Where diagnoses come from

If you have a diagnosis, you probably want to know what it is a diagnosis of. That turns out to be a harder question than it sounds, and the most useful answers come not from critics of psychiatry but from the people who ran it.

The short version, and the register we will hold to for the whole piece: we treat what we can describe, not what we understand.

This is not medical advice, and nothing here is a reason to doubt a diagnosis or stop a treatment. A description can be extremely useful without being a complete explanation. If you are in crisis, call or text 988 (US), 24/7, free.

The NIMH director's own account

In April 2013, weeks before DSM-5 was published, the sitting director of the National Institute of Mental Health, Thomas Insel, wrote this [1]:

"While DSM has been described as a 'Bible' for the field, it is, at best, a dictionary, creating a set of labels and defining each. The strength of each of the editions of DSM has been 'reliability' — each edition has ensured that clinicians use the same terms in the same ways. The weakness is its lack of validity. Unlike our definitions of ischemic heart disease, lymphoma, or AIDS, the DSM diagnoses are based on a consensus about clusters of clinical symptoms, not any objective laboratory measure. In the rest of medicine, this would be equivalent to creating diagnostic systems based on the nature of chest pain or the quality of fever."

That is not an anti-psychiatry pamphlet. That is the head of the agency that funds most American mental-health research, describing the manual his field uses.

And the same post contains the sentence that keeps the criticism honest, which almost nobody quotes alongside it [1]:

"It became immediately clear that we cannot design a system based on biomarkers or cognitive performance because we lack the data."

So the problem is not that someone chose description over explanation out of laziness. The explanation is not available yet. Insel's answer was RDoC — a research framework meant to build, eventually, the data a different kind of diagnosis would need.

A note on where we found that. The post is no longer on nimh.nih.gov; that URL returns 404. We read it, and quote it, from an archived capture on the Internet Archive dated 27 June 2021 [1]. This is the second time in this series we have gone to a private archive for a primary US government document — the FDA's pre-2020 safety communications are the other. Worth noticing as a pattern.

Reliability and validity, and why the difference matters to you

Insel's two words carry the whole argument.

Reliability is agreement. If four clinicians assess the same person, do they reach the same label? DSM's operational criteria were built to make that happen, and by and large they did.

Validity is whether the label corresponds to a real, distinct thing in the world — something with a mechanism, a course, a test.

You can have the first without the second. Four astronomers can reliably agree on which constellation a star belongs to; the constellation is still a pattern drawn from Earth. That is a real gain — agreement lets research accumulate and lets a treatment plan mean the same thing in two cities — and it is not the same as having found a natural boundary.

The practical consequence for a patient is small but real. Your diagnosis is a well-defined, agreed description of a pattern. It is a good enough basis to choose a treatment and track whether it helps. It is not a statement that a specific thing has been found inside you.

How the manual actually gets revised

The DSM is a book, produced by a professional association, revised by committees, and published on a schedule. A peer-reviewed history of psychiatric nosology puts the timeline this way [2]: the series began with Medical 203 in 1945, ran through DSM-I in 1952 and DSM-II in 1968 under psychoanalytic influence, and then DSM-III in 1980 represented what the author calls "a massive 'turning of the page' in nosology." DSM-IV was published in 1994, and DSM-5 in 2013 [2].

DSM-III is the hinge. It is the point at which diagnosis became a checklist of observable criteria rather than an inference about underlying dynamics — which is precisely the move that bought reliability.

One thing we are deliberately not telling you. You will often see a figure for how many diagnoses each edition contains, usually a rise from about 106 to about 300. We went looking for a citable primary source for those counts and did not find one; the numbers vary depending on whether subtypes, "other specified" residual categories, and provisional conditions are counted. It is the most quotable statistic in this subject and the easiest to refute, so we are leaving it out rather than repeating it.

What changed in DSM-5, in the APA's own words

For ADHD, here is the American Psychiatric Association's own summary of what it did, verbatim [3]:

"The diagnostic criteria for attention-deficit/hyperactivity disorder (ADHD) in DSM-5 are similar to those in DSM-IV. The same 18 symptoms are used as in DSM-IV, and continue to be divided into two symptom domains (inattention and hyperactivity/impulsivity), of which at least six symptoms in one domain are required for diagnosis. However, several changes have been made in DSM-5: 1) examples have been added to the criterion items to facilitate application across the life span; 2) the cross-situational requirement has been strengthened to 'several' symptoms in each setting; 3) the onset criterion has been changed from 'symptoms that caused impairment were present before age 7 years' to 'several inattentive or hyperactive-impulsive symptoms were present prior to age 12'; 4) subtypes have been replaced with presentation specifiers that map directly to the prior subtypes; 5) a comorbid diagnosis with autism spectrum disorder is now allowed; and 6) a symptom threshold change has been made for adults... with the cutoff for ADHD of five symptoms, instead of six required for younger persons."

Note what the APA says first: the criteria "are similar," and "the same 18 symptoms are used." The changes sit at the margins of eligibility — onset age 7 to 12, adult threshold 6 to 5, autism comorbidity now permitted — and one of them (the strengthened cross-situational requirement) pushes the other way. "DSM-5 loosened the ADHD definition" is a claim the APA's own document does not support as flatly as it is usually made.

For anxiety, DSM-5 did something structural rather than criterial [3]:

"The DSM-5 chapter on anxiety disorder no longer includes obsessive-compulsive disorder (which is included with the obsessive-compulsive and related disorders) or posttraumatic stress disorder and acute stress disorder (which is included with the trauma- and stressor-related disorders)."

Separation anxiety disorder and selective mutism moved into the anxiety chapter at the same time [3]. So the box did not simply shrink — two things left and two arrived. Any "any anxiety disorder" statistic that spans 2013 is counting different diagnoses on either side of that line, with no change in anyone's symptoms.

And in a detail that complicates the lazy reading of all this, the APA describes one of its own changes as intended to reduce diagnoses [3]:

"the 6-month duration, which was limited to individuals under age 18 in DSM-IV, is now extended to all ages. This change is intended to minimize overdiagnosis of transient fears."

The step nobody sees: a definition only counts if a survey implements it

Here is the finding that changed how we built this whole series, and it comes from a federal statistical agency reasoning about its own data.

The obvious story is: the DSM changes a criterion, so the counts change. SAMHSA's technical report on exactly this question shows the story needs an extra step. On the removal of the bereavement exclusion from major depression — the single most discussed DSM-5 depression change — SAMHSA writes [4]:

"NSDUH already does not apply the bereavement or mixed-episode criteria."

"Although the diagnostic changes from DSM-IV to DSM-5 regarding mixed episodes and the bereavement exclusion will have no effect on NSDUH estimates of MDE, the addition of the term 'hopeless' as a subjective descriptor of low mood (Criterion A1) may have an effect on these estimates."

The criteria changed. The national survey's depression numbers could not move, because the survey's instrument never implemented the criterion that changed.

So "the definition changed" and "the counts changed" are joined by an implementation step that is usually invisible and sometimes simply absent. A DSM change reaches a number only if the thing doing the counting was using that criterion. Where the count comes from clinicians — as with "has a doctor ever told you your child has ADHD" — there is a real path. Where the count comes from a survey algorithm, there may be none.

And when you look, the DSM change often left no mark

The best public data on any of this is childhood ADHD: an unbroken annual series from the National Health Interview Survey, parent-reported ever-diagnosed, ages 3–17, running 1997 to 2018 [8].

DSM-5 was published in 2013. Here is what the series did around it [8]:

2012: 9.5% → 2013: 8.8% → 2014: 8.9% → 2015: 9.8% → 2016: 9.4% → 2017: 9.4% → 2018: 9.8%

There is no step. 2013 went down, and the whole 2012–2018 stretch wanders inside about a point. Whatever DSM-5 did to childhood ADHD diagnosis, it is not visible as a discontinuity in the longest clean public series we have — and no honest annotation can claim otherwise.

Meanwhile the clearly visible steps in that data sit at survey redesigns, where the agencies themselves say not to compare across. We built the full chart, with every break annotated and every agency warning quoted, at how childhood ADHD got counted.

Allen Frances, who chaired the DSM-IV task force, went on to become one of the most prominent critics of diagnostic expansion, publishing on it in the general medical literature — including a 2013 piece in Annals of Internal Medicine titled "The new crisis of confidence in psychiatric diagnosis" [5]. We cite it by title and venue rather than quoting it, because the full text sits behind a paywall we could not read from a primary source, and we would rather say that than paraphrase something we have not verified.

Where the drugs came from, which is its own answer

If diagnoses were derived from mechanisms, you would expect the drugs to have been designed against those mechanisms. Mostly, it happened the other way around.

A 2022 analysis in the World Journal of Psychiatry applied formal criteria for serendipity to the discovery of the classical antidepressants. Its finding on the two founding drugs [6]: imipramine was developed as an antipsychotic and iproniazid as an anti-tuberculosis agent, and both were found to have antidepressant properties through unexpected observation. The paper classifies both as a mixed pattern in which "initial serendipitous discoveries... led secondarily to non-serendipitous discoveries" — later drugs in each family were designed as antidepressants, by modifying the chemistry of the accident [6].

Chlorpromazine ran the same way round. A history of the drug records that it was synthesized in December 1951 and became available on prescription in France in November 1952 — and that the demonstration that it blocks dopamine receptors came afterwards, once the recognition of chemical transmission at the synapse and the necessary instruments and receptor assays existed [7].

Lithium is the oldest example still in daily use: John Cade's 1949 report in the Medical Journal of Australia, "Lithium salts in the treatment of psychotic excitement," reached the modern literature partly through later reprints, and its mechanism is still not settled.

The pattern is mechanism-agnostic discovery, with the mechanism story built backwards. A drug was observed to help. The receptor it acted on was identified later. And the identified receptor was then read back into the disorder as its cause — which is how "this drug affects serotonin" became "the illness is a serotonin problem." We took that particular inference apart at the serotonin story.

None of that means the drugs do not work. It means the explanation arrived after the observation and was shaped to fit it, which is a normal thing in the history of medicine and an abnormal thing to present to patients as settled biology.

The manual is still moving

DSM-5-TR was published in 2022. The American Psychiatric Association now runs a standing proposal-and-comment process, issuing criteria and text updates between editions — most recently in September 2025, alongside ICD-10-CM code changes.

That means "which edition are we on" is no longer a complete answer to "what are the current criteria." Any statistic you read about a diagnosis carries an implicit date, and the criteria behind it may have moved since.

What this does and does not mean

• It does not mean your diagnosis is fake. It means it is a description of a pattern, agreed on carefully, that predicts useful things about treatment.
• It does not mean diagnosis is useless. Reliability is a real achievement. Without it, no one could study anything.
• It does not mean the categories are arbitrary. They came from clinical observation over more than a century, and they are revised — sometimes toward narrowing, as the APA's own 6-month duration change shows [3].
• It is not a reason to stop treatment. Nothing on this page speaks to whether a treatment is working for you. If you want to change something, that belongs in a conversation with your prescriber, with a plan — and stopping psychiatric medication abruptly can be dangerous. See stopping antidepressants: withdrawal and relapse.

What to ask your prescriber

• What specifically led you to this diagnosis, and what would have pointed somewhere else?
• Which criteria did I meet, and were any of them borderline?
• Does the diagnosis change what we would try, or is it mainly a label for what we are already treating?
• If treatment does not work, does that make you reconsider the diagnosis?
• Would you expect this diagnosis to still apply in five years?

The fourth is the one clinicians tend to appreciate most, because it is the question they ask themselves.

Sources

1. Insel T. Transforming Diagnosis. NIMH Director's Blog, 29 April 2013. The original URL (`nimh.nih.gov/about/directors/thomas-insel/blog/2013/transforming-diagnosis`) returns HTTP 404 as of 2026-08-25; quoted from the Internet Archive capture of 27 June 2021. archived copy — verified 2026-08-25.
2. Shorter E. The history of nosology and the rise of the Diagnostic and Statistical Manual of Mental Disorders. Dialogues in Clinical Neuroscience, 2015;17(1):59–67. PMID 25987864, PMC4421901. Open access. Source for the edition years: Medical 203 (1945), DSM-I (1952), DSM-II (1968), DSM-III (1980), DSM-IV (1994), DSM-5 (2013). doi:10.31887/DCNS.2015.17.1/eshorter — accessed 2026-08-25.
3. American Psychiatric Association. Highlights of Changes from DSM-IV-TR to DSM-5. 19 pp. ADHD passage p. 2; anxiety chapter passages p. 5. psychiatry.org PDF — retrieved 2026-08-25.
4. Substance Abuse and Mental Health Services Administration. Impact of the DSM-IV to DSM-5 Changes on the National Survey on Drug Use and Health, 2016, §3 Mental Illness. NCBI Bookshelf `mrdsm52016`. ncbi.nlm.nih.gov — accessed 2026-08-25.
5. Frances A. The new crisis of confidence in psychiatric diagnosis. Annals of Internal Medicine, 2013;159(3):221–222. PMID 23685989. Cited by title and venue; full text not open access and not quoted here. doi:10.7326/0003-4819-159-3-201308060-00655 — citation verified via PubMed 2026-08-25.
6. López-Muñoz F, et al. Role of serendipity in the discovery of classical antidepressant drugs: applying operational criteria and patterns of discovery. World Journal of Psychiatry, 2022;12(4):588–602. PMID 35582332. Open access. doi:10.5498/wjp.v12.i4.588 — accessed 2026-08-25.
7. Ban TA. Fifty years chlorpromazine: a historical perspective. Neuropsychiatric Disease and Treatment, 2007;3(4):495–500. PMID 19300578. Source for the December 1951 synthesis and November 1952 French prescription availability, and for dopamine-receptor blockade being demonstrated subsequently. pubmed.ncbi.nlm.nih.gov/19300578 — accessed 2026-08-25.
8. Centers for Disease Control and Prevention. Trends in ADHD Among U.S. Children. Chart data file `ADHD-diagnosis.json` (published 2024-05-15, updated 2026-07-07), NHIS 1997–2018 segment, parent-reported ever-diagnosed ADHD, ages 3–17. cdc.gov — retrieved 2026-08-25 via Internet Archive (`www.cdc.gov` returns 403 to non-browser clients).

Related reading

• How childhood ADHD got counted — the full chart, every break annotated
• The serotonin story: what the evidence said, and what patients were told
• What FDA approval actually means — and what it doesn't
• Why your doctor gets paid for procedures, not outcomes