Making Sense of a World Where Most Psychiatric Diagnoses Are False
And why isn’t clinical practice even more chaotic than it currently is?
In a post titled “Are Most Claimed Psychiatric Diagnoses False?,” (July 2025) psychiatrist @AffectiveMedicine (henceforth, “Dr. Affective”) made an unsettling argument modeled on John Ioannidis’s famous 2005 paper on why most published research findings are false. He applies the logic of positive predictive value (PPV) to psychiatric diagnosis and arrives at the conclusion that most psychiatric diagnoses given in real-world clinical settings are likely “false.” False here means that the diagnostic label given to the patient doesn’t actually satisfy the DSM/ICD diagnostic criteria for that diagnosis; if the DSM/ICD diagnostic criteria were properly and rigorously applied, the person would have a different diagnosis (or different diagnoses) or no formal diagnosis at all.
The thesis is straightforward and technically sound, as far as it goes. A diagnosis can be treated as analogous to a positive test result. Its accuracy depends on how common the disorder is in the population being evaluated (prevalence, or the pre-test probability), how good the evaluation is at catching the disorder when present (sensitivity), and how good it is at ruling out the disorder when absent (specificity). Even with respectable sensitivity and specificity (say, 85% each) the positive predictive value drops sharply for diagnoses that have low prevalence. For a condition present in 10% of the population being assessed, an 85/85 evaluation yields a PPV of only 39%. That is, most (>50%) of the positive diagnoses are wrong.
Aside from prevalence, there are many considerations working against the accuracy of diagnoses. Common psychiatric conditions such as depression, generalized anxiety, and ADHD have overlapping symptom profiles and fuzzy boundaries with “normality,” making them hard to distinguish even in rigorous evaluations. Most diagnoses aren’t made by psychiatrists but by primary care providers with less training and less time, lowering both sensitivity and specificity. Systematic biases such as clinician hobbyhorses, patient expectations, edge cases, and the pressure to produce a billable diagnosis shift thresholds toward overdiagnosis. And a multiple comparisons problem emerges as patients see multiple providers who each screen for multiple conditions, inflating the cumulative probability of at least one false positive. The more popular and sought-after a diagnosis is, the more likely it is to be false.
When I first read the post, my reaction was, “Damn!” I think Dr. Affective is right. As I’ve sat with the argument, I’ve come to think that the most important thing about it is that it forces us to ask further questions about what “false” means when applied to psychiatric diagnoses.
Why Aren’t Things More Catastrophic?
If psychiatric diagnoses are wrong more than half of the time, you’d expect the clinical psychiatric enterprise to be in freefall, with widespread chaos. We’d expect most people to receive inappropriate treatment and experience negative outcomes from that. And while psychiatric treatment certainly has its problems and while the radical critics of psychiatry already do believe that most people are being harmed, I genuinely don’t believe that the current state of affairs is as bad as it would be if the majority of diagnoses had no meaningful connection to what patients were experiencing.
The reason, I think, is that specific DSM/ICD diagnoses are doing far less therapeutic work than the system pretends. Most (first-line) psychiatric treatments target broad symptom domains, not specific diagnostic categories. SSRIs are effective for depressive symptomatology, multiple anxiety disorders, OCD, post-traumatic reactions, and (sometimes) eating disorders. CBT and psychodynamic principles apply across the internalizing spectrum and more. Mood stabilizers and antipsychotics are deployed transdiagnostically. If a clinician diagnoses major depression when a more careful evaluation would yield generalized anxiety, or when the “true” picture is an undifferentiated blend of depressive and anxious distress, the initial medication recommendation as well as the psychotherapy approach is likely to be the same.
Careful attention to DSM/ICD diagnoses does matter. I want to acknowledge that. Optimal treatment of OCD diverges from optimal treatment of generalized anxiety, for example. And this is where competent clinicians outperform subpar clinicians who cannot progress beyond a coarse-grained symptom characterization. And I’d say excellent clinicians are those who can go beyond DSM/ICD categories to richer forms of clinical characterization and case formulation. In this sense, both subpar and above-par clinical psychiatric practice involves a certain disregard for DSM/ICD.
Clinicians, in practice, operate at a level of description closer to symptom domains and prototypes. DSM/ICD diagnoses are most useful when a clinical presentation is an excellent fit with a diagnostic prototype… when someone has a “classic” or “textbook” presentation of schizophrenia, bipolar disorder, OCD, BPD, etc. It is very common, however, for clinical presentations to have a fuzzy and polymorphous character that doesn’t really fit well into DSM categories. The DSM diagnoses offer a shared vocabulary and many practical and administrative elements of the system run on them, but actual clinical reasoning proceeds on a different and more granular track.
The problem identified by Dr. Affective, if taken to its conclusion, points toward a mismatch between the categorical structure of our diagnostic system and the dimensional, transdiagnostic complexity of what patients present with in the clinic. The categories are too specific, pseudo-precise in a way, too discrete for the underlying phenomenology, and too fragmented for what the treatments can actually distinguish between. As Dr. Affective wryly comments, “Sometimes I think we should just call everything psychosis or neurosis and be done with it.”
If specific diagnoses aren’t driving treatment decisions, then the “falsity” of a given DSM/ICD diagnosis may be less consequential than one may initially assume. A “false” major depression diagnosis in someone with pervasive internalizing distress (say, with a “correct” DSM diagnosis being dysthymia and generalized anxiety) isn’t clinically catastrophic in the way a false cancer diagnosis would be, because the treatment offered isn’t contingent on the diagnostic specificity in the same way.
The risks of false diagnosis are concentrated along certain points: when misdiagnosis comes with the risk of treatment harms (e.g. confusing bipolar disorder with ADHD and putting patient on stimulant monotherapy inducing a manic episode), or missing something vital like catatonia or secondary medical etiology, or failure to update diagnosis beyond initial treatment (e.g. failing to recognize personality disorder after poor response to multiple medications) or when diagnoses carry prognostic weight (schizophrenia), or when they determine access to specific services or accommodations (autism), etc.
Diagnostic Metaphysics vs Diagnostic Math
The deeper issue is the metaphysics rather than the math of diagnostic accuracy. The PPV framework requires a determinate fact of the matter: either the patient “has” MDD or they don’t, and the diagnostic evaluation is trying to correctly determine which. While a person has MDD or not in a strict DSM operational sense, in reality, the person exists on a spectrum of approximate fit to the prototype of major depression (and that prototype itself breaks down into more statistically homogeneous dimensions). Any picture that presupposes that DSM/ICD categories correspond to discrete entities that patients either have or lack gets things wrong at a very fundamental level.
When we talk about the falsity of diagnoses, we can talk at the level of particular diagnoses (while assuming the classification system to be valid) or we can talk at the level of the classification system itself. We could mean that the patient doesn’t meet the DSM criteria for the assigned condition on careful assessment, but diagnoses can also be false in the sense that the entire exercise of forcing a dimensional, hierarchical picture into a discrete categorical scheme is misguided at the level of the classificatory framework. The first problem is addressed by more comprehensive and rigorous evaluations and diagnostic refinement/correction over time. The second problem suggests that we need to update our entire way of thinking about diagnosis to account for the fact that there may be no determinate fact of the matter for a categorical diagnosis to be true or false about.
Asking whether a diagnosis is “true” or “false” in the correspondence sense (does this label accurately map onto a discrete entity the patient has or lacks) may be the wrong question. A better question might be: is this diagnosis a good enough fit for this patient’s symptom profile and descriptive psychopathology, given our current categories, to usefully guide clinical decision-making?
Diagnostic Truth as Convergence
Another way we can think of diagnostic “truth” is in terms of convergence rather than correspondence. The relevant thought experiment goes something like this: if multiple competent clinicians, with comprehensive information available, including detailed developmental and psychiatric history, collateral information from family, longitudinal observation, treatment response data, and adequate time to conduct a thorough evaluation, with agreement on the diagnostic criteria being applied, and with opportunities to refine/update diagnoses, would they converge on the same diagnosis?
When convergence is high, we have something worth calling a “true” diagnosis because it represents a stable judgment under favorable epistemic conditions. Florid mania, contamination OCD, severe anorexia nervosa, melancholic depression, classic paranoid schizophrenia, these are cases where convergence would be high under favorable epistemic conditions and where calling a discordant diagnosis “false” seems appropriate. The clinical presentation is distinctive enough and the consequences substantial enough that competent evaluators would reliably agree.
When convergence is low even under epistemically favorable conditions, then the language of true and false diagnoses becomes misleading in a way. What we have is irreducible diagnostic uncertainty or non-specific psychopathology, reflecting not (or not primarily) the incompetence of the evaluators but the genuine ambiguity of the clinical picture, the overlap between the available categories, or the fuzziness of clinical boundaries.
Even under ideal convergence conditions, DSM/ICD diagnoses encompass heterogeneous symptom profiles and show high comorbidity. Even when convergence is high, even when all competent evaluators would agree that this patient “has MDD,” the diagnosis is still a lossy compression.
This is a thought experiment, in a sense, because in reality clinicians rarely have comprehensive information, adequate time, or opportunities for convergence. Real-world evaluations are typically brief, conducted under time pressure, influenced by biases, and performed by clinicians with varying levels of training. The PPV calculations accurately describe the de facto state of diagnostic practice. But the convergence thought experiment helps us distinguish between two very different sources of diagnostic error: errors that would be corrected by better information and more careful evaluation (and thus are genuinely “false” in a meaningful sense), and indeterminacies that would persist even under ideal epistemic conditions (and thus reflect limitations of the categories themselves rather than failures of the clinicians). When a diagnosis is uncertain, whether epistemically or metaphysically, saying so is both more honest and more clinically useful than conveying false diagnostic certainty.
See also:
Thanks for this thoughtful essay. One reaction (among many others, unspoken!) from my own experience: Some years ago, after a few months of back and forth with my psychiatrist, I convinced him (or at least, I convinced him to say—but I believe he was being genuine and honest) that being diagnosed with schizophrenia means nothing more than exhibiting some disjunctively specified set of experiences or behaviors. Once we agreed on that basic point, *both* of us began to think about treatment differently, and I'd say for the better. (For my part, it was really just a matter of now being willing to work with him.) In particular, he stopped worrying about treating some underlying 'disease' or 'condition' or whatever (called ‘schizophrenia’, or is it ‘really’ ‘schizoaffective disorder’ or..., or…, or does the name really matter?) and we focused on what could be achieved with regard to living well with those experiences and behaviors, or perhaps altering them where that’s helpful and possible.
In other words, if you take DSM/ICD for what they (mostly) are—disjunctive lists of ‘symptoms’ to which certain names are attached—then it is easy to stop worrying about the names and focus on the ‘symptoms’ (a misnomer in this case, but no further comment on that point here). And then an ‘incorrect’ diagnosis starts to matter a lot less because one is not asking “how do we ‘deal with’ this schizophrenia?” but “how do we cope with these experiences?” (regardless of what we might call them, taken collectively).
This is a great article, and great comment above. I almost always have one or more diagnoses that I’m using for insurance purposes and a completely different picture in my head. We are so concerned about being more specific and getting it right.
I wonder if it’s better to acknowledge the ambiguity of psychiatric diagnosis, if there’s value in having a diagnosis that just allows you a lot of wiggle room as a Dr or a patient. It’s counter to science and medicine, it might make us embarrassed. But I personally would sort of prefer to not be described exactly. I’d prefer to think that I can’t be captured by a
code.
We might instead say to patients, “Our diagnoses are quite overlapping and connected. I’m going to suggest this medication because I think it targets your main issue or targets the overlap.”