To the Editor: Gordon Parker, M.D., Ph.D., D.Sc., F.R.A.N.Z.C.P., et al. (1) challenged "the DSM-IV definition of the atypical features specifier in major depressive disorder as a valid entity" (p. 1477). This challenge was based on their analysis, which suggested that the symptoms we refer to as vegetative atypical features—overeating, oversleeping, a leaden feeling, or rejection sensitivity—did not co-occur and were not associated with mood reactivity.

Their critique’s rationale has two flaws. The first misconception has to do with latent class analysis. Latent class analysis can be considered a technique to unmix data or uncover taxonomies or nonarbitrary classes. Kendler et al. (2) (cited by Dr. Parker et al.), in an epidemiologically defined twin sample, performed a latent class analysis, identifying atypical depression as a distinct subgroup. Once latent classes are identified, Dillon and Goldstein (3) noted that "within a cluster, the items are independent" (uncorrelated). Dr. Parker and colleagues reported the expected low correlation of vegetative symptoms. Actually, when the entire group was examined, Dr. Parker et al. found three significant correlations (of a possible six): rejection with hypersomnia (p=0.02), weight gain with leaden paralysis (p=0.03), and another by inference (r=0.12, df=158, p<0.07). From the manner in which the data were presented on the third correlation, it is unclear which two symptoms had this correlation. Within the patient subset with reactive mood and one accessory symptom, there were no significant correlations. The anticipated occurred: a less homogeneous group exhibited significant correlations, and a cluster (homogeneous group) had uncorrelated symptoms. Angst et al. (4) reported a relevant analysis after the study by Dr. Parker and colleagues was accepted for publication. In a Zurich epidemiological sample, which, by definition, was heterogeneous, a high association between atypical depressive symptoms was found.

The second misconception in the article was the suggestion that correlation (symptom interdependence) and not predictive validity is the gold standard for validating a phenomenologically derived syndrome. Height and weight are correlated. Although this correlation exists for men and women, it does not help differentiate the sexes. Predictive validity in medicine has clear heuristic and practical syndrome relevance, especially predictions of treatment outcome.

Do data support the predictive validity of the atypical subgroup? In fact, the validation for the DSM-IV atypical depressive parenthetical modifier is unusually robust for psychiatry (2, 4–10). Our group at Columbia University (8) performed six independent trials in which patients with varying degrees of atypical depression had superior response to phenelzine over imipramine. In two trials, patients lacking atypical features did no better taking phenelzine than imipramine. Sotsky and Simmens (6) also noted that atypical features were associated with a poor imipramine response. Subsequently, Kendler et al. (2) and, independently, Sullivan et al. (7), in epidemiological samples involving thousands of patients, defined a cluster of depressed patients with hypersomnia and hyperphagia who were distinct from other subgroups of depressed patients. Kendler et al. (2) reported that atypical depression bred true.

Dr. Parker et al. also questioned the relevance of mood reactivity. They failed to note the independent analysis of the National Institute of Mental Health Treatment of Depression Collaborative Research Program data by Sotsky and Simmens (6). A syndrome definition that included mood reactivity coupled with vegetative atypical symptoms best identified a group with poor response to imipramine. However, without the mood reactivity context, predictions were much weaker.

Reconciling disparate findings is a recurrent problem in clinical research. It is likely that differences in group selection explain discrepancies between the study by Dr. Parker et al. and the findings of the Columbia group. Roughly 65% of our patients were chronically ill, with a duration of illness of approximately 20 years; all were outpatients. Dr. Parker et al. selected patients with major depressive disorder "present less than 24 months"; "69%…were outpatients" (p. 1473) (we assume that 31% were inpatients). It is unclear why Dr. Parker et al. chose to exclude patients who were ill more than 2 years since atypical depression is a chronic illness. The relevance of chronic depression has been noted (11, 12).

Dr. Parker et al. inaccurately suggested that we define "elephants" (atypicals) as "not giraffes" (melancholics); therefore, elephants are poorly described. Atypicals (elephants) are characterized without reference to melancholia. Patients with major depression who have reactive mood and atypical symptoms respond best to monoxamine oxidase inhibitors (MAOIs) (poorly to tricyclic antidepressants). Patients with otherwise identical symptoms (i.e., those with nonautonomous mood and no atypical symptoms), referred to as having simple mood-reactive depression, do equally well taking tricyclic antidepressants and MAOIs (8). In the most recent "iterative refinement," we (9) demonstrated that only chronically ill patients with vegetative atypical symptoms had a superior response to MAOIs.

We also wish to correct several erroneous assertions by Dr. Parker et al. In the study by Mannuzza et al. (13), 36% of the patients with social phobia met criteria for atypical depression, not two-thirds, as Dr. Parker et al. suggested. Joyce and Paykel (14) did not present new data supporting the efficacy of MAOIs in patients with anxious depression but referred to Columbia data (15) suggesting that MAOIs were particularly effective for anxious patients with atypical depression. However, the relevance of anxiety was no longer evident in the expanded study group, suggesting that a fortuitous, nonreproducible finding had been produced. In the Robinson et al. studies (e.g., 16), phenelzine was never shown to be superior to tricyclics for anxious depression. Dr. Parker et al. stated that we (17) indicated that selective serotonin reuptake inhibitors (SSRIs) are equivalent to MAOIs for atypical depression. However, we did not compare SSRIs and MAOIs.