In 1973, Akiskal and McKinney published an article in Science(1) that was so influential as to result in an extended version subsequently being published in the Archives of General Psychiatry(2). In both articles, the authors had two key objectives: first, to integrate a large number of disparate conceptual models of depression and second, and as a logical extension, to propose a "unified hypothesis" with "the depressive syndrome…conceived as the psychobiological final common pathway" (p. 286). At that time (1975), challenges to the longstanding North American unitarian view were appearing, but the parsimony of the "final common pathway" proposal was so attractive that it reasserted the unitarian model of depression within North America, cementing it shortly thereafter in the DSM-III depression paradigm that continues to dominate clinical practice and research.

Classification of the depressive disorders has long been contentious, with three principal models argued for on the basis of the presumed number of types—one, two, and many. The binary model posited two principal types (i.e., "endogenous/psychotic" and "neurotic/reactive") and is somewhat predictable, for as the humorist Benchley observed, there are two classes of people in the world: those who divide the people in the world into two classes and those who do not. Nevertheless, it had its influential proponents, including St. Paul, who, according to Altschule (3), distinguished between two types of depression: one "from God" and the other "of the world."

In 1926, however, British psychiatrist Mapother argued (see reference 4) that the long-standing binary distinction made in clinical practice was pointless because both "psychotic" and "neurotic" forms of depression lie along a continuum; he so espoused the unitarian position (i.e., there is only one type of depression, which varies by severity).

Influential British academics engaged in the unitary/binary debate for decades without clear resolution. The failure by Lewis (5) to find any clear demarcation in his clinical observation study (undertaken in the early 1930s and before the introduction of multivariate statistical approaches) was highly influential in cementing the unitarian paradigm. Subsequently, in the 1960s, members of the so-called Newcastle school argued strongly (e.g., reference 6) that their use of multivariate analyses supported the binary view. However, their analytic tool (principal components analysis) effectively derives dimensions rather than symptom groups and may have distinguished "depressive" from "anxiety" dimensions, as the removal of anxiety items from such data sets effectively changes the "bipolar factor" held to define the contrasting depressive types (7). The debate also became highly technical, with Kendell (8), for instance, arguing that if two types of depressive disorders exist, a discriminant function analysis plot should demonstrate a bimodal rather than unimodal distribution of scores—but that few data sets showed any such pattern. Although Kendell was unpersuaded by a binary model, Eysenck (9), by contrast, stated that multivariate analyses allowed a clear conclusion ("the unitary hypothesis is wrong, and the binary hypothesis is supported very strongly indeed" [p. 242]).

At the same time, North American writers also grappled with the competing models. Klerman (10) noted that the introduction of effective antidepressant therapies had resulted in many clinicians reporting that patients with "the endogenous symptoms patterns respond better to somatic therapies…than do psychoneurotic patients" (p. 311) with subsequent research findings weakening the unitary position, which, "derived from Meyer and Freud, had been the dominant American and British approach to phenomenology" (p. 312). Although Klerman explored the concept of "endogenous depression" extensively, he judged its status as "unsettled," being either "a well-defined group, clearly separated from other groups, or merely an issue of gradation on multiple dimensions" (p. 315).

Klein (11), by contrast, observed:

He held that any final common pathway model might "blur data derived from an underlying categorical situation into an approximation of a dimensional situation" (p. 447). He was one of the few to immediately challenge the final common pathway model, concluding firmly that the unitarian position could not be maintained. He observed, however, that this did not, ipso facto, make the binary viewpoint correct ("It is possible to adopt a trinitarian or even more pluralistic stand" [p. 447]). His conceptual subtyping paradigm was weighted toward both phenomenological and treatment response distinctions. He stated that an "endogenomorphic group" was categorically distinct on phenomenological grounds because of an "inhibited pleasure mechanism." In regards to treatment differentiation, he hypothesized that endogenomorphic depressive patients would show a delayed but superior response to a tricyclic antidepressant, whereas those with "chronic neurotic depression" would show a poor response—and both would show a minimal placebo response rate. By contrast, those in the third group (with "reactive depression") would show a high response rate to both antidepressant and placebo.

Looking back, there were three issues that prevented clear resolution of the unitary/binary debate. First, the either/or forced-dichotomy problem, which failed to consider one of life’s truisms: there is always (in the least) a third option. Here, and as addressed by Klein (11), the third option effectively concedes multiple depressive disorders (ranging variably in majesty or in salience to the clinician and researcher). It is regrettable that the importance of the indicative studies in identifying meaningful subgroups is more evident in retrospect. For instance, Paykel (12) reported several cluster analyses involving a wisely selected heterogeneous sample of depressed patients. His four-cluster solution essentially identified 1) a psychotic group with features held to be typical of endogenous depression, 2) anxious and "neurotic" patients, who scored high on neuroticism measures, 3) younger, hostile depressed patients, and 4) younger patients with a personality disorder. It is regrettable that that study—as in many subsequent ones using cluster analysis or other analytic techniques capable of identifying groups or classes rather than dimensions—encouraged few immediate replication studies. Researchers lacked confidence in such statistical approaches (e.g., arguing that cluster analysis will always generate clusters and that, as the decision ruling for the "true" number of meaningful clusters remains problematic, applied solutions should be doubted).

A second—and ongoing—issue emerges from a number of confounding sources, with only several noted here. Assume that there are, for example, three principal clinically meaningful groups with depressive disorder (i.e., psychotic, melancholic, and a heterogeneous residue of nonmelancholic disorders) with frequencies that vary significantly across samples selected from the community, general practice, general hospital and liaison units, psychiatric outpatient and inpatient units, as well as tertiary psychiatric facilities dealing with treatment-resistant patients. Any attempt to distinguish such types is influenced by their quite varying site prevalences, with the first two even nonexistent in a number of settings. In addition, the "white noise" of any associated comorbid condition—especially personality disorder—will contaminate any capacity for pristine and readily replicable solutions unless, as detailed later, the analytic strategy encompasses both axis I and axis II features.

The third issue is perhaps the most fundamental. "Types" are best distinguished by the identification of unique, as against merely overrepresented, features. Thus, any attempt to distinguish cars by tire size is unlikely to be helpful. Similarly, we (13) have demonstrated that most of the so-called "endogeneity symptoms" historically examined to distinguish melancholic and nonmelancholic depression are shared across the depressive types, with variation in prevalence largely due to variation in depression severity. If melancholic depression is, in general terms, more severe than nonmelancholic depression—as is generally conceded—those with melancholia are more likely to affirm or to return more "severe" scores on such variables. Clear differentiation of any "true" melancholic and nonmelancholic subtypes is not, however, possible on such a secondary (i.e., severity) basis. The DSM-IV criteria for major depression with melancholia are not immune to this concern. The key criteria (e.g., consummatory anhedonia and nonreactive mood), as well as symptom reports of psychomotor disturbance (e.g., feeling slowed down), are virtually ubiquitous in clinical samples of significantly depressed patients (13)—in themselves rebutting the view that they have unique specificity to melancholia.

Concerns about DSM criteria sets are not new. Although the loose DSM-II definition of "endogenous depression" (i.e., the earlier term for melancholia) allowed prediction of treatment response, Zimmerman and colleagues (14) noted a paradox some years after the introduction of DSM-III: "all published studies examining the relationship between (DSM-III) melancholic subtyping and response to antidepressant medications or electroconvulsive therapy were negative" (p. 361). The relevant DSM-III-R committee addressed this problem by deleting two problematic items (the concatenated descriptor "excessive or inappropriate guilt" and the negatively defined "distinct quality" descriptor). In return, three nonsymptom items held to reflect some of the clinical givens about melancholia (i.e., no significant premorbid personality disturbance, previous episodes followed by recovery, and previous good response to antidepressant drugs and ECT) were added. Inclusion of the latter would clearly improve treatment prediction, but, although scientifically valid, it is a pseudoprofound criterion. The relevant DSM-IV committee revisited the question and essentially returned to the DSM-III criteria set. The consequences are that the DSM-III and DSM-IV criteria sets have failed to generate any consistent neurobiological findings or to demonstrate any treatment specificity for DSM-distinguished melancholic and nonmelancholic depression.

If melancholic and nonmelancholic depression are differing categorical types, we require delineation by features that are either distinctly overrepresented or, preferably, unique to one type or the other. We (13) have argued that there exists one clear specificity candidate—observable (and not merely reported) psychomotor disturbance—whereas other endogeneity features are prevalent in both melancholic and nonmelancholic subjects and are thus nonspecific. The proposition is not new, with Jackson (15) noting the definitional importance of psychomotor disturbance to melancholia from early Roman and Greek times to early in the twentieth century, while Berrios (16) has informed us that in classical antiquity, melancholia was defined in terms of overt behavioral features and that "symptoms…were not part of the concept" (p. 298). The remarkable stereotypic presentation of psychomotor disturbance across cultures and races attests to its clinical utility, in as much as a neurologist would consider Parkinson’s disease on observing the signs of a mask-like face and a resting hand tremor.

There were other pointers to the potential utility of considering psychomotor disturbance as a surface marker rather than as a symptom. Lewis’ failure (5) to demonstrate separate depressive categories in his clinical sample was noted previously. Kiloh and Garside (17) subsequently coded Lewis’ data set to undertake multivariate analyses, and we (18) subsequently reanalyzed the Kiloh-coded data set. Our mixture analysis, including numerous historically suggested endogeneity variables, produced a clear unimodal distribution of summed scores, arguing for only one type of depression. When, however, we limited the analysis to a refined set of distinguishing features (including observable psychomotor disturbance), a striking bimodal distribution of scores was evident, arguing against the presence of a single type of depression in the Lewis sample. Klein (11) foreshadowed such a result when, considering the long-standing failure to demonstrate bimodality, he stated that if analyses were "limited to those scale items that only apply to endogenomorphic patients, bimodality may be evident" (p. 450). We (18) suggested that melancholic and nonmelancholic depression may only be distinguishable on the basis of a limited set of clinical features (such as psychomotor disturbance) but that the inclusion of a large number of "other" nondifferentiating variables (the common strategy) could swamp the capacity of psychomotor disturbance and any other truly differentiating variables to demonstrate those separate categories, and thus occasion a unimodal distribution.

We therefore developed the CORE measure (13), an operationalized 18-item observer-rated system assessing cognitive processing problems, as well as agitation and retardation. Scores on that measure (in comparison to symptom reports of psychomotor disturbance) indicated quite marked specificity of "substantial" observable psychomotor disturbance being present in melancholic depression and absent in nonmelancholic depression—and thus in sharp contrast with the more ubiquitous "endogeneity symptoms." "Substantial" was quantified in our study as a CORE score of 8 or more, and validation was supported by examination against a range of sociodemographic, psychosocial, and biological variables, as well as by demonstrating a strong capacity to predict ECT response. CORE scores can also be viewed dimensionally, either as quantifying the severity of psychomotor disturbance or by providing probability estimates of melancholia. In a validation study involving dexamethasone suppression testing (DST) of 100 depressed patients (19), DST nonsuppression rates rose linearly (from 30% to 90%) as CORE scores increased, supporting the latter point. In comparison to other measures of melancholia (e.g., DSM-III, DSM-III-R, Newcastle), CORE scores have shown superior discrimination across a range of clinical givens or ascriptions to melancholia (13), including illness variables (e.g., older age at onset), psychosocial risk factors (e.g., decreased prevalence of personality disorder, lower exposure rates to deprivational experiences in childhood), neuropsychological testing (e.g., higher rates of cognitive processing problems), and treatment response (e.g., strong prediction of ECT response).

Thus, we regard observable psychomotor disturbance (as defined) as the surface (or recordable) marker of an underlying neuropathological process so allowing identification of a neurobiologically discrete depressive subtype—i.e., melancholia. As a marker, it points to the likely site of a neurobiological lesion or perturbation. Our formulation (20) respects the neural circuit model detailed in the mid-1980s by Alexander et al. (21, 22). The authors have demonstrated in primate studies that if the circuit connecting the basal ganglia and the prefrontal cortex is disrupted, three consequences result: psychomotor disturbance, cognitive impairment, and depression. Our model (20) posits two contrasting mechanisms. In those who develop depression with psychomotor disturbance at a young age, there is usually a strong family history of depression, structural imaging studies rarely reveal an abnormality, and patients usually respond in a straightforward manner to antidepressant drugs and ECT. Here we assume that stress precipitates a physiological disruption of those circuits in subjects who have a genetic diathesis to melancholia. By contrast, in those who first develop depression with psychomotor disturbance in later life, a family history of depression is less common, whereas subjects are more likely to have a family history of and/or increased risk factors for cerebrovascular disorders. Here, the neural circuit is more likely to be structurally disrupted (i.e., the caudate and putamen tend to be reduced in volume, and there are hyperintensities indicative of microvascular damage disrupting the circuits), in addition to any functional disturbance. Such patients tend to respond less well to physical treatments and may develop delirium in response to a standard dose of a tricyclic or with ECT, and their depression may presage subsequent dementia.

The status of psychotic depression (i.e., as a separate type of depression or as a subtype of melancholia) is less clear. This mood disturbance generally appears severe to observers but may be denied or minimized by such patients. Most classic endogeneity features of melancholia are present, but some (e.g., diurnal variation of mood and energy) may be replaced by persistent and unvarying depression and anergia throughout the day as the condition worsens. Psychomotor disturbance (as quantified by the CORE measure) is distinctly more severe than in nonpsychotic melancholia (13) and is so severe in some—especially mute patients—that actual psychotic features may be denied or unacknowledged. Although severe psychomotor disturbance (or very high CORE scores) can provide a proxy measure of psychotic depression, it is the specific presence of psychotic features (i.e., delusions and hallucinations) that distinguish this category.

Although the dimensional view of depression argues that "there are no sharp qualitative distinctions of a categorical nature" (11), the presence of two specific and categorical features (psychomotor disturbance and psychotic features) logically allows a tiered hierarchical model (F1) to be operationalized. Here, all three features (psychotic, melancholic, and residual nonmelancholic depression) share a mood disorder component, for they are quintessentially "depressive" disorders. Overall, the mood disorder is most severe in psychotic and least severe in nonmelancholic depression. But, as it is common to all three categories, it is a weak tool for differentiating the conditions, although it is relied on by many current indices of melancholia. Thus, we have reason for the limited utility of the DSM-III and DSM-IV definitions of "melancholia" and the ICD-10 model.

Proceeding from nonmelancholic to melancholic depression, our contrast model ignores any greater mood state severity and argues for the necessity of a feature that is absent in nonmelancholic depression (i.e., observable psychomotor disturbance). Proceeding from melancholic to psychotic depression, psychomotor disturbance and the observed mood state are more severe overall, but these features (here reflecting dimensional differences) are minimized in model-derived diagnostic decisions that weight the specific presence of psychotic features. It is of interest that those with bipolar disorder have been reported to be distinctly more likely to have melancholic and psychotic expressions of depression when in a depressed phase (23, p. 43); this is also clearly evident in our data sets (24).

The tripartite model itself does not extend to subdividing the presumed heterogeneous residue of nonmelancholic depression, for this class possesses only the shared (and nondifferentiating) depressed mood component. The options, then, for developing a clinical algorithm might be to 1) regard the tripartite model as sufficient in and of itself, 2) search for unique clinical markers of separate nonmelancholic disorders, or 3) extend the model to include the nonmelancholic class but, of necessity, adopt a differing conceptual paradigm for operational development. The latter might involve a dimensional approach, with perhaps severity-derived groupings of major depression, minor depression, and even subsyndromal depression. Alternately, one could attempt to circumscribe categories on the basis of clinical observations—for instance, fleshing out or redefining class descriptors of previously fashionable diagnostic groupings that proceed beyond the axis I phenomenological definition (e.g., reactive and neurotic depression).

There are a number of arguments for involving axis II components in our model. First, the clinical concept of reactive or situational depression resists identification in multivariate analyses of depressed patients who are seen for psychiatric assistance (25–27), presumably because here—and in comparison with transient mood states experienced by nonclinical subjects in response to stress—the disordered response is induced and/or maintained by predisposing factors. This indirectly argues that the "soil" is of greater definitional utility than the "seed" and that the subgroup definition by "seed" (i.e., the presence of severe life event stressors) will fail when tested empirically.

Second, the definition of neurotic depression has resisted clarification, despite several attempts (28, 29). Akiskal and McKinney (2) articulated commonly held frustrations with the term, questioning whether it was synonymous with "reactive" or "psychogenic" or referred to "neurotic symptoms" such as "anxiety, obsessions, and inordinate fatigue superimposed on a primary disturbance of mood" or to "a characterological propensity to overreact with depressive symptoms when confronted with minor frustrations," and asked how such concepts might be operationalized.

An alternative approach to pursuing potential nonmelancholic subgroupings is to first establish whether meaningful syndromes or disorders have been identified clinically and with some observational consistency. This is an appropriate time to consider an older paradigm. Kendell (30) has described how "For most of medical history, syndromes have been identified intuitively by gifted physicians on the basis of their personal clinical experience. They saw a pattern where others saw only confusion, or they saw a different pattern than had their predecessors" (p. 308). Such syndrome identification was "essentially an imaginative insight which could not be reduced to formal rules" (p. 308). In comparison, Kendell judged that psychiatry’s more recent strategies (e.g., structured interviews, multivariate analyses) have produced "meager" returns. Psychiatrists recognize but have perhaps exaggerated the limitations to clinician-based subgroupings (e.g., invalid constructs opined by the feudal barons of psychiatry and reified by acolytes). Consistency of independent observation, as well as durability of description across time and culture, merits respect.

Of importance for the strong historical emphasis on an association between psychomotor disturbance and "melancholia," clinical observation has identified candidate nonmelancholic subgroupings. Quite variable multivariate analytic approaches, including factor, principal component, cluster, and grade of membership (see reference 27) have assisted labeling of "anxious" and "irritable/hostile" depressive disorders. Their clinical identification, however, has been quite vague, suggesting limitations to symptom-based definition alone. Descriptors of class members in relevant studies argue more for a spectrum paradigm, linking symptom state and underlying personality style or temperament—a concept variably described (31–33). In our study of defined nonmelancholic subjects (27), the "anxious worrier" patients were more likely to have a family history of anxiety, to have a history of school phobia and behavioral inhibition in childhood, to have become dependent on anxiolytic drugs and alcohol, to meet lifetime criteria for an anxiety disorder, to be viewed by relatives as "anxious worriers," to "act in" when stressed, and to have a cluster C personality style. By contrast, our "irritable/hostile" patients were more likely to "act out" and have a "short fuse" in response to stress, to have difficulty in maintaining relationships, and to have a cluster B or volatile personality style.

Such descriptors are more likely to describe differences based on temperament than on axis I clinical symptoms, thus explaining why any subtyping model cannot be based on only symptoms and signs. Temperament is likely to be essential for subtyping the nonmelancholic disorders. As we have also shown that "anxious worrier" patients are more likely—when depressed—to show anxiety (34), whereas "volatile" temperament patients are more likely to appear irritable when depressed, some vertical continuity across axes is observable. Just as it was earlier suggested that psychomotor disturbance is a surface marker of an underlying neuropathological process, we similarly view these two spectrum disorders—albeit with more noise distortion in the signal-to-noise ratio—so that "temperament style" does not lead invariably to an equivalent "symptom pattern." To the extent, however, that there is some continuity, we assume that certain biological factors shape both these contrasting temperament styles and surface symptoms—or, more likely, that the contrasting biologically determined temperament styles act like a lens for focusing the contrasting surface symptom patterns crudely described as "anxious" and "irritable/hostile" depression. Although dysthymia is clearly a heterogeneous construct, subsuming disorders that achieve duration and severity criteria by multiple pathways, we assume that the spectrum model may provide superior definition of many instances of "dysthymia," recognizing a contribution of personality style (including anxious) to both the onset and persistence of depression.

Within the overall nonmelancholic class, we presume that multiple other expressions and admixtures of symptoms and temperament styles await delineation. Although the definition of the nonmelancholic disorders will always be imprecise and certainly require greater clinical clarification, we favor pursuing the utility of such subgroupings according to a spectrum model rather than by definition on a severity basis. Treatment may then also become more logical and rational.

The final common pathway model put forth by Akiskal and McKinney (1) stands in sharp contrast with the model outlined here, with each having major implications for classification and clinical practice, as well as for research study design and results. Reference here to the final common pathway model will focus on the authors’ more detailed article (2). In essence, the authors identified five broad processes or domains (a genetic predisposition, developmental factors such as early parental loss, psychosocial events or stressors, physiological stressors, and personality traits) and held that varying admixtures of these processes "conceivably converge in those areas of the diencephalon that modulate arousal, mood, motivation, and psychomotor function" (p. 290). They postulated that a "diencephalic final common pathway" interlocks "processes at chemical, experiential, and behavioral levels that, in the language of neurophysiology, translate into a functional impairment of the diencephalic centers of reinforcement" (p. 300). Furthermore, they stated that the "diencephalic final common pathway accounts for the shared clinical features seen in the heterogeneous group of depressive disorders—especially when they attain the melancholic phase" (p. 300), thus encouraging a distinct unitarian interpretation. However, by adding that "phenomenological heterogeneity (unipolar, bipolar, and secondary) is explainable by differences in the pathogenic factors and the multilevel interaction possible among these factors" (p. 300), they also appeared to allow different "types" of depression.

F2 and F3 contrast the final common pathway and currently proposed models. The final common pathway model holds, in essence, that differing etiological and determining factors act upstream then funnel down to emerge through a spout and exert some level of dysfunction on a single system (the diencephalon). Thus, any upstream differences (e.g., variable pathogenic factors) are consolidated and expressed along a single conduit, allowing principally for dimensional differences in the end product (i.e., depression severity).

In the contrasting model (F3), we suggest that certain neurobiological processes lead to an obligatory depressed mood component (I), alone or via recruitment of other neurobiological processes determining the dimensional expression of, for example, anxiety (II), irritability/hostility (III), and fatigue (IV), with such features as II–IV joining with depression to define, in part, mood state parameters. Depending on their variable presence and admixture, these mood state components define the surface clinical expression of the nonmelancholic depressive disorders. Activation of additional (if not independent) processes is required to induce psychomotor disturbance (component V) and thus aggregate to a "melancholic" disorder, whereas activation of further processes (VI) is similarly required to aggregate psychotic features to create psychotic depression. The model allows a logical but deceptively simple framework for conceptualizing and then operationalizing the classification of separate major depressive disorders.

Two major reasons for changing the current unitarian diagnostic paradigm—impact on research and treatment utility—can be advanced. As noted earlier, current diagnostic systems have failed to deliver consistent neurobiological research findings or to demonstrate clear treatment specificity when contrasts are made between melancholic and nonmelancholic depression—not unexpected if the unitarian model is valid. The unitarian model encourages treatment decisions to be made on a simplistic severity basis (e.g., ECT for severe depression, antidepressant medication for moderate, and psychotherapy for mild depression).

If qualitatively different and clinically meaningful depressive subtypes actually exist, waiting like the Pietà to have some surrounding stone chipped away, it is likely that there will be substantive treatment-specific implications. As Kendell (30) has noted, it was not until cardiac and renal forms of dropsy (edema) were distinguished that prediction of who would respond to digitalis became clear. It is salutary to read Klein’s article of a quarter century ago (11) and to observe the general treatment implications that might have followed acceptance of his categorical model at that time and realize how much time has been lost and how many research endeavors wasted by subscription to the unitarian paradigm.

In terms of research, the hierarchical model (relying on the presence or absence of specific features) alone allows a powerful research study paradigm, particularly for mounting and pursuing neuroimaging and treatment-specificity studies. Beyond advocacy, let the contrasting models compete as research paradigms. Both have to explain the ubiquitous feature of depression and why some combinations of the components deconstructed here are clinically uncommon or nonexistent (e.g., psychotic depression without psychomotor disturbance) and identify the sites of three potentially separate constructs—mood, psychomotor disturbance, and psychotic features—and how each might be activated. For instance, are there independent sources for each, or is each component activated disparately from a single source? Is activation simultaneous or cumulative? If, by contrast, the unitarian model is valid, we need a more sophisticated explanation for how a depressed mood state could evolve or bud off into the expression of psychomotor disturbance and psychotic features.

If, in contrast, the hierarchical model is superior, then iteration between research findings and clinical initiatives should refine the rules for diagnostic classification. If it is to be extended by delineation of nonmelancholic spectrum disorders, even greater iteration will be required, as the boundaries between nonmelancholic disorders are less precise, and definition will not come easily.

In terms of treatment implications, we already have both definitive and suggestive data revealing that depressive subtypes do show differential naturalistic and treatment outcomes. Two meta-analyses (35, 36) have quantified the response rate of psychotic depression to antidepressant medication alone at 20%–30%, to antipsychotic medication at 30%–40%, and to combination antidepressant/antipsychotic medication or to ECT at approximately 80%. Those treatment differences are substantive, and the data appear definitive. There is the suggestion that melancholic depression (37–39) may be less responsive to selective serotonin reuptake inhibitors than to the tricyclic drugs, but when the definition of melancholia is problematic, no definitive answer can be expected. Clinically diagnosed melancholic and nonmelancholic subjects have demonstrated substantive differences in long-term outcome (40, 41). When, before DSM-III, descriptive reports considered clinical subtypes, Blashfield and Morey (42) concluded that "anxious depressives respond well to major and minor tranquilizers but not to tricyclics, while hostile depressives show little improvement with conventional drug therapies" (p. 522). And although placebo and spontaneous remission rates appear substantive in nonmelancholic depression, spontaneous improvement is quite uncommon in melancholic and psychotic depression (43). Such indicative data have emerged far more clearly from studies that have assessed the depressive disorders categorically rather than dimensionally. Irrespective of that paradigm base, such studies argue for a more rigorous examination of the proposition that distinctive depressive subtypes may each have their own hierarchy of treatment effectiveness.

The final common pathway was truly integrative and was an important heuristic model. It merged well with, and helped to shape, the DSM-III model (and subsequent DSM-IV and ICD-10 models) for conceptualizing the depressive disorders dimensionally. This was not inappropriate then as, at the time that the DSM-III system was introduced, the binarians had failed to prove their case. The DSM-III model reflected the then-classificatory stalemate. It made clinical diagnosis easy—by creating homogenized entities such as major depression and dysthymia—but failed to deliver substantive and acceptably replicable research findings. Akiskal and McKinney (1, 2) did the profession a service in providing an exhaustive overview of the models and data sets available in the early 1970s and in seeking to integrate such data into the parsimonious final common pathway model, but the model developed by their scholarship was neither fully used nor its limitations pursued.

Of most importance, progress might be expected if research studies were to contrast results generated by DSM-IV, ICD-10, and categorical paradigm models into etiological and treatment outcome studies, as comparative results would allow an empirically driven diagnostic paradigm to be determined and operationalized. It is not too late to debate and, more important, to examine the research implications of the competing paradigms.

In Milton’s Paradise Lost, a palace of pandemonium was built and paradise was lost by yielding to temptation. In the sequel, resistance to temptation regains paradise. The unitarian view of the depressive disorders is temptingly straightforward, but it is time to reject its simplicity.

Received March 11, 1999; revisions received Aug. 16, 1999, and Jan. 31, 2000; accepted Feb. 16, 2000. From the Mood Disorders Unit, School of Psychiatry, University of New South Wales, Randwick. Address reprint requests to Prof. Parker, Psychiatry Unit, Prince of Wales Hospital, Randwick 2031, N.S.W., Australia; g.parker@ unsw.ed.au (e-mail).Supported in part by National Health and Medical Research Council grant 993208 and a grant from the New South Wales Department of Health Infrastructure.The author thanks Dusan Hadzi-Pavlovic, Kerrie Eyers, and Heather Brotchie for their help.