The change solved a genuine problem: before it, two psychiatrists examining the same patient often reached different diagnoses, and operational criteria gave them a far better chance of agreeing. Those criteria made trials, epidemiology, and standardized records far easier to produce. Built into the categories, though, was a stronger claim, one that reached back to Emil Kraepelin’s separation of dementia praecox from manic-depressive illness by their different courses: that these labels name discrete conditions, each with its own boundary and, eventually, its own biology. The working hope was that reliability would mature into validity.
Several features of the data sit awkwardly against that hope. Comorbidity is the first. In the National Comorbidity Survey Replication, a large U.S. epidemiological study, 45 percent of people who met criteria for any disorder in a given year also met criteria for at least one more in the same year, with nearly a quarter meeting criteria for three or more. If these labels marked independent disease processes, this degree of systematic overlap would be hard to explain; instead it is the norm, which suggests the categories are lines drawn across continuous territory, and the overlap is what you would expect if the conditions are not truly separate.
Within-category heterogeneity creates the same problem inside a single diagnosis. Because most DSM diagnoses require some number of symptoms from a longer menu, two people can receive the identical label while sharing almost none of the same problems. In the STAR*D depression study, Eiko Fried and Randolph Nesse found 1,030 distinct symptom profiles among 3,703 depressed outpatients, and the single most common profile fit only 1.8 percent of them. One patient may be sleepless, agitated, and unable to eat, another sleeping constantly, slowed, and overeating, both carrying the same diagnosis and, in principle, the same treatment. Post-traumatic stress disorder is more extreme still: the DSM-5 criteria admit 636,120 distinct symptom combinations that all qualify. A category holding that much internal variety is hard to treat as one thing with one cause.
Field trials also showed uneven reliability for some common diagnoses. When the DSM-5 was tested before publication, several of its most familiar diagnoses agreed with themselves only modestly on repeat assessment. Major depressive disorder reached a kappa around 0.28 and generalized anxiety disorder around 0.20, values that fall in the range clinicians themselves call questionable. The reliability Insel praised turned out to be uneven for exactly the disorders most people receive.
The biological evidence has not produced individual-level tests that map cleanly onto DSM labels. Decades of imaging, genetics, and blood work have produced real findings, but none that reliably diagnose a single person and line up with a category. Large genomic analyses keep finding that the same inherited risk variants raise the odds of several disorders at once; a study of autism, ADHD, bipolar disorder, major depression, and schizophrenia found substantial shared genetic risk across all five, so the liability spreads across diagnostic boundaries instead of tracking one category. There is no scan or assay that confirms depression the way a biopsy confirms a lymphoma, which is the comparison Insel drew when he pointed to ischemic heart disease, lymphoma, and AIDS as diagnoses anchored in something a laboratory can measure.
The Research Domain Criteria framework, or RDoC, is NIMH’s answer to this pattern. Instead of asking which disorder a person has, it asks how well-defined functional systems are working: how the brain responds to threat, how it processes reward, how attention and memory operate. RDoC now spans six such domains — negative valence, positive valence, cognitive systems, social processes, arousal and regulation, and sensorimotor function — each studied dimensionally from genes and circuits up to behavior, and deliberately across diagnostic lines. A study built this way might recruit everyone in a mood clinic and examine the circuitry of anhedonia, the loss of pleasure, wherever it appears, without restricting itself to patients who clear the depression threshold. The wager is that the mechanisms of mental illness follow the brain’s functional systems, which need not line up with the manual’s categories, and that a durable classification has to be built by measuring those mechanisms first. Insel was explicit about the limits: RDoC, he wrote, is a research framework, not a clinical tool.
HiTOP, the Hierarchical Taxonomy of Psychopathology, comes at the same problem from the statistical and clinical side. It starts from how symptoms actually cluster together in large samples and builds upward: individual symptoms combine into narrow components and syndromes, and those into broad spectra such as internalizing, which covers the anxious-depressive range, externalizing, which covers impulsive and disinhibited problems, and thought disorder. Everything in it is a matter of degree, a position on a continuum that a person occupies to some extent. Comorbidity becomes less puzzling here, because related disorders sit near one another under the same higher-order spectrum. Supporting the picture, several groups have found that a single general dimension, sometimes called the p factor by analogy with the general factor in intelligence research, captures a shared liability running through internalizing, externalizing, and thought disorders alike. If psychopathology grades continuously and shares a common core, the boundaries between DSM categories begin to look like thresholds imposed on a gradient.
“Convenient fiction” is a precise description of what the categories have become. A convenient fiction is a category everyone knows is not literally carved out by nature but keeps using because it does real work. Money, legal adulthood, and national borders all rest on convention, and all organize behavior and carry serious consequences. DSM categories work the same way. A diagnosis of major depressive disorder tells a colleague roughly what is going on, unlocks a reimbursable course of treatment, and gives a patient a name for their experience. For those clinical and administrative uses, shared and stable criteria are enough; the category does not have to be a natural kind. What the research frameworks have done is pull apart two roles the manual used to fuse. The categories can keep organizing communication and payment while the science stops treating them as accounts of how illness is actually built. The risk in keeping the labels is that people mistake a description for an explanation: saying a patient cannot work because they have major depression is administratively necessary, but the label only gathers the symptoms under a heading and leaves untouched the harder question of why that person’s fatigue, insomnia, and hopelessness took the form they did.
The clinic cannot simply switch, for two reasons. Dimensional systems are more faithful to the data and much harder to use at the bedside. A clinician needs a decision — admit or discharge, prescribe or wait, code the visit so it can be billed — and a decision is easier to hang on a single label than on a profile of scores across a dozen continua. Insurance, disability law, and prescribing guidelines are all written in categorical language, so the manual persists in part because the institutions around it are built from its terms. Dimensional models also carry their own hazards. They are harder to teach and to communicate quickly, and a system that scores everyone on many axes creates new opportunities to classify ordinary variation in mood, attention, or fear as abnormal along some dimension. Better measurement does not by itself produce more restraint.
The categorical habit also sharpens the long-running worry about overdiagnosis. When a disorder is a yes-or-no box, widening the box quietly reclassifies more people as ill, and a category with no natural boundary to point to has weak defenses against that drift, especially for mild, mixed, or transient problems. Critics such as Allen Frances have argued that successive expansions of diagnostic criteria have pulled ordinary distress into the clinic, and the broader campaign against overdiagnosis in medicine makes the same case about acting on labels applied too readily.
More than a decade after the 2013 announcement, the promised replacement has not arrived. RDoC has generated a large body of research without producing a diagnostic system a clinician can use in an ordinary appointment. In a later interview, Insel was reported as saying that the neuroscience program he led had not measurably reduced suicide or hospitalization at the scale that matters. The DSM, now in its text-revised fifth edition, remains the working manual of psychiatry, as categorical as ever. The important change is quieter: the field’s leading research institutions now organize their work around the premise that the manual’s disorders are not the real units of mental illness. They keep the labels for communication, billing, and treatment access, and look elsewhere for the mechanisms that produce symptoms. Psychiatry has settled into a working arrangement in which the book that defines its diagnoses and the science that studies them no longer make the same claim about what a mental disorder is.