Are you as smart as a 4th grader? Why the prototype‐similarity approach to diagnosis is a step backward for a scientific psychiatry

Drew Westen proposes a prototype-similarity matching approach to diagnosis, in which clinicians classify a patient as having a disorder when the patient's symptoms are judged sufficiently similar to a described prototypical case of the disorder that contains all the standard symptoms. He claims that the prototype approach to diagnosis “better fits the ways humans naturally think and classify”. I will argue that Westen's prototype approach is a step backward in light of research showing that people tend to classify by causal history, not similarity, consistent with psychiatry's goal of etiology-based diagnosis. I will also argue that the prototype approach undermines diagnostic validity by eliminating restraints on false positive diagnoses. Inspired by Wittgenstein's “family resemblance” account of concepts, prototype-similarity theory challenges traditional notions that concepts consist of necessary-and-sufficient criteria for category membership. The theory holds instead that an entity is a member of a category if it is overall more similar to that category's “prototypical” member (e.g., robins for “bird”) than to prototypes of competing categories 1. Prototype-similarity theory was initially hailed as a breakthrough in the psychology of classification. Studies showed that people judge some members that are more similar to typical members as better examples of a category than others (“prototypicality effects”). For example, woodpeckers are better “bird” examples than ostriches due to more closely resembling robins. But ostriches are still birds because they resemble robins more than non-bird prototypes. Prototypicality effects supposedly indicated that category membership itself is dimensional and based on degree of similarity to a prototype. Westen thus equates degree of diagnostic-category membership with degree of similarity to a clinical prototype. However, the prototype-similarity theory's fundamental assumption that prototypicality effects (i.e., degree of being a “good example” of a category and degree of similarity to a prototype) equate with degree of category membership has turned out to be incorrect. For example, the category “even number” is defined by necessary-and-sufficient criteria (“divisible by 2”) that equally apply to all members, but nonetheless this category displays prototypicality effects; subjects judge some even numbers as more similar to the prototype “2” than others and as better examples of even numbers than others. Yet the same subjects, when asked, do not consider there to be degrees of category membership in “even number” 2. Whatever prototypicality effects represent psychologically, they are not equivalent to the way people think about category membership. Research suggests that when people categorize, they are not merely similarity judgers but causal theoreticians who infer shared underlying natures (or “essences”) as determinative of classification. Classic experiments show how similarity judgments and category membership judgments diverge based on causal knowledge. For example, subjects judge a 3-inch-diameter round object to be more similar to a quarter than to a pizza, but more likely to be a pizza than a quarter, understandably given the causal laws governing quarters and pizzas 3. When various-aged subjects are told that a raccoon was altered to appear similar to a skunk by painting it black with a white stripe down its back and adding odor sacs, kindergartners, using an approach analogous to prototype-similarity theory, categorize the creature as a skunk, but 4th graders and older subjects classify it as a raccoon based on the “etiology” of its features 4. Surely clinicians’ diagnoses should be as conceptually sophisticated as 4th graders’ “diagnoses” of the raccoon! A causal-modeling/essentialist understanding of concepts explains why category membership so often deviates from prototype similarity 5. We classify steam, ice, and liquid water together as the substance “water” despite dissimilarity and despite steam and ice not at all resembling the presumed prototype of liquid water, based on shared molecular structure. Whales resemble prototypical fish more than prototypical mammals, yet are mammals based on evolutionary history. There are small round red carrots that look more like cherries, and elongated orange cherries that look more like carrots, all classified based on causal inferences, not superficial similarity. Throughout science and physical medicine, underlying essential properties determine category membership. This is the way we think and classify deliberatively, and this is the approach optimal for scientific progress. In principle, the DSM works this way, with symptom criteria used to infer shared internal dysfunctions responsible for symptoms. Little is yet known about etiology, so at present the DSM relies heavily on syndromal similarity. However, the DSM's descriptive criteria are designed to be transitional until research reveals etiologically distinct disorders among current syndromes. Such causal inferences occur implicitly now, for example, when clinicians classify prototypical depressive symptoms that occur during bereavement as normal and atypical depression as major depression, or adolescent antisocial behavior in the course of avoiding sexual abuse as normal 6. If sheer prototype similarity determined diagnosis, illiteracy and delinquency would likely be forms of dyslexia and conduct disorder, respectively, rather than normal variations. Identifying distinct underlying dysfunctions is what scientific study of mental disorder is mostly about, so understanding diagnosis as inference about shared dysfunction is both how people naturally think conceptually and the optimal basis for scientific development of psychiatry 7. Diagnostic criteria allow for flexible development towards an etiologically based system, whereas prototype-similarity diagnosis freezes diagnosis at the symptom-similarity stage. Couldn't the prototype view simply be extended to encompass etiological considerations? The problem is that etiology often forms a necessary-and-sufficient criterion for standard diagnostic category membership judgments, eliminating the relevance of the degree-of-similarity judgments. The prototype approach would thus be transformed into a criterial approach 8. There is a tension between the fuzziness of disorder/normality boundaries and the need to validly distinguish disorders from normal distress. The DSM addresses this tension via artificially rigidly bounded high-threshold diagnostic criteria that limit false positives, combined with “not otherwise specified” (NOS) categories for subsyndromal conditions that would not be judged disorders based on symptoms alone but in context are judged to be disorders. Westen criticizes the precision of DSM symptom and duration thresholds, but fails to explain how the prototype approach prevents false positives. False positives usually arise because normal distress and mental disorder resemble each other symptomatically. Prototype diagnosis, combining symptom similarity and very vague boundaries, is particularly liable to false positives. The only restraint lies in how the line is drawn between the global judgments “has this disorder” and “has significant features of this disorder” (but not the disorder). How this distinction is validly made based on global similarity judgments remains obscure. Westen's response to the false-positives problem is to dismiss it: “In light of the dearth of research on the treatment implications of clinical versus subthreshold symptoms and of data suggesting that subthreshold variants often produce similar levels of functional impairment, satisficing is not an irrational diagnostic strategy in clinical practice.” By “satisficing”, Westen means “to make a ‘good-enough’ assessment”. According to Westen, because normal/disorder boundaries are uncertain, prototype diagnosis can ignore the syndromal/subsyndromal boundary and thus largely ignore the disorder/normal-distress boundary, and still yield “good-enough” psychodiagnosis. More precise diagnosis using all available decision rules only occurs if “the need arises”. But when does the need arise if not at initial diagnosis and treatment selection? And, who would knowingly entrust their psychological or medical fate to “satisficing” clinicians who prefer more fallible but easier-to-use heuristics to full deliberative analysis, especially when, as Westen notes, the prototype heuristic inherently biases towards certain errors (“encouraging clinicians to match patients to prototypes could make them more likely to gloss over disconfirming data…”)? Couldn't the prototype view simply be modified to encompass sharp cutoffs to prevent false positives? The problem is that adding sharp diagnostic boundaries strikes at the heart of the prototype view's equation of category membership with dimensional similarity judgments. Westen argues that the prototype approach has clinical utility. Whether this is true depends on what you mean by “clinical utility”. All else being equal, clinician-friendliness of criteria makes sense. But clinician-friendliness is secondary to validity and patient utility. False-positive diagnoses are neither good for the patient nor ethically defensible. So all else is not equal, and clinical utility becomes irrelevant.