Bitwise

by David Auerbach

  Machine Psychiatry

  [Freud] has not given a scientific explanation of the ancient myth. What he has done is to propound a new myth.

  —LUDWIG WITTGENSTEIN

  The discrete taxonomies of the DSM-III were quite amenable to computers, and the desire to increase quantification of mental illness guided (and misguided) the latest revision, the DSM-5. Frances, in his book Saving Normal, describes trying to refine the existing categories in the DSM without letting special interests corrupt them.

  We saw DSM-IV as a guidebook, not a bible—a collection of temporarily useful diagnostic constructs, not a catalog of “real” diseases. We tried to make this abundantly clear in the introduction to DSM-IV and at greater length in the DSM-IV Guidebook. Unfortunately, I am not sure anyone ever reads the introduction, and I know that few people have read the Guidebook.

  If the DSM is not a catalogue of real diseases, you wouldn’t know that from the extent to which categories like bipolar and borderline are thrown around clinically and the extent to which they are researched as though they are real. The computers processing those DSM codes certainly don’t know whether they’re real.

  A 2012 Canadian Medical Association Journal study showed that within a grade, schoolkids were far more likely to be diagnosed with ADHD merely by being younger. As the study put it: “Boys who were born in December were 30% more likely to have a diagnosis of ADHD than boys born in January, and girls born in December were 70% more likely to have a diagnosis of ADHD than girls born in January.” As family therapist Joan Lipuscek observes, immaturity was being misdiagnosed as ADHD.

  If childhood bipolar is arguably not a real diagnosis, what’s to say that ADHD is the right rubric for what may be multiple ailments or even no ailment at all? What is the reason for sticking with current DSM categories rather than the alternative paradigms proposed by DSM opponents like Edward Shorter? Are mania, depression, and bipolar three separate things, or all aspects of one overarching mood disorder, which Shorter terms Kraepelin’s disease, or “melancholic syndrome”? How do we even adjudicate such issues, in the absence of decisive neurobiological evidence?

  Frances advocated a conservative approach, believing that changes to the DSM would exacerbate diagnostic ambiguity by inflating the number of diagnoses without helping to ground them more firmly in neurobiology. Frances’s advice, then, is to diagnose less and treat less, using the DSM provisionally: “The right goal for DSM-5 would have been diagnostic restraint and deflation, not a further unwarranted expansion of diagnosis and treatment.” In this, however, he was at war with the DSM-5 leaders who both found the existing categories inadequate and had the ambition to set things right. And this is where computers enter the DSM story.

  Computers were a driving force behind this latest, troubled revision of the DSM. From the DSM-5’s conception in 1999, the American Psychiatric Association’s DSM-5 architects, led by chairman Darrel Regier, had grand ambitions to adopt new dimensional measures of diagnosis instead of categorical ones. Diagnoses were to be graded on a sliding scale (How depressed are you? How severe is your schizophrenia?) rather than by yes-or-no criteria. The DSM-5 team, including Regier and Alan Schatzberg, rebutted an objection by Frances by appealing to a 2009 article co-authored by Regier:

  As documented in the recent American Journal of Psychiatry (AJP) article, the use of dimensional assessments to reconceptualize psychopathology represents the most practical and evidence-based way of moving our field forward.

  That 2009 article shares two of the same authors and promotes dimensional measures without providing any evidence that they’re more successful:

  Our immediate task is to set a framework for an evolution of our diagnostic system that can advance our clinical practice and facilitate ongoing testing of the diagnostic criteria that are intended to be scientific hypotheses, rather than inerrant Biblical scripture. The single most important precondition for moving forward to improve the clinical and scientific utility of DSM-V will be the incorporation of simple dimensional measures for assessing syndromes within broad diagnostic categories and supraordinate dimensions that cross current diagnostic boundaries.

  In plain English, this passage claims dimensional measures will be a crucial improvement to the DSM.*13 The justification for dimensional measures tended to center on a 2006 paper coauthored by another DSM-5 architect, Helena Kraemer, a biostatistician in Stanford’s department of psychiatry whose background is in statistics rather than medicine. As the dimensional measures mostly did not make it into the DSM-5, I won’t relitigate this dispute. What interests me more is that the implicit rationale for this shift to dimensional measures was increased quantitative specificity. Kraemer sought to make the diagnostic process as predictable and standardized as possible. To her, dimensional measures offer

  a new set of advantages such as greater statistical power, improved predictive validity, more focused treatments, and new opportunities for genetic and other etiological research.

  These “advantages” share several characteristics. All of them restrict the individual initiative of a physician. All appeal to harder science than psychiatry has offered to date. And all of them require computers. The statistical analyses and research opportunities Kraemer cites all require the collection of data in order to assess the reliability and utility of treatments. The goal is to increase the granularity of wide-scale (and narrower-scale) data collection, so that uncontrolled field studies can be performed to a far greater degree of precision, supplementing controlled studies. Kraemer bemoans the “forfeit of clinical information” caused by the existing DSM categories:

  Perhaps the greatest disadvantage of a categorical system is the limitations, both clinical and statistical, imposed by the forfeit of clinical information inherent in labeling patients based solely on whether their signs and symptoms collectively rise above a defined threshold.

  To get a grasp on an individual patient, one may need to read a physician’s notes or even speak to him or her. But for Kraemer, “clinical information” is only what can be represented by the DSM categories. It is purely quantitative. Kraemer focuses on the transcription of patient data into a computer, even offering a vision of clinical assessment by computer:

  Consistency in the collection of clinical and epidemiological data would be enhanced if structured interviews and/or questionnaires were developed and offered as part of the DSM-V. Patient self-administered questionnaires could be developed to gather relevant symptom data. Such efforts would offer even greater benefit if users had the choice of paper or computer administration. There is considerable evidence that responses to computerized interviews are more candid than face-to-face responses.

  Kraemer’s goal, then, is the quantification of the diagnostic process itself so that diagnosis could be possible even in the absence of a human physician. Diagnosis becomes another version of a standardized test one takes in school. The benefits are uniformity of data being collected and removal of potential bias in the physician. Unfortunately, this isn’t possible. Kraemer speaks of “a direct, dimensional reflection of the categorical definition that could be used for genetic and other analyses to increase statistical power.” But if the categories are bunk, there’s no way that chopping them into spectra is going to make them better. The categories will only become more arbitrary and detached from the everyday realities that spawned them. Psychotherapist Gary Greenberg’s account of the dismal and sometimes comical results of the DSM-5 task forces, where the members often could not agree on what it was they were trying to accomplish, shows this detachment stymieing the participants repeatedly.

  The DSM-5 project did not break down just because it was a cultural construct. We need cultural constructs to organize our lives, and we inevitably create them in their absence. Rather, the DSM-5 couldn’t satisfy several competing interests at the same time. The DSM today is used for physician diagnosis, act
uarial insurance practices, longitudinal research studies, drug regulation, and more. Robert Spitzer and his colleagues architected the DSM-III with one primary purpose in mind: standardizing diagnoses across physicians. Yet because its classifications are so provisional, the DSM’s comparative success at that goal doesn’t mean that it was or is suited for the many other purposes to which it is being put. The differing interests and rationales make consensus impossible. There will be a tug-of-war between people trying to pull these provisional categories in different directions to suit differing agendas, and no way to claim that one direction is “right,” because there is no agreed-upon truth to measure a classification against. The DSM has passed its expiration date not because it is intrinsically nefarious, but because the context in which it was created no longer exists, having been replaced with newer models of psychiatry, and it is failing to meet the requirements currently being put upon it.

  The National Institute of Mental Health is abandoning the DSM as a diagnostic taxonomy in favor of the less politically loaded and less specific categories of psychiatric disorders listed in the International Classification of Diseases—a classification owned and managed by the United Nations’ World Health Organization rather than the American Psychiatric Association. In 2013, NIMH head Michael Insel wrote of the decision:

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

  The diagnostic system has to be based on the emerging research data, not on the current symptom-based categories.

  To Gary Greenberg, Insel was even more frank: “There’s no reality. These [disorders] are just constructs. There’s no reality to schizophrenia or depression….Whatever we’ve been doing for five decades, it ain’t working.” And yet even while recognizing this, Insel did not toss out taxonomies entirely, but turned to what he thought of as a better provisional taxonomy, the ICD.

  Without scientific consensus on the nature of mental illness, we cannot do without folk taxonomies like the DSM or the ICD. But we must recognize, as Frances says, both their limitations and their dangers. Given the increasing quantification of every aspect of daily life, we will suffer similar difficulties considerably more often. Even if the DSM is recognized as obsolete, terabytes of medical records containing DSM-III, -IV, and -5 codes remain, ready for research and actuarial purposes. Computers make it so that any taxonomy like the DSM will hang on far longer than we may wish it to. Computers reify existing taxonomies, making them harder to displace, while simultaneously making them appear more adequate—and accurate—than they actually are.

  *1 Other strong proponents include psychologists John Digman and Lewis Goldberg.

  *2 “Ashton’s model basically divides FFM A[greeableness] into two factors, the second called Honesty-Humility.”

  *3 Anyone who ever owned one of the classic Epson printers will remember how loud they were, the banshee screech of each printed line coming out like laser fire. A line of underscores (______) had a thinner, more monotone sound than the high-pitched chatter of letters and numbers. The English composer, improviser, and music maker Hugh Davies took advantage of the sonic variability to construct dot-matrix printer compositions in the 1980s.

  *4 Some remain with me via muscle memory: F5 brought up the file system interface. F10 saved the current document. F7 (followed by an “N” and a “Y”) exited the program. F2 was the spell-checker. There was a thin, long strip of function key shortcuts that came with WordPerfect and sat above the function key row of our keyboard for years.

  *5 In fact, I’d learned “anxiety” from Peanuts, specifically from Charlie Brown’s line “My anxieties have anxiety.” Lucy came to symbolize many Americans’ ambivalent image of psychiatry for many years. My father, however, bemoaned how poorly psychiatrists were portrayed in popular culture, from Dr. Mabuse to Hannibal Lecter.

  *6 The term of art for classification of diseases is “nosology,” which I see no need to use any more than one needs to use the correct collective noun for every sort of bird. The difference between “a murder of crows” and “a group of crows” is the difference between an in-group and an out-group’s vocabulary. So too with the difference between “nosology” and “taxonomy” or “classification.”

  *7 A label is not a fixed abbreviated description. A label has many conflicting definitions, each of which is no more than a grouping of overlapping labels itself, and there is no adjudicating process for determining which one applies in a particular circumstance. Labels are extremely fuzzy and circularly interrelated entities. There is little that cannot be thrown into confusion by taking a particular term, whether it’s “borderline” or “mania” or even “sick,” and asking, “What do you mean by that?”

  *8 The DSM-5 reorders the symptoms to put the two fears last, but otherwise leaves the list from the DSM-IV-TR as is. It also puts “going crazy” into quotes. It adds, “Culture-specific symptoms (e.g., tinnitus, neck soreness, headache, uncontrollable screaming or crying) may be seen. Such symptoms should not count as one of the four required symptoms.” I don’t see how those symptoms are culture-specific, but I am not a psychiatrist.

  *9 What is the difference between a sensation and an emotion? French conflates the two in the word sensibilité, which is sensation rather than sensibility per se, and since everything is neurochemistry anyway, the distinction appears to be more by convention than anything else. Is pain an emotion? I can distinguish the heavy, pillowy weight of sadness and despair from the spiky intrusion of guilt, but both are a great distance from the malignant, irresistible, and nonnegotiable steamroller assault of biochemical depression.

  *10 I had spent some time reading about the Buddhist philosopher Nagarjuna and his philosophy of emptiness, and the Buddhists’ radically contingent view of phenomenal experience came as a great aid to me by helping me stand at a remove from the bizarre things my brain and body were doing to me. It also had the less salutary effect of teaching me how tenuously each of us is connected to an ordinary, manageable experience of the world, such that we stand, by the grace of our bodies, only a few steps away from unbearable agony or madness. David B.’s graphic novel Epileptic is a harrowing and hermetic portrayal of that thread of coherence being severed. The psychologist Louis Sass has written perceptively on these issues, and the distortions through which language puts them, in The Paradoxes of Delusion: Wittgenstein, Schreber, and the Schizophrenic Mind and Madness and Modernism: Insanity in the Light of Modern Art, Literature, and Thought.

  *11 The brief history of the DSM I give here draws heavily on the writings of Donald Goodwin, Samuel Guze, Allen Frances, Gary Greenberg, Michael Alan Taylor, Edward Shorter, and David Healy. With the exceptions of Goodwin and Guze, who anticipated and influenced the DSM-III, and Frances, who was the architect of the DSM-IV before becoming one of the preeminent critics of the DSM-5 (alongside Robert Spitzer himself), they all think poorly of the DSM. In pursuit of balance, I read the defenses of the DSM methodology offered by people such as DSM-5 architect Darrel Regier, William Carpenter, and Helena Kraemer. To my inexpert eye, they mostly fail to rebut the most serious charges of their critics, namely with regard to the fundamental lack of validity of the existing DSM taxonomy. I do not mean to litigate that dispute. Disputes on whether medications are being overprescribed will not be solved by appeals to unsound taxonomies—or by attacks on them. If a patient feels better after taking Prozac, the quest
ion is not whether he or she is mentally ill, but whether the prescription is responsible and safe. Since historical accounts of the DSM have been authored overwhelmingly by its critics, I have attempted to correct for that by avoiding the more contentious and subjective aspects of their criticisms where possible.

  *12 New York is the only city I have lived in where people talk openly about their therapists and their relationships with them.

  *13 The numbing verbiage of the leaders of the DSM-5 contrasts to the considerably more forthright writings of their predecessors. A similar apologia by Regier’s colleague William Carpenter similarly masks a paucity of content in vague obfuscation: “The challenge for DSM-V is whether an approach to domains of pathology can supplement syndrome diagnostic categories. This does introduce a new paradigm—domains of pathology—and the scientific field is moving rapidly to deconstruct heterogeneous syndromes.” The use of the word “deconstruct,” signaling a heady revisionary mind-set, is almost always a red flag—the presence of “paradigm” doubly so.

  6

  GAMES COMPUTERS PLAY

  Dungeons and Dice

  What is it that you see when you see? You see an object as a key, a man in a car as a passenger, some sheets of paper as a book. It is the word ‘as’ that must be mathematically formalized, on a par with the connectives “and,” “or,” “implies,” and “not.”…Until you do that, you will not get very far with your AI problem.

  —STANISLAW ULAM