Imagine practicing oncology without pathology. No biopsy, no receptor status, no molecular profiling, no circulating tumor DNA, no tissue specimen that can be reexamined later. Instead, the physician asks a series of questions, observes the patient, determines whether enough findings have persisted long enough, and assigns a diagnosis. Modern oncology would consider such a system radically incomplete. Something resembling it remains ordinary practice across much of psychiatry.
This is not a claim that psychiatric illness is imaginary. The brain is biological, severe mental illness produces profound disability, and genetics and neuroscience have demonstrated substantial biological contributions to psychiatric disease. Medications, electroconvulsive therapy, and psychotherapy produce real clinical benefit. The harder question is what kind of scientific knowledge a psychiatric diagnosis actually represents. Psychiatry is a hybrid science, and nowhere is that clearer than when the symptoms are also the test.
In much of medicine, symptoms initiate an investigation that moves toward an independently measurable biological process. In psychiatry, the patient’s experiences and behaviors often remain both the phenomena requiring explanation and much of the evidence establishing the diagnosis. The symptoms do not merely point toward the diagnosis; in important respects, they constitute it. In many areas of medicine, the clinician interprets a measurement obtained independently of that interpretation. In psychiatry, the interpretation often helps produce the evidence.
This does not apply uniformly across the field. Some neuropsychiatric presentations can be traced to identifiable medical causes, and advances in dementia, genetics, immunology, and neurology continue to create stronger biological anchors. But for many of psychiatry’s most common diagnoses, there is no validated laboratory, imaging, or molecular test that independently establishes the diagnosis in an individual patient.
Real physician voices, twice a week
Free, and one click to unsubscribe.
None of this is news inside the field. It is psychiatry’s own long-running self-critique, argued for decades by figures at the center of the discipline. What has not always caught up is the language in which diagnoses reach patients, courts, disability adjudicators, and the public, where a useful clinical construct can gradually be spoken of as though it were an independently demonstrated biological entity.
The Lindsay Clancy trial, ongoing as of this writing, has again placed postpartum mental illness, psychosis, and criminal responsibility in the national conversation. Whatever conclusions belong to that particular case, the broader scientific question extends far beyond it: What can psychiatry measure, what must it infer, and how confidently should those different forms of knowledge be communicated?
A science that must explain and understand
German philosopher Wilhelm Dilthey distinguished erklären, the causal explanation of physical processes, from verstehen, the interpretive understanding of human experience and meaning. Psychiatrist and philosopher Karl Jaspers later applied this distinction to psychiatry. Psychiatry is required to do both.
A malignant cell can be characterized through its appearance, molecular pathways, and genetic alterations. A delusion also arises from the brain, but understanding it requires more than knowing what is happening biologically. The clinician must determine what the person believes, how strongly the belief is held, whether it makes sense within the person’s culture, and how it affects behavior. Reducing such experiences entirely to biology leaves out clinically important information, while treating them purely as personal narratives ignores the biological organ producing them. Psychiatry therefore occupies a genuinely hybrid position. Treating it as purely a natural science overstates the certainty of its current diagnostic categories. Treating it as merely social construction ignores substantial evidence from genetics, neuroscience, pharmacology, development, and treatment response.
Reliability is not validity
Modern psychiatric classification emerged partly from a problem of reliability. Before DSM-III introduced more standardized diagnostic criteria in 1980, clinicians could disagree substantially about the diagnosis of the same patient. Standardized criteria improved communication and made diagnosis more reproducible. That was a real advance, but reliability and validity are not interchangeable.
Define “tall” as 6 feet or greater and observers can agree almost perfectly about who qualifies. The classification is highly reliable. Nature has not thereby created two biologically distinct populations separated at that threshold. The boundary is ours.
Psychiatric classification faces a more consequential version of the same problem. Clinicians can become increasingly consistent in identifying major depressive disorder without establishing that everyone crossing its threshold shares one underlying disease. In 1970, psychiatrists proposed several ways to evaluate whether psychiatric diagnoses represented valid disease categories: clinical features, laboratory findings, separation from other disorders, what happened to patients over time, and patterns within families. Psychiatry has since accumulated enormous evidence across these areas, yet clear biological separation remains elusive for many major disorders.
Many psychiatric diagnoses are therefore best understood, at least at present, as clinical constructs. They are categories created to organize recurring patterns of symptoms, illness course, and impairment. The unresolved question is whether a given construct corresponds to a distinct disease that exists in nature rather than primarily being a useful way for humans to classify similar presentations. A psychiatrist and neuroscientist has written specifically about the danger of reifying DSM categories. Reification occurs when a useful diagnostic label gradually begins to be treated as though the label itself proves that a distinct biological disease has been discovered. The diagnostic manuals themselves have often been more cautious about what their categories represent than the language surrounding those categories in medicine and society.
When the phenotype becomes the diagnosis
Chest pain leads to an electrocardiogram, or ECG, a troponin blood test, and potentially coronary angiography. Polyuria, or excessive urination, leads to glucose measurement. Unexplained weight loss can lead to imaging and tissue pathology. The symptom begins the investigation; it generally does not constitute the disease.
Major depressive disorder is different. It is identified through combinations of depressed mood, loss of interest or pleasure, changes in sleep and appetite, changes in movement or activity, fatigue, guilt or worthlessness, difficulty concentrating, suicidal thoughts, duration, and functional impairment. No laboratory value, brain scan, tissue specimen, or molecular test independently confirms it in an individual patient. The patient’s experiences are therefore simultaneously the symptoms being explained and much of the evidence used to establish the diagnosis.
This raises a fundamental question about what the diagnosis actually represents. Is major depressive disorder one biological disease, several diseases producing overlapping symptoms, a syndrome that can result from multiple causes, or a dimension of human functioning that becomes pathological beyond a certain point? Psychiatry does not yet have definitive answers.
Anemia offers a useful comparison. Anemia is unquestionably real, but it is generally not the end of medical reasoning. Iron deficiency, bleeding, hemolysis, kidney disease, bone marrow failure, cancer, and inherited disorders can all produce it. Medicine attempts to move from the observable abnormality toward the mechanism causing it.
Psychiatry may still be earlier in that process for some of its major diagnoses. Two people can meet criteria for major depressive disorder while sharing surprisingly few symptoms. One may sleep excessively while another cannot sleep. One may overeat while another loses weight. One may be slowed while another is agitated. Both may appropriately receive the same diagnosis, but whether they have the same underlying biological disease is a separate question.
Perhaps major depression will prove to be a coherent biological entity. Perhaps it will eventually be divided into disorders defined by different molecular, circuit, developmental, inflammatory, hormonal, or environmental mechanisms. Or perhaps it will settle into something more like anemia, an important syndrome identifying a clinically meaningful state without necessarily identifying its cause. The diagnosis may not be wrong. It may be unfinished.
Biology is not respecting the boundaries
Large genetic studies have repeatedly found substantial overlap among psychiatric diagnoses. Schizophrenia and bipolar disorder share significant genetic risk, while depression, anxiety, and post-traumatic stress disorder also overlap. This does not mean these disorders are identical. It does suggest that their biological boundaries are less distinct than their diagnostic names can make them appear.
The then-director of the National Institute of Mental Health (NIMH) highlighted this problem shortly before publication of DSM-5. In an April 2013 post titled “Transforming Diagnosis,” he argued that the manual had succeeded in improving reliability but continued to struggle with validity because psychiatric diagnoses rested on agreed-upon clusters of symptoms rather than objective laboratory measures. The deeper point was not that psychiatric diagnoses are useless. It was that nature has no obligation to organize mental illness according to categories humans developed from clinical observation.
The NIMH Research Domain Criteria (RDoC) initiative grew partly from that recognition. Instead of starting only with diagnoses such as schizophrenia or major depression, RDoC allows researchers to study areas of functioning across multiple levels, including genes, molecules, cells and circuits, physiology, behavior, and patients’ own reports of their experiences. Its limitations should be stated alongside its promise. NIMH itself describes RDoC as a research framework intended to improve future understanding and classification, not a diagnostic guide or a replacement for existing diagnostic systems. More than a decade later, it has not produced one. RDoC reframes the question well. It has not yet answered it.
When mechanism changes the diagnosis
Anti-N-methyl-D-aspartate (NMDA) receptor encephalitis illustrates why mechanism matters. It is an autoimmune disease in which the immune system produces antibodies against an important receptor in the brain. In a substantial share of affected young women, the disease is associated with an ovarian teratoma, a usually benign tumor that can contain neural tissue expressing the same receptor targeted by the immune response.
The illness can unfold through stages. A nonspecific flu-like illness may be followed by a psychiatric phase involving hallucinations, delusions, agitation, disorganized behavior, or catatonia, which can resemble a primary psychiatric illness. Some patients first enter psychiatric care. Neurological features can then emerge, including seizures, abnormal movements, and instability of functions such as heart rate and blood pressure. Detection of anti-NMDA receptor antibodies in cerebrospinal fluid is a key part of confirming the diagnosis, and many patients improve with immune treatment and, when a tumor is present, its removal.
Once the autoimmune mechanism is identified, medicine does not describe the patient as having a biologically confirmed subtype of schizophrenia. The diagnosis becomes autoimmune encephalitis, and treatment targets the immune process causing the illness. The psychiatric symptoms were real; what changed was access to their cause. This is not an argument that most psychosis is undiscovered autoimmune disease. It demonstrates a more important principle: Similar psychiatric symptoms can arise from different biological mechanisms, and symptoms alone do not necessarily reveal their cause. How many conditions currently organized according to psychiatric symptoms will eventually be divided once their mechanisms become measurable?
Where biology meets meaning
Psychiatry must also determine when human experience becomes pathological. How much grief is abnormal after a death? When does fear become an anxiety disorder rather than a reasonable response to circumstances? When does ordinary distractibility become attention-deficit/hyperactivity disorder (ADHD)? When does an unusual belief become a delusion rather than a belief accepted within a person’s culture?
Science can inform these boundaries, but measurement cannot fully create them. Distress, impairment, dysfunction, and proportionality require judgments about human functioning. Psychiatric history shows how consequential those judgments can be. Homosexuality was once classified as a psychiatric disorder before scientific evidence and changing understandings of pathology led to its removal. The DSM’s handling of bereavement and depression has also changed as the field reconsidered whether depressive symptoms following the death of a loved one should be treated differently from otherwise similar episodes.
This history does not mean psychiatric diagnoses are arbitrary or simply products of social consensus. It shows that deciding what counts as a psychiatric disorder involves more than biology alone. Scientific evidence matters, but so do judgments about dysfunction, impairment, culture, and context. Those considerations are unavoidable when medicine is deciding when human thoughts, emotions, or behaviors have become pathological. They matter in a way they do not when measuring something like serum sodium.
Acknowledging these social and interpretive dimensions does not make psychiatric illness any less biological. Bereavement is a social event, but the brain processing loss is biological. Trauma is experiential, but memory formation and stress responses are biological. Psychotherapy operates through language, attention, memory, learning, and relationships, all of which involve a biological brain. The social affects biology, biology affects behavior, behavior changes the environment, and the environment feeds back onto biology. A psychiatrist and geneticist has described psychiatric illness as arising from influences across many levels, including genes and cells, brain function, development, psychology, relationships, and culture. Multiple levels of explanation may not represent competing theories. They may simply reflect the complexity of the system being studied.
Treatment can work while the cause remains uncertain
The limitations of psychiatric classification should not obscure the effectiveness of psychiatric treatment. Lithium can be remarkably effective in bipolar illness. Antipsychotics reduce symptoms and relapse in psychotic disorders. Electroconvulsive therapy can produce dramatic improvement in severe depression and catatonia. Evidence-based psychotherapies produce meaningful benefits across multiple conditions.
Effectiveness, however, is not proof of cause. A treatment can alter a biological system and improve a condition without correcting whatever originally caused it. Aspirin relieves a headache; a headache is not an aspirin-deficiency disorder. How a drug works and what caused a disease are different questions, and that distinction is easy to lose.
Consider the reaction to a 2022 umbrella review, which concluded that the available evidence did not convincingly support the familiar hypothesis that depression results from reduced serotonin activity. The paper drew enormous public attention and substantial methodological criticism from other researchers, and that dispute is worth following on its merits. But much of the coverage treated it as a verdict on whether antidepressants work, which is a separate claim entirely. A medication can be clinically effective without the illness having resulted from a deficiency in the system the medication affects.
Differences in treatment response may themselves eventually provide clues. If patients sharing the same diagnosis respond to fundamentally different treatments, one possibility is that treatment is not yet precise enough. Another is that the diagnostic category contains several different conditions grouped together because they produce overlapping symptoms.
When interpretation enters the courtroom
Nowhere are the stakes of this uncertainty higher than in court. Andrea Yates drowned her five children in June 2001. Twelve mental health professionals testified at her 2002 trial in Houston. Five testified that she did not know right from wrong or that she believed what she did was right. One, the forensic psychiatrist Park Dietz, testified that she knew her conduct was wrong. He was the state’s only mental health expert.
On cross-examination, Yates’s own attorney asked Dietz about his consulting work for the television program Law & Order. Dietz answered that an episode had depicted a woman with postpartum depression who drowned her children in a bathtub and was found insane, and that it had aired shortly before the killings. The prosecution then used that answer twice: to cross-examine a defense expert about whether she knew Yates watched the program, and in closing argument, where it connected her depression, her viewing habits, and the idea that she had seen a way out. The jury rejected her insanity defense and convicted her of capital murder.
No such episode existed. The error came to light after the guilty verdict but before the punishment phase. Yates moved for a mistrial, and the trial court denied the motion. In January 2005, the Texas First Court of Appeals reversed, concluding that there was a reasonable likelihood the false testimony could have affected the jury’s judgment and that denying the mistrial had been an abuse of discretion. The court was explicit on two points: The record did not show that Dietz had intentionally lied, and the prosecution’s use of his answer was not misconduct. The false testimony had nonetheless given his opinion greater weight. In a footnote, the court observed that had the jury known during deliberations that the testimony was false, it would likely have found the state’s only mental health expert less credible. At a second trial in 2006, another jury found Yates not guilty by reason of insanity.
That is the part worth sitting with. There was no demonstrated fraud and no misconduct. An experienced forensic psychiatrist made an ordinary human error of recollection inside a proceeding in which expert interpretation of an unobservable mental state carried enormous weight on the central question. Criminal responsibility is ultimately a legal judgment rather than a psychiatric diagnosis, but psychiatric experts often supply much of the evidence used to reach it.
There was no stored specimen capable of retrospectively producing a “psychosis level,” no biological threshold above which reality testing disappears, and no brain scan that could later establish definitively what Yates understood at a particular moment. Experts had to reconstruct her mental state from medical records, reported experiences, statements made at the time, medication history, behavior, and accounts from other people. They could disagree not only about what the evidence meant but about what her mental state had actually been.
That uncertainty is not necessarily incompetence or bad science. It reflects the nature of the evidence. Psychiatry frequently must infer an internal mental state rather than independently measure it. When those inferences influence criminal responsibility, involuntary commitment, disability determinations, competency, or parental rights, the distinction between inference and measurement stops being academic.
Medicine is not as objective as it imagines
Psychiatry should not be caricatured as uniquely subjective while the rest of medicine is treated as perfectly objective. Migraine remains primarily a clinical diagnosis. Pain is inherently subjective. Irritable bowel syndrome is diagnosed largely from symptoms. Radiologists disagree about images and pathologists about difficult specimens. Thresholds for hypertension and diabetes turn continuous biological measurements into categories. Many diseases were recognized clinically long before their mechanisms were understood.
The distinction is therefore one of degree. Medical knowledge exists along a continuum. At one end are diseases with directly demonstrated pathology or well-validated biomarkers. Further along are physiologic syndromes and symptom-defined clinical syndromes. At the other end are diagnostic constructs that depend heavily on behavior, subjective experience, and clinical interpretation. Psychiatry has an unusually large share of its major diagnoses toward the latter end.
That matters because diagnostic language conveys certainty. “You have bipolar disorder” sounds grammatically equivalent to “you have type 1 diabetes.” Scientifically, the two statements rest on different kinds and degrees of evidence. Bipolar disorder is inferred primarily from patterns of mood, energy, cognition, behavior, and functioning over time. Type 1 diabetes can be connected to a comparatively well-understood failure of insulin production, measurable metabolic abnormalities, and often identifiable autoimmunity. Both can describe serious illness and justify medical treatment. The certainty and type of evidence behind them are different.
What follows
The practical consequence is not that psychiatry should diagnose less. It is that medicine should say what kind of claim it is making.
In clinical practice, this can be remarkably simple: “This is a clinical diagnosis based on your symptoms, their course, and their effect on your life. There is no laboratory test that confirms it, and the diagnosis may be revised as we learn more.” That is not a hedge or a loss of medical authority. It is an accurate description of the evidence, and it may prevent a later revision from being experienced as a contradiction or a betrayal.
In forensic, disability, competency, and custody settings, the distinction should be explicit and on the record. Experts can separate which parts of an opinion rest on objective measurements, which come from documented history, and which represent an interpretation of a mental state that cannot be directly observed. Nothing about that weakens legitimate psychiatric testimony. It clarifies the basis of the conclusion and makes overreach easier to identify, whichever party has called the witness.
In biological research, DSM categories should not automatically serve as the fundamental units of investigation. They remain useful for clinical communication, epidemiology, treatment trials, and health services research, but usefulness does not guarantee that a category maps neatly onto a single biological disease. That was one of the central insights behind RDoC, and it remains important even though RDoC has not produced a replacement.
Psychiatry may eventually undergo something resembling oncology’s molecular revolution. Tumors once grouped by location and appearance have been divided according to molecular mechanism, while cancers arising in different organs are sometimes grouped together because they share the same actionable abnormality. Major depression may similarly fragment. Some forms of psychosis may eventually be classified according to specific genetic, immune, developmental, molecular, or circuit abnormalities. Other categories may survive largely intact because biology ultimately validates them.
Yet psychiatry will probably never become oncology, because its subject includes something a biopsy does not: meaning. Perfect knowledge of the circuitry involved in a delusion would still leave the questions of what the person believes, why the belief matters, and what it means within that person’s life and culture. Better biology may dramatically reduce psychiatric uncertainty without eliminating interpretation.
The important question is not whether mental illness is real. It is whether the names given to mental illnesses correspond to biological entities as precisely as medical language often implies. Until that correspondence is clearer, precision about the limits of a diagnosis is not an admission of failure. In a hybrid science, where the symptoms are also the test, it is part of the diagnosis.
This essay is cited in the KevinMD record on mental health.
Laurel A. Coons is a scientist with a background in genomics and biomedical research. She completed her doctoral training in pharmacology and cancer biology at Duke University and conducted research at the National Institute of Environmental Health Sciences. Her work has focused on genomic regulation, endocrine signaling, and translating complex scientific data into insights relevant to medicine and patient care.
She shares professional updates on X at laurelcoons.
