The Semantic Memory of Physicians - and More...

 

I have the somewhat grandiose plan to model psychiatric diagnosis based on the cognition of a physician rather than focusing on the externals.  By the externals I mean classification systems and critiques of classification systems.  At a later date – I might try to comment on how this approach compares with AI.  For now, I will try to keep it focused on human diagnosticians.  I have an interest in this is because I have made and witnessed incredible diagnoses and treatments by physicians and psychiatrists who I have been affiliated with. I don’t think there has been much of a focus on the process.  A secondary consideration is that cognitive neuroscience is a neglected subject in psychiatry and I hope to make the point that should change. I would go as far as suggesting that cognitive neuroscience should be taught to all psychiatrists more urgently than focusing on another DSM.    

Since the early 1970s, memory functions are divided along various lines clinically and functionally. The first division is long term memory and working memory (also called short term memory).  On the long-term side there is a further division to declarative and procedural memory.  Declarative memory is divided into episodic and semantic memory.  Episodic memory is the ability to recall discrete events.  Semantic memory can have a number of graded definitions.  A minimalist definition is factual knowledge independent of the source (7). A definition more informed by recent research in cognitive psychology: “General (encyclopedic) knowledge as well as schematic representations of events distilled from lifelong experiences, retrieved independently from their original spatial or temporal context” (9).  The authors in that case give examples of knowing who wrote the book “1984” and what generally happens at a birthday party.  That naturally raises the question how does all of this freestanding knowledge occur in the first place?  And also – does that imply a connection to episodic memory? In other words, does semantic memory occur when the context surrounding episodic memory is forgotten?

In the case of physicians there is a very long list of formative experiences across the course of one’s career.  The ability to recall them often assists in making diagnoses and provides an advantage over a physician who has not experienced that event.  Semantic memory is about concepts, words, and their relationship independent of a specific event or experience.  It typically consists of a collection of general facts and word meanings.  For example, it would include facts that apples can be red, green, or yellow and what a mechanic does.

Anyone familiar with cognitive screening examinations has probably asked questions focused on semantic memory.  Naming, word similarities, verbal fluency by word generation, general knowledge questions, are all examples. 

The semantic memory of a physician will contain many unique concepts and they will vary based on experience and exposure to clinical scenarios.  The general categories can be described as the following:

1:  Meaningful prior experiences – even though episodic memory stores specific events at specific intervals, semantic memory contains the specific meaning.  In the case of psychiatry an example would be seeing the effects of CMV encephalitis in a major university transplant unit and a decade later seeing similar behavior and consulting on a case in a general community hospital for similar findings.  That similarity triggers non-analytic hypothesis generation.

2:  Prototypes - the patterns noted in the above example can be averaged over a group of patients and those averages can be consolidated into prototypes.  In the above case a psychiatrist may have seen many cases of encephalitis and many cases of meningitis resulting in encephalitis and meningitis prototypes.  Similar prototypes may exist for all major neurological, medical, and psychiatric condition that they have encountered.  Note that the prototype differs from diagnostic criteria (the typical focus) because it is recall of all of the relevant and in many cases unique clinical features that were experienced.

3:  Specific patient memories (exemplars) – all physicians recall specific patients.  These memories are important for non-analytical reasoning like pattern matching.

4:  Knowledge Encapsulation – medicine like most professions is based on a system of graduated learning.  Basic science transitions rapidly into clinical medicine and then into clinical practice and lifelong learning.  At each stage prior knowledge is reorganized in a more efficient way.  In this case – general biomedical knowledge from basic science is organized under higher level concepts. 

An example in one of the references is a person with an infection who is experiencing progressive physiological problems.  At the medical student/basic science level the analysis might proceed from the basic science level and pathophysiology first.  At the clinician level the relevant pathophysiology is organized as sepsis and that provides a more immediate pathway for intervention.  The encapsulation encompasses and efficiently organizes the lower-level information.  At the same time experts must retain a significant amount of that earlier information.  

5:  Illness Scripts – are mental representations of diseases containing three different dimensions.  The first is enabling conditions like risk factors, demographics, predisposition, and context.  The second is fault or underlying pathophysiology.  The third is consequences including signs, symptoms, lab findings, and course or natural history.  Experts have a significant collection of these features.   

One of the questions in this area is what kind of illness script do physicians have?  Should they all be from their particular specialty or should these scripts encompass the totality of their training?  Some authors suggest that the pathophysiological mechanisms from basic science needs to be retained for true expertise – so my conclusion is that the illness scripts from the entirety of a physicians training probably remain relevant.

This is important in psychiatry because the general pathophysiology important in today’s environment was probably not taught is any detail in medical school and most conditions that are not secondary to medical conditions or the effects of drugs do not easily lend themselves to physiological explanations.  I would suggest that medical stability, generalized seizures and seizure variants, increased intracranial pressure, meningitis, encephalitis, cerebral localization, cerebellar dysfunction, peripheral neuropathies, coma, confusion/stupor/delirium, intoxication, and cranial nerve deficits are some of the illness scripts that every psychiatrist must have.

6:  Semantic Qualifiers - every physician has a lexicon of semantic qualifiers acquired in both medical school and post graduate training. They include anatomic descriptions (areas, more specific locations), pathological descriptions, disease course descriptions, and many others. Framing clinical scenarios with these qualifiers is often all that is needed to acquire associations to the disease of interest.

7:   Base rates and Context – experts by way of their clinical practice have an intuitive grasp of the base rates of various clinical conditions and how they typically present in their practice.  These rates of presentations and findings are integrated with the other features of semantic memory (disease scripts, patterns, etc) for more analysis and hypothesis generation.

These features of semantic memory are of course models of brain function for the most part determined by experimental models in cognitive psychology. Examples include testing for specific functions and seeing how those modelled functions vary among trainees and experts at various stages of development. 

Apart from the descriptive approaches used in many studies on physicians at various levels of training are there any more general models that could apply?  Cognitive neuroscience and cognitive psychology offer a more complete model of memory and knowledge structures as well as the underlying biology.  The lead figure for this post is a case in point and has the potential to consolidate many of the descriptions under a more comprehensive model based on experimental validation.

At levels B and C in the diagram we see a perceptual episode being processed from the left to the right in the diagram.  The activated or instantiated schema is a template for extracting relevant features and repressing irrelevant features.  In the diagram circles represent general concepts and squares are action scripts. Gist in the case of the model is a representation of a single episode where much of the detailed information is removed.  The overall sequence at level B depicts how a schema serves to form semantic type memory (gists) and at the same time can be altered or accommodated by new information.

Level A in the diagram illustrates what is known about the localization of these processes largely from human fMRI and preclinical studies.  Memory schemas are stored in various sites including the retrosplenial cortex (RSPL), middle and superior temporal gyrus (MTG/STG), anterior temporal lobe (ATL), and temporoparietal junction (TPJ).  These sites are bound per the diagram to the ventromedial prefrontal cortex (vmPFC).    Solid lines are context sensitive associative pathways biased by the vmPFC. Broken lines in the diagram represent context irrelevant associations that are not activated or inhibited.

How might all of this model work for psychiatry?  In general physicians are seeing a lot of patients in their training and practice.  In the course of that work - schemas are developed for diagnoses, signs, symptoms, and situations.  Here is a comparison of two scenarios that all psychiatrists are trained to recognize acute encephalitis and bipolar disorder, manic with psychotic features. 

	Encephalitis	Bipolar disorder, manic with psychosis
Schema	Acute illness, acute altered mental status, fever, seizures, focal neurological deficits, CSF/MRI abnormalities	Acute illness, euphoria/irritability/anger, hyperactivity, functional impairment, psychosis, temporal pattern, exclusion features
Subschema	Predisposing factors, pathophysiology patterns, temporal pattern	Euphoric expansive Irritable dysphoric Spontaneous v. precipitated
Gist	Acute confusion + fever + temporal lobe MRI changes = treat as HSV until proven otherwise" "Young woman + new psychosis + movement disorder = think anti-NMDAR, look for teratoma" "Summer encephalitis + flaccid paralysis = arboviral, likely West Nile" “Immunocompromised man with acute agitation = think CMV encephalitis	Episodic psychosis +/- mood changes (diagnosis gist) Mood stabilizer + antipsychotic (treatment gist) Severe postpartum psychosis = think bipolar disorder, manic with psychotic features Catatonia – think bipolar disorder, manic/depressed/mixed with psychotic features.

 

I came up with the following graphic (click to enlarge) based on the descriptive categories and the cognitive neuroscience model of Gilboa and Marlatte (12).  From left to right – the  “heterogenous construct supported by clinical utility” characterization is probably the most charitable one from philosophers.  Others like “this disorder does not exist” or “this disorder is not real” are two additional examples.  The central semantic memory category includes investigations and models of diagnostic reasoning conducted largely on medical students and physicians.  The cognitive neuroscience model contains schema and I have attempted to show how the concepts and actions map from the semantic memory to the schema model.  In both the semantic memory and cognitive neuroscience model, although the focus is memory the conceptualizations are really knowledge structures emphasizing a dynamic role for the schema in incorporating features of reality – in this case patient encounters. The cognitive neuroscience and semantic memory models also map on to brain anatomy – with a more comprehensive map for the cognitive neuroscience model as illustrated in the figure at the top.

What have I learned about this so far:

1:  The pattern matching of yesterday is more complicated today – I taught a course in diagnosis and diagnostic reasoning for 15 years into the early part of this century.  Pattern matching and pattern completion was a big part of that course.  The patterns were fairly simple and involved visual diagnoses (diabetic retinopathy, rashes) comparing physicians at various levels of training.  The most dynamic aspect was the implication that experts were better at matching incomplete patterns than novices.  Today’s conceptualizations of knowledge structures and schemas contain concepts, actions, and dynamically alter what is retained in memory and what is not. 

2:  There are clear implications for psychiatric diagnosis -   the DSM classification and all of the criteria do not capture the reality of medical and psychiatric diagnoses.  There is a qualifier in the manual that it is not a substitute for experience but that is never defined.  That reason becomes a lot clearer looking these cognitive models.  Classification systems attempt to operationalize the diagnostic reasoning of a physician by averaging a verbal description of those events.  I don’t think that is possible and I will cite a couple of examples.

Example 1:  A psychotherapist refers a 27-year-old woman to a psychiatrist because of concerns that she has histrionic personality disorder.  She has not been able to make progress in therapy.  The psychiatrist seeing the patient knows within minutes that she is manic.

Example 2:  An intern is presenting the history of a 68-year-old man to his psychiatric attending.  The patient is extremely depressed to the point that he believes that he is cursed based on a trivial event that occurred in his childhood. Within the first 5 minutes the attending realizes that the patient is delusional and communicates that to the intern. The intern acknowledges that this is true and wonders how he failed to make that diagnosis.

Both cases highlight that knowledge of a classification system is not enough.  The psychotherapist and the intern both know the DSM and use it regularly. They have both had didactics in classification of mental disorders.  The only difference is that the psychiatrist in both cases has experienced cases of the disorder and had knowledge structures and schema to make the diagnosis.  Written descriptions of schema and knowledge structures are an incomplete approach to diagnostic reasoning. 

3: Classifications artificially separate actions from concepts – any reading of the DSM gives the impression that “this is the universe of psychiatric disorders – in order to function as a psychiatrist, pick one and then come up with a treatment plan.”  This is problematic at two levels.  First, if the cognitive neuroscience model of memories and knowledge structures is correct – a classification system is operating at a sublevel that averages features.  It is blind to the overall gist that despite this averaging no two people are alike.  Second, it removes action features that are necessary to function as a physician.  That would include top level schemas like “This patient is medically unstable and requires medical or surgical care first” or “This is a life-threatening problem that requires a safe and closely monitored environment." Some will argue that is not the goal of classification.  I would argue that many consider classification to be a diagnosis and in order for it to function that way – it needs to include action items in addition to a general rule out of causative intoxication states and medical problems. The DSM as it exists is classification without diagnosis.

4:  Cognitive neuroscience models highlight the fact that the separation between diagnosis and treatment is artificial.  All physicians are taught to do exhaustive evaluations of medical problems.  That is the initial step in a career.  It is also critical to learn when that exhaustive process needs to be immediately interrupted to focus on a more acute problem. I can still recall seeing a 7-year-old boy who have been hit by a car while playing in the street. He was alert but had significant abdominal pain.  The car bumper struck him just below his left rib cage.  It took me less than 5 minutes to determine that he had an acute abdomen and call the trauma surgeons. That non-linear process happens frequently in acute care psychiatry and in outpatient psychiatry with patients in crisis who need verbal interventions to assist in the diagnostic and treatment process.  

5:  Psychotherapy – there are recent perspectives on how cognitive psychology applies to the psychotherapeutic process at both the psychological and biological levels using these models.  Basically, maladaptive schemas are confronted and modified during the therapy.  There is some empirical evidence that this may happen particularly in the area of positive and negative self-schemas.  Much of this literature draws on existing cognitive behavioral therapy.  That leads to a question of what is the difference between a therapy focused on a cognition or an isolated memory compared with a schema focused therapy?

At the highest level of analysis memory focused therapies generally involve isolated autobiographical memories and schema focused therapies are about knowledge structures abstracted across multiple events that involve emotion, cognition, and behavior.  In theory the schema focused therapies may be useful in cases where the memory focused therapy is not effective, but a competing consideration is that schemas can be entrenched and difficult to change.  The memory focused therapy could be considered a bottom-up type of approach and the schema focused a top-down approach. 

6:  Criticisms – Criticizing the DSM as a diagnostic system is a cottage industry in the US and the UK.  As we approach a new version of the DSM expect most media sites to start months and even years of criticism. Practically everybody does it rarely discussing their motivations, understanding, and the limitations of their proposed system if they have one.  If diagnostic reasoning is a complex process consistent with the cognitive neuroscience models and requires direct experience, criticism of the manual rings hollow.  It is equivalent to reading about things that might exist and proclaiming you are an expert.  Psychiatrists with criticisms are also limited if they have insufficient experience in the areas they are criticizing.  Psychiatrists with the broadest experience will produce the best criticism. If you are criticizing a list of diagnostic criteria in a classification system in isolation – that is exactly what you are doing.  It is trivial compared with an actual diagnosis by a trained and experienced psychiatrist.        

This brief focus on the cognitive neuroscience of diagnosis should highlight that psychiatric education and practice is seriously lagging in this knowledge base.  If we are taking the “diagnosis” in DSM seriously it has to be modified to include this important brain science.  All of the current competing models face the same criticism.  A diagnosis by a physician is much more than typed criteria attempting to capture a dynamic process.  Secondly, psychiatry needs modern approaches to the mind. Approaches that correlate with neurobiology and have a clear empirical basis. Much of the DSM claims a sketchy atheoretical basis that should no longer be acceptable when powerful explanatory theories may exist.  Philosophy is no substitute.  Finally, we must find a way to implement these across all of our training programs and practitioners.  We should be devoting as many resources to integrating cognitive neuroscience into psychiatry as we do modifying the DSM.

And that should be the first step.  What does a DSM looked like with cognitive neuroscience baked in?  The answer goes a lot farther than “dimensions”.      

   

George Dawson, MD, DFAPA

 

 

References:

 

1:  Norman G, Young M, Brooks L. Non-analytical models of clinical reasoning: the role of experience. Med Educ. 2007 Dec;41(12):1140-5. doi: 10.1111/j.1365-2923.2007.02914.x. Epub 2007 Nov 13. PMID: 18004990.

2:  Brush JE Jr, Sherbino J, Norman GR. Diagnostic reasoning in cardiovascular medicine. BMJ. 2022 Jan 5;376:e064389. doi: 10.1136/bmj-2021-064389. PMID: 34987062.

3:  Custers EJ. Thirty years of illness scripts: Theoretical origins and practical applications. Med Teach. 2015 May;37(5):457-62. doi: 10.3109/0142159X.2014.956052. Epub 2014 Sep 2. PMID: 25180878.

4:  Koufidis C, Manninen K, Nieminen J, Wohlin M, Silén C. Unravelling the polyphony in clinical reasoning research in medical education. J Eval Clin Pract. 2021 Apr;27(2):438-450. doi: 10.1111/jep.13432. Epub 2020 Jun 22. PMID: 32573080.

 5:  Binder JR, Desai RH, Graves WW, Conant LL. Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies. Cereb Cortex. 2009 Dec;19(12):2767-96. doi: 10.1093/cercor/bhp055. Epub 2009 Mar 27. PMID: 19329570; PMCID: PMC2774390.

6:  Duff MC, Covington NV, Hilverman C, Cohen NJ. Semantic Memory and the Hippocampus: Revisiting, Reaffirming, and Extending the Reach of Their Critical Relationship. Front Hum Neurosci. 2020 Jan 24;13:471. doi: 10.3389/fnhum.2019.00471. PMID: 32038203; PMCID: PMC6993580.

7:  Insaustu R, Amaral DG. Hippocampal Formation. In: Mai JK, Paxinos G (eds) The Human Nervous System, 3^rd ed.  Elsevier, London, 2012: p. 933.

8:  Mazoué A, Gaultier A, Rocher L, Deruet AL, Vercelletto M, Boutoleau-Bretonnière C. Does a rabbit have feathers or fur? Development of a 42-item semantic memory test (SMT-42). J Clin Exp Neuropsychol. 2022 Sep;44(7):514-531. doi: 10.1080/13803395.2022.2133088. PMID: 36269845.

9:  Renoult L, Irish M, Moscovitch M, Rugg MD. From Knowing to Remembering: The Semantic-Episodic Distinction. Trends Cogn Sci. 2019 Dec;23(12):1041-1057. doi: 10.1016/j.tics.2019.09.008. Epub 2019 Oct 28. PMID: 31672430.

10:  Brown TI, Rissman J, Chow TE, Uncapher MR, Wagner AD. Differential Medial Temporal Lobe and Parietal Cortical Contributions to Real-world Autobiographical Episodic and Autobiographical Semantic Memory. Sci Rep. 2018 Apr 18;8(1):6190. doi: 10.1038/s41598-018-24549-y. PMID: 29670138; PMCID: PMC5906442.

11:  Teghil A, Bonavita A, Procida F, Giove F, Boccia M. Temporal Organization of Episodic and Experience-near Semantic Autobiographical Memories: Neural Correlates and Context-dependent Connectivity. J Cogn Neurosci. 2022 Nov 1;34(12):2256-2274. doi: 10.1162/jocn_a_01906. PMID: 36007071.

12:  Gilboa A, Marlatte H. Neurobiology of Schemas and Schema-Mediated Memory. Trends Cogn Sci. 2017 Aug;21(8):618-631. doi: 10.1016/j.tics.2017.04.013. Epub 2017 May 24. PMID: 28551107.

13:  Reyna VF, Edelson S, Hayes B, Garavito D. Supporting Health and Medical Decision Making: Findings and Insights from Fuzzy-Trace Theory. Med Decis Making. 2022 Aug;42(6):741-754. doi: 10.1177/0272989X221105473. Epub 2022 Jun 23. PMID: 35735225; PMCID: PMC9283268.

14:  Wilhelms EA, Fraenkel L, Reyna VF. Effects of Probabilities, Adverse Outcomes, and Status Quo on Perceived Riskiness of Medications: Testing Explanatory Hypotheses Concerning Gist, Worry, and Numeracy. Appl Cogn Psychol. 2018 Nov-Dec;32(6):714-726. doi: 10.1002/acp.3448. Epub 2018 Sep 1. PMID: 30686857; PMCID: PMC6345391.

15:  Hawke LD, Provencher MD, Parikh SV. Schema therapy for bipolar disorder: a conceptual model and future directions. J Affect Disord. 2013 May 15;148(1):118-22. doi: 10.1016/j.jad.2012.10.034. Epub 2012 Dec 4. PMID: 23218898.

16:  Lane RD, Ryan L, Nadel L, Greenberg L. Memory reconsolidation, emotional arousal, and the process of change in psychotherapy: New insights from brain science. Behav Brain Sci. 2015;38:e1. doi: 10.1017/S0140525X14000041. Epub 2014 May 15. PMID: 24827452.

Graphics Credit:

1:  The lead graphic as noted is from Cell Press and reference #12.  It is reproduced here with permission from Elsevier and this is their acknowledgement:

Reprinted from Trends in Cognitive Sciences, August 21(8), Gilboa A, Marlatte H. Neurobiology of Schemas and Schema-Mediated Memory, p. 618., Copyright 2017, with permission from Elsevier.  License 6278000229455, May 29, 2026 

2:  Second graphic was made by me using Microsoft Visio.

Supplementary 1:  Nobel Laureate and Psychiatrist Eric Kandel noted the importance of cognitive neuroscience years ago and this was a quote from his book:  The Age of Insight.

 

The Majority of DSM Diagnoses Are Never Used...

 

The landscape of medical and psychiatric diagnoses that are actually used by clinicians has always interested me.  Diagnostic classifications like the DSM and the ICD are generally used for the purpose of billing and generating statistics.  There is also an implicit research function that is probably why the number of diagnoses are so expansive.  I wrote a brief comment about this on my other blog almost exactly 3 years ago.  In a study of 1,260,097 psychiatric diagnoses reported from hospital care between 2001-2007 only 16 or 4.2% of the available diagnoses accounted for 50% of the reported activity (1).  Forty-nine diagnoses accounted for accounted for 75% of the activity and 108 diagnoses accounted for 95% of the activity.  Of the total diagnoses available most were used infrequently if at all.  In a separate abstract, 32 diagnoses were not used at all and 121 diagnoses were used in less than 0.1% of cases (2).  This is an important issue that I intend to use in further posts about diagnostic reasoning in psychiatry.  This is intended as an update and moving the concept over to my main blog.

The first question when it comes to either DSM or ICD diagnoses in different clinical settings is – how many are there?  In the case of the DSM – I personally counted the diagnoses and came up with 281 diagnoses using the methods outlined in that post.  Since then, I have encountered a reference that lists the total diagnoses as 245 (3).  In an earlier DSM-III study of 11,292 general psychiatric admission 296 of 329 available diagnoses were used and the 9 most frequent accounted for 35.8% of all diagnoses (4).

Surveys of Psychiatric Diagnoses Used In Practice
N	Classification	Used/Available (%)	Skew
11,292 adults	DSM-III	296/329 (90%)	73% of diagnoses were from 6 diagnostic categories with major depression the predominate category at 23%
214,206 adults	ICD 9/10-CM	----	mood disorders (22%), anxiety disorders (21%), and substance use disorders (16%) together accounted for the majority of documented psychiatric diagnoses
13,684,154 children and adolescents	ICD 9/10 – grouped as 13 diagnostic groups and 1 other	-----	Diagnostic groups were trauma/stressor-related disorders (27%), anxiety disorders (19%), and depressive disorders (17%)
7,076 adults	DSM-III-R	------	41.2% of the adult population under 65 experienced at least one DSM-III-R disorder in their lifetime, 23.3% within the preceding year. Depression, anxiety, and alcohol abuse and dependence were most prevalent
1:  Mezzich JE, Fabrega H Jr, Coffman GA, Haley R. DSM-III disorders in a large sample of psychiatric patients: frequency and specificity of diagnoses. Am J Psychiatry. 1989 Feb;146(2):212-9. doi: 10.1176/ajp.146.2.212 2:  Barr PB, Bigdeli TB, Meyers JL. Prevalence, Comorbidity, and Sociodemographic Correlates of Psychiatric Diagnoses Reported in the All of Us Research Program. JAMA Psychiatry. 2022;79(6):622–628. doi:10.1001/jamapsychiatry.2022.0685 3:  Mojtabai R, Olfson M. Trends in Mental Disorders in Children and Adolescents Receiving Treatment in the State Mental Health System. J Am Acad Child Adolesc Psychiatry. 2025 Aug;64(8):906-920. doi: 10.1016/j.jaac.2024.08.008. Epub 2024 Aug 28. PMID: 39214290. 4:  Bijl, R., Ravelli, A. & van Zessen, G. Prevalence of psychiatric disorder in the general population: results of the Netherlands Mental Health Survey and Incidence Study (NEMESIS). Soc Psychiatry Psychiatr Epidemiol 33, 587–595 (1998). https://doi.org/10.1007/s001270050098

 I have listed several additional surveys of diagnoses in various samples.  Comparison across studies is complicated by the classification system used and whether specific diagnoses are counted or diagnostic groups.  If only groups are counted it is more difficult to illustrate the skew by weighting.  Large healthcare systems have these statistics but I am not aware of any of that data being published.  Having worked for one of those systems the data is often considered proprietary.  The data would also be affected by the clinical populations being treated.  I would expect safety net hospitals to have a much higher percentage of disability associated diagnoses than private hospitals.  I would expect the same skew between acute care settings (inpatient units and acute psychiatric services) to have a different distribution of diagnoses than outpatient clinics.  Of the 3 studies that looked at this issue above using DSM criteria – most DSM diagnoses are used infrequently if at all.

What about the criticism of the proliferation of diagnoses?  I expect to see the usual discussion of this issue as the DSM-6 is hyped as a controversial topic over the coming years.  We already know the answer to the question but everyone will need to pretend that we don’t. By my count the DSM diagnoses peaked with the DSM-IV.  A lot of the controversy about diagnostic proliferation will start by saying the DSM-I had 106 diagnoses in 1954 and that number has more than doubled.  Nobody will say that most clinicians are using a set of diagnoses so limited that they have the numerical codes memorized so they do not have to keep looking them up.  

A comparable look at the ICD shows that it started out in in 1893 as the International List of Causes of Death (or the Bertillon Classification of Causes of Death).  There were 44, 99, or 161 codes that could be used depending upon the reporting capabilities of the country.  The 161-code version became the ICD and in 1898 the American Public Health Association (APHA) recommended that Canada, Mexico, and the US adopt it and revise it every 10 years based on advancements in medical knowledge.  The current version ICD-11 has 55,000 codes up from the previous version (ICD-10) 14,000 codes.  

Any comparison of numbers of diagnoses is problematic for several reasons.  The authority proposing the classification system is averse to reporting them. That is true whether it is the DSM or the ICD. When I counted them, I provided the methodology and you can replicate it yourself.  With the DSM there are occasional isolated counts close to mine – but no explanations.  With the ICD – things are more complex and estimated range from 10,000 – 15,000 diagnoses that would be recognized as unique.  In the ICD-11 those diagnoses are included with other biomedical terms in the underlying Foundation of the ICD.  The Foundation is technically a semantic database of terms including symptoms and other findings. 

Before getting into how these codings work relative to diagnoses – a brief introduction to ICD coding terminology since it is impossible to separate out what physicians typically consider diagnoses.  In the example below, I have produced a hierarchical tree diagram that is considered the basis for the ICD.  In the example I am following how an episode of recurrent depressive disorder-severe without psychotic features is coded.  The top category is the grouping of all medical disorders into 28 categories.  The next group is all mental, behavioral, and neurodevelopmental disorders grouped into 24 categories.  From there a mood disorder group, depressive disorder group and recurrent depressive disorder group follows.  The final grouping is the variant of 15 recurrent depressive disorder possibilities that we are looking for.  In ICD jargon, that final group is called a leaf code because it is the ultimate result of the hierarchy and it cannot be split any farther.  The branching above that level is called stem codes.   

 

A more interesting comparison is how the diagnostic codes in the rest of medicine have increased.  

Version	Approximate Leaf Codes	Notes	References
ICD-10 (WHO)	~10,607	Base international version	[1]
ICD-10-CM (US)	~71,932	US clinical modification with extensive granularity	[1]
ICD-11-MMS	~14,622	Moderate increase over ICD-10; post coordination expands expressivity	[1]
ICD-11 Foundation	Much larger	Includes 5,500+ rare diseases; serves as semantic knowledge base	[2-3]
1:  Fung KW, Xu J, Bodenreider O. The new International Classification of Diseases 11th edition: a comparative analysis with ICD-10 and ICD-10-CM. J Am Med Inform Assoc. 2020 May 1;27(5):738-746. doi: 10.1093/jamia/ocaa030. PMID: 32364236; PMCID: PMC7309235. 2:  Feinstein JA, Gill PJ, Anderson BR. Preparing for the International Classification of Diseases, 11th Revision (ICD-11) in the US Health Care System. JAMA Health Forum. 2023;4(7):e232253. doi:10.1001/jamahealthforum.2023.2253 3: Chute CG. The rendering of human phenotype and rare diseases in ICD-11. J Inherit Metab Dis. 2018 May;41(3):563-569. doi: 10.1007/s10545-018-0172-5. Epub 2018 Mar 29. PMID: 29600497; PMCID: PMC5959961.

   

The table shows a direct comparison between the ICD-10 and ICD-11.  The conclusion is that there has been a moderate increase in codes.  Leaf codes can undercount and overcount the diagnoses and are not necessarily strict representations of diagnoses.  For example, a code of type 2 diabetes mellitus can generate many additional codes depending on the complications.  The only equivalent in the DSM are the modifier codes.  Medicine can also code symptoms rather than a specific diagnosis – so those codes like “neck pain, cough, constipation, etc) also generate codes that have no DSM equivalent.  There is residual or not-otherwise-specified (NOS) codes in the ICD that meet no diagnostic criteria.  The DSM-5-TR has replaced NOS codes with other specified disorder or unspecified disorder that are probably not much better.  The ICD-11 added complexity codes for severity, histopathology and other features to increase specificity.

The structure of the ICD is relevant to counting diagnoses.  The basic a hierarchical tree structure that can be viewed at the following link.  In this case the diagram illustrates the hierarchy Mental, Behavioral, or Neurodevelopmental disorders (category) -> Mood Disorders (3) -> Bipolar Disorder -> Bipolar Type 1 Disorder (16) -> Bipolar type I disorder, current episode manic, without psychotic symptoms (32) -> Bipolar type I disorder, current episode manic, without psychotic symptoms, with prominent anxiety symptoms (32).  The numbers in parentheses indicate the total branching of the hierarchical tree diagram.  The branching is graphically represented in the center panel.

A comparison of leaf codes in the DSM is possible by estimating leaf codes as 3-5-digit total billable codes.  That would include about 350 leaf code like endpoints and 150 environmental codes for total of about 500.  That is only about 3% of the total ICD-11 codes in the above table – a number made more significant by the fact that the DSM includes diagnoses that the ICD codes in other categories – most notably neurocognitive disorders.   

In conclusion – all of the controversy about the proliferation of diagnoses (or codes) in the DSM as excessive does not match the reality of how diagnoses in general have increased in the rest of medicine. If anything, it seems to be lagging.  It also misses the point why this happens in the first place as it was well put by the American Public Health Association in 1898 – to revise the ICD every 10 years “based on advancements in medical knowledge.”   

 

George Dawson, MD, DFAPA

 

Supplementary 1:  Leaf code approximation:  I counted all of the diagnoses listed in the chapter "Numerical Listing of DSM-5 Diagnoses and Codes (ICD-10-CM)” Total codes listed in that appendix are 760 but it is a mapping of DSM diagnoses onto the ICD-10 and that is not an exact match.  As a result, there are 148 duplicate codes bringing the total down to 612.  The list also contains parent or sub-stem codes such as F79 (Unspecified intellectual disability) that requires an additional digit to become a leaf code.  There are 23 sub-stem codes bringing the total number of leaf codes to 589.

Applying that number to the approximate total leaf codes in the above table yields the following:

589/10,607 = 5.6%

589/14,622 = 4.0%

589/71,932 = 0.8%

Those numbers are consistent with the number of codes and diagnoses in psychiatry are certainly not excessive compared with the rest of medicine and the estimated disease burden (see fig 3).

 

References:

1:  Munk-Jørgensen P, Najarraq Lund M, Bertelsen A. Use of ICD-10 diagnoses in Danish psychiatric hospital-based services in 2001-2007. World Psychiatry. 2010 Oct;9(3):183-4. doi: 10.1002/j.2051-5545.2010.tb00307.x. PMID: 20975866; PMCID: PMC2948730. 

2:  Müssigbrodt H, Michels R, Malchow CP, Dilling H, Munk-Jørgensen P, Bertelsen A. Use of the ICD-10 classification in psychiatry: an international survey. Psychopathology. 2000 Mar-Apr;33(2):94-9. doi: 10.1159/000029127. PMID: 10705253.

3:  Leucht S, van Os J, Jäger M, Davis JM. Prioritization of Psychopathological Symptoms and Clinical Characterization in Psychiatric Diagnoses: A Narrative Review. JAMA Psychiatry. 2024;81(11):1149–1158. doi:10.1001/jamapsychiatry.2024.2652

4:  Mezzich JE, Fabrega H Jr, Coffman GA, Haley R. DSM-III disorders in a large sample of psychiatric patients: frequency and specificity of diagnoses. Am J Psychiatry. 1989 Feb;146(2):212-9. doi: 10.1176/ajp.146.2.212. PMID: 2783540.

5:  Chute CG, Çelik C. Overview of ICD-11 architecture and structure. BMC Med Inform Decis Mak. 2022 May 16;21(Suppl 6):378. doi: 10.1186/s12911-021-01539-1. PMID: 35578335; PMCID: PMC9109286.

6:  Harrison JE, Weber S, Jakob R, Chute CG. ICD-11: an international classification of diseases for the twenty-first century. BMC Med Inform Decis Mak. 2021 Nov 9;21(Suppl 6):206. doi: 10.1186/s12911-021-01534-6. PMID: 34753471; PMCID: PMC8577172.

7:  Quan H, Steinum O, Southern DA, Ghali WA. Coding mechanisms for main condition in ICD-11. BMC Med Inform Decis Mak. 2025 Jul 10;21(Suppl 6):387. doi: 10.1186/s12911-025-03069-6. PMID: 40640794; PMCID: PMC12243148.

What Does ERISA Say About AI Guardrails?

 

 

A colleague sent me a news article this morning about a couple suing a major AI firm for advice given by their chatbot to their son resulting in a fatal overdose.  As a psychiatrist most of what I read about problematic AI comes in the form of AI hallucinating false medical references (1), AI induced psychosis in people who either use it excessively or who are predisposed, or AI facilitating its own use by excessive praise or obsequiousness.  In the latter case it can result is emotional attachment to the AI that of course is unwarranted.  I have also flagged a couple of cases that illustrate the problems when AI is applied to moral and political decision making.

I decided to do a little more research on the subject.  I was surprised to find a Wikipedia page titled Deaths Linked To Chatbots. Thirty-three deaths are listed not including the case I was investigating. The suggested pathways to violence generally include overuse, emotional attachment, and bad advice biased toward reinforcing irrational decisions.  The evidence contained on this page highlights a couple of concepts that might not be apparent to most people including the architects of AI.  The first is the importance of emotion in human decision making. This was articulated by Bechara in the past who demonstrated that if there is a disruption between emotional and cognitive systems in the human brain – even basic decisions become impossible.  Other disruptions in the same systems can lead to an array of emotional dysregulation and the associated irrational and often socially inappropriate decisions.  Second, emotional biases clearly affect decision making in the case of intact brains.  There is perhaps no better example than the current American political system installing a less competent government that is clearly not in support of the wants and needs of most Americans.

Secondly, humans can form intense attachments to inanimate objects that are unable to reciprocate.  The classic example is developmentally normal transitional objects (stuffed animals, toys, blankets).  Winnicott theorized that in infancy – this object is recognized as not part of the self or external reality.  It is a fantasized relationship that represents a future “illusion”(2).  According to Winnicott’s theory the transitional object loses meaning during normal development and becomes irrelevant.  Persistence into later stages may indicate a normative transition like object attachment during grieving, to a way to compensate for the lack of interpersonal attachments, to personality or psychopathology. 

Chatbots can be significant attachment figures and this is currently an area of study (4-6).  The area of human – digital object transference is also being explored (6) as well as the projection of human needs onto a digital object (8), and more complex models of human-machine connectedness (9).  This literature is referenced primarily to indicate that there is a lot that is not known about the array of human responses to interactions with these machines and what the possibilities are.

Apart from my previous concerns that machines lack consciousness and have demonstrated a lack of adequate moral decision-making there is always the question of programming and algorithms. Both of the features are the bane of most Internet users who find that their most mundane interests are often amplified to result in a barrage of advertisements and sales offers.  And then there is the army of misinformation bots spreading foreign and national political propaganda 24 hours a day.  None of that requires AI but is there any doubt that AI will make it worse and harder to detect?

It is no secret that the current AI explosion is a multitrillion dollar enterprise being run by a handful of men who have shown no interest in the environment, social equity, or human rights. They immediately aligned themselves with an autocratic government at the highest levels and so far, have had no regulation of their AI.  As a result, that AI is spewing out massive amounts of information that the average citizen is taking as legitimate if not some type of advanced advice. The complications of that advice include the deaths, environmental damage from the required power generation, and societal damage from unemployment.  There is additional damage based on inequity from wealth concentration.  The barrage of pro-AI hype in the media greatly exceeds any realistic discussion of the downsides.  The only clear benefit that most people see is their ability to sit at home and entertain themselves with a chatbot or see if an AI can do their homework or other projects.  The purported efficiency seems offset by a tremendous amount of time wasted.

At the minimum – in the case that started this post there is a stark contrast between human decision makers and AI.  In 40 years of practice – I never recommended kratom by itself or with alprazolam (Xanax) or Benadryl (diphenhydramine).  In fact, I spent a considerable amount of time getting people off of alprazolam and later kratom. But I am not unique in this – I don’t know of any physician who would make these recommendations.  But those recommendations form the basis for the AI lawsuit. 

That highlights the danger of the current hype that AI will replace physicians or the predictable studies that comparing AI to physicians shows that AI can be safely consulted.  There are even stories that AI is prescribing drugs in some settings without physician input.  The question of agency is never addressed and that seems like the basis for this lawsuit.  Corporations always seem to do good job of avoiding responsibility in healthcare.  The classic example is the Employee Retirement Income Security Act of 1974 (ERISA).  The pre-emption clause of ERISA means that in employer-sponsored health plan covered employees cannot bring state malpractice or negligence claims against their managed care organization (MCO) for injuries from denial of plan benefits, utilization review decisions, failure to use qualified physicians, or improper plan administration.   The reviewing physicians working for MCOs are also generally protected and the associated arguments are that utilization review is not the practice of medicine and/or the reviewers have no accountability/duty to the patient. Several studies have documented the patient harms related to this accountability gap and despite several attempts at amelioration it remains largely intact and a considerable source of financial success for managed care organizations.

The critical question is whether this kind of accountability gap will exist with AI.  It is easy to envision a scenario where AI is implemented to review charts and prescribe low risk medications like many online services do now.  Will AI eventually take the place of physician reviewers employed by MCOs? Will consumers and patients be led to believe that AI is making decisions that affect their medical care based on the best available information or in the interest of the corporation. Current statistics suggest that there are tens of millions of these decisions made every year.  AI can greatly increase that as well as the harassment factor if decisions are being appealed.

With all of the political talk about guardrails for AI – it is important to recognize that these guardrails need to exist at several levels.  Right now, it is not much of a stretch to say that AI is out there practicing medicine without a license. In the majority of cases like the initial example, the user does not know if the search result if strictly from medical literature or something else.  The user does not know if the AI is exercising the judgment of an average physician or in malpractice parlance using the community standard of care.  The user does not know if their psychology in terms of defense mechanisms or attachment style to inanimate objects or AI is being exploited.  The user does not know if the AI is just telling them what they want to hear.  And the user does not know if the AI is providing information in their best interest or the interest of corporations or the government.

I read a study doing research for this post and subjects were asked to rate the professionalism of the AI.  In my opinion the single-most significant determinant of professionalism for physicians is accountability and duty to their patients.  It fuels not only the immediate encounter but the concept of life long learning and service to patients. It is usually evident over time but only indirectly in the form of positive results and a positive relationship over time.  AI in its current form does not have it and I am not convinced that a society or culture that came up with ERISA can construct physician-like guardrails around medical AI.   

 

George Dawson, MD, DFAPA

 

Supplementary 1:  It came to my attention after posting this that managed care organization (MCOs) have already implemented AI for utilization review and care denials.  Part of the problem in getting an accurate estimate of how much AI is involved is that this is an area where algorithms have been in place for a long time. Some of the care denials may be algorithmic and some may be due to a new AI interface.  This is what I have so far.  If you have additional references or data – please send it my way and I will add it to this post.

Automated prior authorizations is an early application for triage based on various data sources, medical necessity, and machine assistance on the provider side. A Congressional Investigation of the 3 major companies providing Medicare Advantage insurance plans showed that over the course of 4 years (2019-2022) – denials of prior authorization requests for post-acute care increased and was consistently larger than the denials for all other types of care (11 – see page 19 Figure 1). As United Health Care automated the process the denial rate increased.  The document is clear that prior authorization by these companies is highly profitable and even though a small percentage of denials are appealed – most of those appeals are also denied. The overriding concern is that AI or other automatic of the prior authorization process will greatly increase the number of denials overwhelming whoever is on the physician-patient side who needs to make the appeal.   It is more than a little ironic that a process that so clearly favors the managed care industry needs additional leverage from AI.

Disclaimer:  I have made the argument several times on this blog that prior authorization should just be made illegal since it serves no useful purpose other than making money for companies that do not actually provide patient care and it forces physicians and nurses to work for free while addressing these denials.  The total cost of that work was estimated to be worth $31B in 2009.  The estimated cost of drug utilization management alone is $93 billion (14).       

Supplementary 2:  In the past 5 years I have fielded many complaints about authorization for post-acute care (PAC) from friends, relatives, and people contacting me here to figure out what to do about it.  A typical scenario is a 70+ year old adult hospitalized for a a significant problem.  The hospital team wants them discharged ASAP of course even though in many cases their primary problem has not been adequately treated.  They realize the patient cannot care for themselves at home and there are often no caregivers available and want to transfer them to a skilled nursing facility (SNF) for rehabilitation.  In many cases it is specialized rehabilitation like post stroke, heart attack, or traumatic brain injury rehabilitation and the patient lacks basic skills to care for themselves.  I am fielding selected complaints but all of these transfers were denied - often repeatedly to the point the patient and family were demoralized and gave up.  In one case the patient was dead within 48 hours of discharge.  Given the results in reference 11 - this appears to be a financial strategy.       

References:

1:  Topaz M, Roguin N, Gupta P, Zhang Z, Peltonen LM. Fabricated citations: an audit across 2·5 million biomedical papers. Lancet. 2026 May 9;407(10541):1779-1781. doi: 10.1016/S0140-6736(26)00603-3. PMID: 42107362.

2:  Kernberg OF.  Object relations theories and techniques.  In:  Textbook of Psychoanalysis, 2^nd ed.  Person ES, Cooper AM, Gabbard GO, eds.   Washington DC: American Psychiatric Association Publishing, 2025: 57-75. 

3:  Bachar E, Canetti L, Galilee-Weisstub E, Kaplan-DeNour A, Shalev AY. Childhood vs. adolescence transitional object attachment, and its relation to mental health and parental bonding. Child Psychiatry Hum Dev. 1998 Spring;28(3):149-67. doi: 10.1023/a:1022881726177. PMID: 9540239.   

4:  Cheng N, Yu R. Measuring and understanding emotional attachment in human-AI relationships. Ergonomics. 2026 Feb 2:1-20. doi: 10.1080/00140139.2026.2622539. Epub ahead of print. PMID: 41622967. 

5:  Liu T, Lo TY, Wen KH, Sun Y, Wei ZQ. Pathways of long-term AI virtual companion app use on users' attachment emotions: a case study of Chinese users. Front Psychol. 2026 Jan 12;16:1687686. doi: 10.3389/fpsyg.2025.1687686. PMID: 41602682; PMCID: PMC12833267.

6:  Koles B, Nagy P. Digital object attachment. Curr Opin Psychol. 2021 Jun;39:60-65. doi: 10.1016/j.copsyc.2020.07.017. Epub 2020 Jul 22. PMID: 32823244.

7:  Holohan M, Fiske A. "Like I'm Talking to a Real Person": Exploring the Meaning of Transference for the Use and Design of AI-Based Applications in Psychotherapy. Front Psychol. 2021 Sep 27;12:720476. doi: 10.3389/fpsyg.2021.720476. PMID: 34646209; PMCID: PMC8502869.

8:  Saracini C, Cornejo-Plaza MI, Cippitani R. Techno-emotional projection in human-GenAI relationships: a psychological and ethical conceptual perspective. Front Psychol. 2025 Sep 29;16:1662206. doi: 10.3389/fpsyg.2025.1662206. PMID: 41089650; PMCID: PMC12515930.

9:  Boyd RL, Markowitz DM. Artificial Intelligence and the Psychology of Human Connection. Perspect Psychol Sci. 2026 Mar;21(2):192-220. doi: 10.1177/17456916251404394. Epub 2026 Jan 29. PMID: 41608879; PMCID: PMC12960742.

10:  Sahni NR, Carrus B. Artificial Intelligence in U.S. Health Care Delivery. N Engl J Med. 2023 Jul 27;389(4):348-358. doi: 10.1056/NEJMra2204673. PMID: 37494486.

11:  US Senate Permanent Subcommittee on Investigations. Refusal of Recovery: How Medicare Advantage Insurers Have Denied Patients Access to Post-Acute Care. October 17, 2024. Accessed March 24, 2025. hsgac.senate.gov/wp-content/uploads/2024.10.17-PSI-Majority-Staff-Report-on-Medicare-Advantage.pdf

12:  Mello MM, Trotsyuk AA, Mahamadou AJD, Char D. The AI Arms Race In Health Insurance Utilization Review: Promises Of Efficiency And Risks Of Supercharged Flaws. Health Aff (Millwood). 2026 Jan;45(1):6-13. doi: 10.1377/hlthaff.2025.00897. PMID: 41494115.

13: Casalino LP, Nicholson S, Gans DN, Hammons T, Morra D, Karrison T, Levinson W. What does it cost physician practices to interact with health insurance plans? Health Aff (Millwood). 2009 Jul-Aug;28(4):w533-43. doi: 10.1377/hlthaff.28.4.w533. Epub 2009 May 14. PMID: 19443477.

14: Butcher  L.  Can legislation save the day for challenges related to prior authorization?   Neurol Today. 2022;22(1):1-25. doi:10.1097/01.NT.0000817608.36002.47