It can fairly be said that modern psychiatric diagnosis was “born” in a 1970 paper on schizophrenia.
The authors, Washington University psychiatry professors Eli Robins and Samuel B. Guze, rejected the murky psychoanalytic diagnostic formulations of their time. Instead, they embraced a medical model inspired by the careful 19th-century observational work of Emil Kraepelin, long overlooked during the mid-20th-century dominance of Freudian theory. Mental disorders were now to be seen as distinct categories, much as different bacterial and viral infections produce characteristic diseases that can be seen as distinct “natural kinds.”
Disorders, Robins and Guze argued, should be defined based on phenomenology: clinical descriptions validated by long-term follow-up to demonstrate the stability of the diagnosis over time. With scientific progress, they expected fuller validation of mental disorders to derive from laboratory findings and studies of familial transmission.
This descriptive approach to psychiatric diagnosis -- based on lists of symptoms, their timing of onset, and the duration of illness -- undergirded the American Psychiatric Association’s widely disseminated and highly influential Diagnostic and Statistical Manual of Mental Disorders, first published in 1980. Since then, the original “DSM-III” has yielded two relatively conservative revisions, and right now, the DSM-5 is under construction. Sadly, it is clear that the optimistic predictions of Robins and Guze have not been realized.
Four decades after their seminal paper, there are still no widely validated laboratory tests for any common mental illness. Worse, an enormous number of family and genetic studies have not only failed to validate the major DSM disorders as natural kinds, but instead have suggested that they are more akin to chimaeras. Unfortunately for the multitudes stricken with mental illness, the brain has not given up its secrets easily.
That is not to say that we have made no progress. DNA research has begun to illuminate the complex genetics of mental illness. But what it tells us, I would argue, is that, at least for the purposes of research, the current DSM diagnoses do not work. They are too narrow, too rigid, altogether too limited. Reorganization of the DSM is hardly a panacea, but science cannot thrive if investigators are forced into a cognitive straitjacket.
Before turning to the scientific evidence of fundamental problems with the DSM, let’s first take note of an important problem that the classification has produced for clinicians and patients alike: An individual who receives a single DSM diagnosis very often meets criteria for multiple additional diagnoses (so-called co-occurrence or “comorbidity”), and the pattern of diagnoses often changes over the lifespan. Thus, for example, children and adolescents with a diagnosis of an anxiety disorder often manifest major depression in their later teens or twenties. Individuals with autism spectrum disorders often receive additional diagnoses of attention deficit hyperactivity disorder, obsessive-compulsive disorder, and tic disorders.
Of course, there are perfectly reasonable explanations for comorbidity. One disorder could be a risk factor for another just as tobacco smoking is a risk factor for lung cancer. Alternatively, common diseases in a population could co-occur at random. The problem with the DSM is that many diagnoses co-occur at frequencies far higher than predicted by their population prevalence, and the timing of co-occurrence suggests that one disorder is not likely to be causing the second. For patients, it can be confusing and demoralizing to receive multiple and shifting diagnoses; this phenomenon certainly does not increase confidence in their caregivers.
Family studies and genetics shed light on the apparently high rate of co-occurrence of mental disorders and suggest that it is an artifact of the DSM itself. Genetic studies focused on finding variations in DNA sequences associated with mental disorders have repeatedly found shared genetic risks for both schizophrenia and bipolar disorder. Other studies have found different sequence variations within the same genes to be associated with schizophrenia and autism spectrum disorders.
An older methodology, the study of twins, continues to provide important insight into this muddy genetic picture. Twin studies generally compare the concordance for a disease or other trait within monozygotic twin pairs, who share 100% of their DNA, versus concordance within dizygotic twin pairs, who share on average 50% of their DNA. In a recent article in the American Journal of Psychiatry, a Swedish team of researchers led by Paul Lichtenstein studied 7,982 twin pairs. They found a heritability of 80% for autism spectrum disorders, but also found substantial sharing of genetic risk factors among autism, attention deficit hyperactivity disorder, developmental coordination disorder, tic disorders, and learning disorders.
In another recent article in the American Journal of Psychiatry, Marina Bornovalova and her University of Minnesota colleagues studied 1,069 pairs of 11-year-old twins and their biological parents. They found that parent-child resemblance was accounted for by shared genetic risk factors: in parents, they gave rise to conduct disorder, adult antisocial behavior, alcohol dependence, and drug dependence; in the 11-year-olds these shared factors were manifest as attention deficit hyperactivity disorder, conduct disorder, and oppositional-defiant disorder. (Strikingly, attention deficit disorder co-occurs in both the autism spectrum cluster and disruptive disorder cluster.)
These and many other studies call into question two of the key validators of descriptive psychiatry championed by Robins and Guze. First, DSM disorders do not breed true. What is transmitted across generations is not discrete DSM categories but, perhaps, complex patterns of risk that may manifest as one or more DSM disorders within a related cluster. Second, instead of long-term stability, symptom patterns often change over the life course, producing not only multiple co-occurring diagnoses but also different diagnoses at different times of life.
How can these assertions be explained? In fairness to Robins and Guze, they could not have imagined the extraordinary genetic complexity that produces the risk of many common human ills, including mental disorders. What this means is that common mental disorders appear to be due to different combinations of genes in different families, acting in combination with epigenetics -- gene expression varies even if the underlying DNA sequence is the same -- and non-genetic factors.
In some families, genetic risk for mental disorders seems to be due to many, perhaps hundreds, of small variations in DNA sequence -- often single “letters” in the DNA code. Each may cause a very small increment in risk, but, in infelicitous combinations, can lead to illness. In other families, there may be background genetic risk, but the coup de grace arrives in the form of a relatively large DNA deletion, duplication, or rearrangement. Such “copy number variants” may occur de novo in apparently sporadic cases of schizophrenia or autism.
In sum, it appears that no gene is either necessary or sufficient for risk of a common mental disorder. Finally, a given set of genetic risks may produce different symptoms depending on broad genetic background, early developmental influences, life stage, or diverse environmental factors.
The complex nature of genetic risk offers a possible explanation for comorbidity: what the DSM treats as discrete disorders, categorically separate from health and from each other, are not, in fact, discrete. Instead, schizophrenia, autism-spectrum disorders, certain anxiety disorders, obsessive-compulsive disorder, attention deficit hyperactivity disorder, mood disorders, and others represent families of related disorders with heterogeneous genetic risk factors underlying them. I would hypothesize that what is shared within disorder families, such as the autism spectrum or the obsessive-compulsive disorder spectrum, are abnormalities in neural circuits that underlie different aspects of brain function, from cognition to emotion to behavioral control, and that these circuit abnormalities do not respect the narrow symptoms checklists within the DSM.
The first DSM had many important strengths, but I would argue that part of what went wrong with it was a fairly arbitrary decision: the promulgation of a large number of disorders, despite the early state of the science, and the conceptualization of each disorder as a distinct category. That decision eschewed the possibility that some diagnoses are better represented in terms of quantifiable dimensions, much like the diagnoses of hypertension and diabetes, which are based on measurements on numerical scales.
These fundamental missteps would not have proven so problematic but for the human tendency to treat anything with a name as if it is real. Thus, a scientifically pioneering diagnostic system that should have been treated as a set of testable hypotheses was instead virtually set in stone. DSM categories play a controlling role in clinical communication, insurance reimbursement, regulatory approval of new treatments, grant reviews, and editorial policies of journals. As I have argued elsewhere, the excessive reliance on DSM categories, which are poor mirrors of nature, has limited the scope and thus the utility of scientific questions that could be asked. We now face a knotty problem: how to facilitate science so that DSM-6 does not emerge a decade or two from now a trivially revised descendant of DSM-III, but without disrupting the substantial clinical and administrative uses to which the DSM system is put.
I believe that the most plausible mechanism for repairing this plane while it is still flying is to give new attention to overarching families of disorders, sometimes called meta-structure. In previous editions of the DSM, the chapters were almost an afterthought compared with the individual disorders. It should be possible, without changing the criteria for specific diagnoses, to create chapters of disorders that co-occur at very high rates and that appear to share genetic risk factors based on family, twin, and molecular genetic studies.
This will not be possible for the entire DSM-5, but it would be possible for certain neurodevelopmental disorders, anxiety disorders, the obsessive-compulsive disorder spectrum, so-called externalizing or disruptive disorders (such as antisocial personality disorder and substance use disorders), and others. Scientists could then be invited by funding agencies and journals to be agnostic to the internal divisions within each large cluster, to ignore the over-narrow diagnostic categories. The resulting data could then yield a very different classification by the time the DSM-6 arrives.
Psychiatry has been overly optimistic about progress before, but I would predict that neurobiologically based biomarkers and other objective tests will emerge from current research, along with a greater appreciation of the role of neural circuits in the origins of mental disorders. I would also predict that discrete categories will give way, where appropriate, to quantifiable dimensions. At the very least, the science of mental disorders should be freed from the unintended cognitive shackles bequeathed by the DSM-III experiment.