Costello syndrome

Ten years ago Clarkin and Kendall (1992) defined the study of comorbidity as "the premier challenge facing mental health professionals in the 1990s." At that time they called for studies of the frequency of comorbidity and related conditions, examination of symptom overlaps and their potential role in defining boundaries between related disorders, and studies of differential effects of treatment of children with "pure" versus comorbid disorders and of children who, within the same disorder, differ on etiological factors. But as best as I can tell by my Psychlit and Medline literature search (1992-2002, using terms "comorbidity," "psychopathology," and "child or adolescent" in all combinations of any two), apart from one special section on ADHD comorbidities in the MTA study (Jensen, 2001a), this special section is the only one devoted to the general issues of comorbidity in any of our major mental health journals since Clarkin and Kendall issued their call to arms. Moreover, extremely few systematic studies h ave been done that clearly respond to their challenge--that is, comparing comorbid and pure forms of disorders, in terms of implications for classification, diagnosis, treatment response, and prognosis. I could locate only one book on child/adolescent comorbidity (on ADHD, Brown, 2000). In short, this much needed special section appears to represent the most systematic exploration of these issues to date. Not only has it been long-awaited--if anything, it is long overdue.

The research challenges posed by comorbidity are daunting (Nottelmann & Jensen, 1995). Perhaps because of these challenges, most of our child psychopathology research over the last decade has not addressed these problems satisfactorily, usually failing on three broad fronts: First, in many previous studies (except epidemiologic surveys) children with comorbid disorders have been excluded, likely making these studies' results inapplicable to children (who are usually comorbid) seen in mental health settings. Fortunately, the tide seems to have turned, as more and more of the studies mounted by NIMH and industry tend to allow children with comorbidity to be included in research studies, even including treatment trials (e.g., MTA Cooperative Group, 1999a, 1999b; RUPP Anxiety Study Group, 2001).

Second, even when comorbid children have been included in studies, the methodologic issues pertaining to definitions of comorbidity have remained unexplored, leading to inexplicable and/or discrepant findings across studies, as well as making most comparisons meaningless. This seemingly straightforward problem seems to have been ignored by most investigators, even those explicitly studying comorbidity. And third, with the possible exception of ADHD (e.g., see Biederman, Baldessarini, Wright, Keenan, & Faraone, 1993; Biederman, Milberger, Farnone, Guite, & Wharburton, 1994; Biederman, Newcorn, & Sprich, 1991; Jensen et al., 2001; Jensen, Martin, & Cantwell, 1997; Jensen, Shervette, Xenakis, & Richters, 1993; Sprich-Buckminster, Biederman, Milberger, Faraone, & Lehman, 1993; Wozniak et al., 1995), comorbidity characteristics, correlates, and outcomes have only rarely been described in most previous studies, leaving readers uncertain as to whether a disorder, when accompanied by one or more others, is still esse ntially the same thing as the pure form of the disorder. Without such studies, clinicians, scientists, and diagnostic system developers alike will remain relatively ignorant as to whether comorbid versus pure forms of the disorder have different etiologies, clinical characteristics, or outcomes, and/or whether they will require different prevention and treatment strategies.

This special section adds welcome new data to inform the last two problem areas, thorny conceptual issues that are all too often avoided: how do rates of comorbidity vary as a function of informant or decision rules employed when combining data from multiple informants? (For preliminary answers, see Youngstrom, Findling, & Calabrese, 2003, this issue). It also considers questions that cut somewhat deeper: in terms of broad symptom types (e.g., internalizing vs. externalizing), do some types of internalizing (INT) and externalizing (EXT) symptoms tend to covary, whereas others remain relatively "pure" and unaffected by the presence of the opposite type? Are some risk factors relatively specific to either "pure" INT, or EXT conditions, whereas other risk factors are more related to co-occurring symptom patterns? (For preliminary answers, see Keiley, Lofthouse, Bates, Dodge, & Pettit, 2003, this issue). In longitudinal studies, do some types of symptoms or disorders (e.g., depression) tend to precede or follow o thers (e.g., delinquency)? Is the disorder precedence pattern predicted by baseline risk factors? Does the disorder precedence pattern have long-term prognostic implications, in terms of symptom improvement or exacerbation? (For preliminary answers, see Beyers & Loeber, 2003, this issue).

In the comments that follow, I outline in broad brush strokes what these special section authors may have and have not achieved. You will readily see that on some points I disagree with either their assumptions and/or their conclusions, but these points tend to be more minor quibbles against the background of the section's major contributions. Also, I offer a modest set of next-step recommendations for the field, if we are to rise to meet the Clarkin and Kendall challenge.

OPERATIONALIZING COMORBIDITY

The study by Youngstrom et al. (2003, this issue) conducts head-to-head tests within a single sample, comparing categorical versus dimensional approaches to definitions of comorbidity. Happily, this is not the same old "which method, categorical or dimensional, works best?" type of study. Rather, these investigators sought to determine the impact on comorbidity rates by using different decision strategies of how data are combined across informants. Their findings revealed great differences in rates of INT + EXT comorbidity (from 5 to 74%), not only when comparing traditional diagnostic interview approaches with checklists converted to clinical cutoffs, but also when comparing comorbidity rates as a function of which informant provides the information. Thus, regardless of informant source, even when using empirically derived dimensional measures converted to their recommended clinical cutoffs, dramatic differences in comorbidity rates and very low rates of agreement on comorbidity status are found (with kappas ranging from .17 to .30).

Youngstrom and colleagues draw useful distinctions among three different decision strategies for integrating discrepant data across multiple informants: disjunctive (the so-called "OR" rule), conjunctive (the "AND" rule), and compensatory (ADDITIVE rule; see Bird, Gould, & Staghezza, 1992) strategies. As they demonstrate, each of these approaches produces quite different comorbidity rates, even though each relies on the same data. To their three strategies, I would add a fourth: the discriminative strategy (elsewhere called the IFs, ANDs, OR BUTs rule; Jensen et al., 1999). Using this approach, one must determine when to rely on one, another, or multiple informants. Usually this requires a well-trained clinician to integrate discrepant data. This approach seems particularly sensible, given evidence indicating that no single information source is always appropriate or valid across all disorders (Angold & Costello, 1996; Hart, Lahey, Loeber, & Hanson, 1994; Jensen et al., 1999; Loeber, Green, Lahey, & Stouthame r-Loeber, 1989), and because the accuracy of an informant source will vary from person to person (e.g., one youth might be an excellent informant whereas another is not). Under such conditions, the judgment of a trained clinician who can evaluate information quality is essential. The process for ensuring valid diagnostic information has been described by Spitzer (1983) as the "LEAD standard," which requires that diagnostic information be longitudinal and obtained by expert clinicians who have access to all available data sources. To strengthen the LEAD standard, I have recommend the addition of the SMARTER standard: using standardized measures, assessing all relevant functioning domains across settings and raters, using trained expert clinicians, trained to achieve reliability (Jensen, 2001b). These additional recommendations take into account the recent advances in our diagnostic and assessment methods, the advent of a wide array of excellent measures, and a growing awareness that even expert clinicians may not be reliable among themselves without careful training and structuring of how they gather information and interview various informants (Piacentini et al., 1993).

In sum, Youngstrom and colleagues' study constitutes an important contribution, even though they did not explore what I suggest would be the most valid combinatorial strategy, the discriminative approach using the LEAD and SMARTER approaches. Note also, their approach to comorbidity was cross-sectional: they did not attempt to explore the impact of order of disorder emergence on comorbidity, nor did they explore the relative impact of various risk factors on comorbidity development, topics left to the next two papers in the section.

RISK FACTOR SPECIFICITY IN "PURE" VERSUS COMORBID STATES

Taking a different tack, Keiley et al. (2003, this issue) attempt to tease out unique versus overlapping symptoms, applying confirmatory factor analyses to data from two informants (mothers, teachers) collected at multiple data collection points over time. These analyses revealed constructs of "pure" INT, EXT, and overlapping (COVARY) symptom dimensions, within and across the two informants. Using structural equation modeling, they then examined the shared and unique impacts of four different types of risk factors on latent constructs of each of the three symptom types. Their analyses not only provide support for unique symptom profiles (INT, EXT, and COVARY), but also identify risk factors that are relatively specific to each symptom type. For example, the authors identify the risk factors of temperamental unadaptability and female gender as specifically related to the INT symptom profile. In contrast, temperamental resistance to control, parental harsh punishment, male gender, low socioeconomic status, peer rejection, and lower temperamental unadaptability were related to pure EXT symptoms. Finally, family stress and peer rejection were related to the expression of covarying symptoms.

By using empirically derived symptom scales for two separate informants, they were able to avoid the nosologic confusion sometimes resulting by using different assessment systems. Similarly, by using multiple time points to capture the latent constructs for both parents and teachers of the pure INT, EXT, and COVARY symptom profiles, they may have been able to reduce measurement error and enhance the reliability of these latent constructs. Although these strategies appear to constitute a strategic methodologic advance for comorbidity studies in children, they do not totally circumvent several other problems-albeit problems that, for the most part, the authors themselves point out.

First, fine-grained patterns, not apparent in broadband INT and EXT symptom profiles may be obscured. Within a given broad band construct, finer symptom! disorder types may in fact have different risk factors, illness correlates, treatment responses, and outcomes. For example, although obsessive--compulsive disorder (OCD) is an anxiety disorder (i.e., TNT), a number of studies have established that one form of OCD arises as an autoimmune response to beta-hemolytic streptococcal infection, with specific and unique response to autoimmune therapies (Singer, Giuliano, Zimmerman, & Walkup, 2000; Swedo & Kiessling, 1994). Similarly, depression and anxiety disorders, both presumably part of an INT profile have treatment responses to SSRIs of quite different magnitudes, suggesting important differences within the INT broad-band category (Keller et al., 2001; RUPP Anxiety Study Group, 2001).

Second, their statistical model cannot tease out the effects of systematic distortion by one or both raters (e.g., the depression-distortion hypothesis, Richters, 1992) that would not show up as an error term, since as conceptualized, it is a reliable characteristic of a rater. The potential problem of systematic distortion is less of an issue for teachers' ratings, since each data point is based on ratings from teachers from different grade years. Thus, any characteristic of a single rater not shared with other teachers would be modeled in the error term.

A third and related problem is the implicit assumption that the phenomena of interest, that is, valid characterizations of symptoms, whether comorbid or not, are appropriately identified by these two types of informants. The problem of course, is that all informants are not equal, nor fully adequate. Thus, as the authors suggest, teachers may not be particularly good reporters of internalizing symptoms, whereas parents may not be good raters of children's peer relations. Moreover, in their final model, maternal and teacher reports of comorbid symptoms were no longer significantly correlated, once all risk factors' effects were entered, including their effects on individual informants' ratings. As the authors note, this could be based on the possibility that parent and teacher reports of youth behavior problems are dependent upon so many of the same risk factors, and once these effects are accounted for, each informant witnesses comorbid symptoms either specific to their setting or based on informant-specific biases.

A fourth important problem in their approach is that any potential effects of time and disorder precedence are ignored, since they used all data time points to model the INT, EXT, and COVARY latent constructs. Consider for example their tentative conclusion that the COVARY profile is most similar to the EXT profile, in that these two symptom dimensions shared similar risk factors: When considering the effects of time, the child who first evinces an INT-only profile, to be augmented later by EXT symptoms (NT + EXT), might be treated in a statistically identical fashion to the child who first shows an INT + EXT profile, only to later show an INT-only profile. Given their approach, such children would be treated as COVARY, even though common sense makes apparent that they are in fact quite different in underlying disease processes and outcome, as well as likely risk factors. One child is likely improving, whereas the other may have a much more pernicious subsequent course.

Lastly, it is not clear that this method as they apply it here is up to the task of teasing out specific versus shared effects of various risk factors, if risk factor timing and disorder precedence are not accounted for in the model, and if we do not have strong knowledge that the risk factor's role is invariant over time. Thus, the same underlying risk factor or disorder process can yield quite disparate results, depending upon the point of the illness or stage of disorder progression.

A medical example makes this quite clear: prolonged heat exposure (the risk factor) may lead to a syndrome of heat exhaustion, characterized by increased heart rate, sweating, and elevated temperature (the symptom pattern). If heat exposure persists, it may be then followed by a much more severe condition, heat stroke, characterized by no sweating, and loss of consciousness, and cardiovascular collapse (a new symptom profile if one ignores the time dimension). Once heat stroke status is reached, further risk factor exposure has little effect. Such a situation, if modeled by collapsing multiple time points, might appear to reflect a comorbid situation (two different symptom patterns) with the risk factor uniformly important, when in fact it is the same underlying illness process (not a comorbid state), and the risk factor's presence and persistence is critical early on, but irrelevant in end stages of the illness.

In fairness to the authors, they do indicate their intent to apply future efforts to tease apart the potential effects of time. Their methods are quite sophisticated, appear to offer clever approaches for identifying comorbid versus "pure" states and for teasing out shared and unique effects of risk factors on these various states, and should have broader application by our field. The sophistication and complexity of these methods--which I had to read and reread several times over--bespeak the need for multidisciplinary team input: from clinicians experienced with the onset, course and outcome of abnormal behavior patterns and their relevant risk factors, from developmentalists able to draw distinctions between normal and abnormal behavioral processes expressed over time, and from biostatisticians acquainted with state-of-art statistical approaches and their limitations.

THE ROLES OF TIME AND DISORDER PRECEDENCE ON COMORBIDITY OUTCOMES

Rounding out the special section, Beyers and Loeber (2003) examine the effects of time, timing, and development on comorbidity impact. They explore the situation when one type of disorder precedes another (vs. the opposite pattern): does such a disorder precedence pattern convey a different meaning in terms of illness course and eventual outcome? Using two-level hierarchical generalized linear models, controlling for the effects of baseline risk factors on comorbidity, and including time-varying covariates, they studied the overlaps and course of depression and delinquency in a high-risk male youth sample. Their overall results suggest that depression affects the concurrent expression, later course, and overall severity of delinquency, just as delinquency affects the expression of depression, even after controlling for various shared risk factors. However, their results also suggest that delinquency affects the expression of depression more than delinquency affects the course of depression.

Moreover, their findings suggested (though they did not directly test) that depression occurring early in the course of development should be considered as different-in-kind than later-onset depression emerging in the presence of already crystallized delinquency. In fact, here it is appropriate to ask the question, when is depression sufficiently different that it should considered qualitatively distinct from another presumably similar depression? Thirty years ago Robins and Guze (1970) offered advice in this regard, proposing a five-phase approach based upon the premises that a new, presumptively valid diagnostic construct should demonstrate at least some (preferably all, though this has not yet been achieved with any psychiatric disorder) of the following characteristics: distinctive clinical/descriptive characteristics, distinctive biologic characteristics (e.g., laboratory studies), evidence that the disorder "breeds true" within families, follow-up studies that demonstrate a distinctive clinical course o r outcome, and evidence that the disorder can be differentiated from other syndromes. Applying this strategy to disorders of childhood and adolescence, Cantwell (1995) modified the model to include eight domains of clinical investigation: clinical phenomenology, demographic factors, psychosocial factors, biologic factors, family genetic factors, family environmental factors, natural history, and intervention response, all areas where distinctions might be made to provide support for the distinctiveness of a proposed clinical entity. While it is not clear that these criteria, originally developed to judge the validity of wholly separate disorder categories, should be used to provide qualitative distinctions between various comorbidity combinations (including pure vs. comorbid forms of disorder), I suggest that these criteria constitute a useful and appropriately conservative approach with which the meaningfulness of any new entities can be explored. Applications of this approach within the NIMH MTA study have recently been published (Jensen et al., 2001).

Which Robins and Guze/Cantwell criteria are addressed by the authors? Given the design of their study, most findings fall into the category of distinctive clinical course or outcome. Here they provide evidence that depression exposes a youth to increased risk for development of eventual delinquency, and that depression may affect the course of remission of delinquency symptoms, such that youth with depression show less symptom improvement in delinquency than youth without depression. In this regard, they note the importance of future research to identify and classify individuals by trajectories using latent mixture models, so that predictions might be made and tested to determine which youth might benefit the most from preventive interventions.

Despite the sophistication of their approach, can we be confident that these methods (or anyone's methods, perhaps?) are yet adequate to explore and test the complex phenomena that clinicians believe that they can observe, such as the differential role of a risk factor at an early versus a later point in development (essentially a two-way interaction), or the impact of a risk factor occurring in the presence (vs. absence) of one disorder at a certain point in development on the emergence of a comorbidity (essentially a three-way interaction)? Are these methods up to the task of identifying the interactions of gender, ethnicity, family history, personal attributions, or genotype in any of these above situations, or worst of all, identifying the specific contributions of all of these factors as they play out concurrently, over time, and interactively on a whole range of different outcomes?

To use an analogy from a friendly commentator in this area (Dan Robinson, PhD, personal communication), if Johannes Kepler were to begin from the perspective that the earth is the center of the universe, and then apply our typical study methods to understand the positions of the planets in the sky, he might enter time of day, planets' height above the horizon, time of year, presence or absence of a full moon, and apparent distances of planets from each other into complex regression models. Doing so, Kepler might happily have found that some small amount of variance of planets' positions could be captured by this approach. Yet he would never approach the degree of understanding and precision captured by an alternative model that begins from the perspective of the sun as the center of the solar system, with separate planets' earth-viewed location modeled by equations that take into account the earth's position around the sun (time of year), the distance of all other planets from the sun, formulas describing the nature of each of their orbits (in three dimensions, because the orbits lie on different planes), and the impact of gravitational fields of some planets on others.

How many of the comorbidity phenomena that we are trying to capture are in essence four-, five-, six-, or more-way interactions that lie beyond traditional detection methods, given measurement error and increasing loss of precision of estimates, as multiple variables and their interactions are added to estimating equations? If some of our psychopathology phenomena are in fact this complex (which I think they are), we have a problematic situation indeed, one that bespeaks the need for close research collaborations between experienced clinicians, psychopathology investigators, and biostatisticians. Intuitive, gifted clinicians may be quite necessary to help us as investigators to discriminate sensible from nonsensical notions and to determine which variables are likely to be of genuine interest (preserving statistical power and precision of estimates), whereas our biostatisticians grapple with the best approaches to test complex study questions, or perhaps develop new methods if necessary. Although our current statistical and design approaches have been good starting points, and while this special section embodies the most sophisticated application of methods and intensive exploration of comorbidity issues to date, as our understanding of the complexity of phenomena increases, these methods are likely to need augmentation by intensive observational methods applied to our subjects over the course of the study, in order to capture glimpses of the essence and impact of emerging forces on psychopathologic development and comorbidity.

RULES OF THUMB FOR FUTURE STUDIES

Based on the issues and problems identified above, below I offer a few suggestions that might move us ahead. No doubt many of these will provoke disagreement among some and controversy among many. As rules of thumb, however, they are not hard and fast guidelines, and instead should be thought of as strategies that might be pursued by at least some in our midst to complement our more traditional, necessarily conservative approaches. Like Johannes Kepler's initial perspective of the sun as the center of the solar system, these rules of thumb begin and proceed from different assumptions than traditional approaches to the study of presumed universal characterisics--rather, they proceed from assumptions that individual difference factors are key to understanding comorbidity--and that time, development, context/setting, meaning, and history often "matter."

To Lump or to Split? When in Doubt, Split Whenever Possible, Then Combine

Current methods fail to give sufficient guidance as to when we should lump or split. But how might Beyers and Loeber's comorbidity findings be different if they took into account dysthymia versus acute depression, or presence of family history of depression, or a broader internalizing (depression and/or anxiety disorder) construct? We do not know. Nor do we know how Keily et al.'s findings might be different if applied solely to a sample of INT children with OCD and a history of streptococcal infection. Although power problems from a "split-first" approach are apparent, staying attuned to the likelihood of such factors will mean that we will be less likely to miss what might seem obvious to experienced clinicians who have a knack for sorting through the many clues pertaining to a given child's outcomes.

Look for and Tease Out Setting- and/or Informant-Specificity

How might Beyers' and Loeber's comorbidity findings differ if their depression construct separated children who were apparently depressed only in presence of their parents in a highly stressed home setting, but who seemed much happier when with their peers? To use a medical analogy, consider the differences in likely etiology and intervention between a child who wheezes only in the presence of cat dander or dog hair (an allergic diathesis), from the child who wheezes frequently and often develops respiratory infections (perhaps chronic asthma or cystic fibrosis). Without information gathered at a certain level of depth and detail, it is not possible to see, much less to explore such differences.

No Combinatorial Strategy Works All the Time

Different decision strategies can be applied to the same data and yield quite different rates of comorbidity, and each strategy has its place. The combinatorial analytic strategies illustrated in this special section may work quite well for one disorder or comorbidity construct, but be totally inapplicable to other disorder types, such as autism, obsessive--compulsive disorder, certain forms of depression or aggression, etc. Also, because both common sense and data indicate that youth and parents are not both always equally valid reporters, why should we combine under some circumstances, such as for ADHD and ODD (Jensen et al., 1999; Loeber et al., 1989)?

Relatedly, Youngstrom and colleagues' conclusion--that current permutations of symptomatology (whether clinical or statistical) are often inadequate--and their resulting four recommendations are very welcome indeed. Some of our constructs likely are best understood as categorical, while others may appear to be dimensional. While our categorical diagnostic systems clearly do not "carve nature at its joints," neither can we assume that our empirically derived checklists adequately capture all of the relevant phenomena. Nor do they clearly separate out the wheat from the chaff of irrelevant phenomena, given the likelihood of individual rater distortion, setting specific behaviors/disorders, and the lack of any well-accepted approach to take the best of these various sources and combine them. Knowing when to use conjunctive, disjunctive, additive, and discriminative approaches is key, and such decisions should be justified first logically , then statistically.

Redraw Disorder Boundaries When Warranted

In some instances apparently comorbid conditions might be more appropriately regarded as a single disorder. When comparing "pure" versus comorbid forms of disorder, it can be reasonably assumed that any differences in pure versus comorbid disorder that are the simple additive effects of shared risk factors, treatment response, or outcomes of the co-occurring disorder should not necessarily be regarded as validational evidence for the existence of a new syndrome. Having a fractured arm and a broken leg need not be given the status of a new syndrome, but if the "syndrome" occurs more frequently than either alone occurs by chance, occurs as the result of a specific/unique set of risk factors, and/or if it conveys to the clinician unique treatment/prognostic information (beyond the additive information of the two fractures), then its consideration as a unique syndrome may be warranted (in this example, the combined fractures might be best considered as the "battered child syndrome"). It is this synergism, that is , the interaction of the two or more conditions that convey unique information that should set the standard for determining whether the comorbid pattern should be regarded as a unique syndrome.

CONCLUSION

We do not know at this point exactly where to draw distinctions between various psychopathology states, such as when a given profile represents comorbidity or multiple aspects of the same underlying disease process. Biologic discoveries will be important, but the possibility for substantive advances need not await neuroscientific findings alone. Thus, some investigators have suggested that the strategy of examining various comorbid and diagnostic groups (e.g., age of onset, the possibility of gender, associated symptoms, neuropsychological test findings) may offer the possibility of substantial gains in understanding. For example, Conners, March, Erhardt, and Butcher (1995) have recommended that future studies of ADHD carefully tease out various subgroups as a function of differences on neuropsychological tests. Similar, promising strategies have been noted with other forms of subdivision and classification (e.g., history of perinatal trauma; Sprich-Buckminster et al., 1993) or methods (i.e., neuroimaging and neuropsychology; Keefe, 1995).

The "lumpers" and "splitters" among us must scrutinize each others' assumptions, and find means to bring more critical thought to bear on our current diagnostic and nosologic practices. Among current diagnostic categories, we must carefully examine for possibly unique diagnostic groups defined by comorbidity, and further refine our categories as evidence proceeds. Similarly, when faced with lack of evidence for separate diagnostic categories, reconsideration of their presumed discrimination may be required. Even when cases appear similar in terms of current behavioral phenomena, we must remember that it is likely that any single form of psychopathology (whether "pure" or comorbid) may arise through quite different routes. Two phenotypically similar persons may actually represent two different populations--one that is drawn from the high end of normal processes, another that reflects some disease process--just as some tall persons appear to reflect a disorder in growth hormone production, whereas others do not reflect any identifiable disease process. Characteristics that might identify etiologically meaningful subgroups might actually lie in the past history of the person, rather than in any current phenotypic expression--for example, early versus late onset conduct disorder.

With these conceptual tools--comparing "pure" and comorbid cases not just on phenotype, but also more broadly on environtype, trajectory-type, history, and biology--further development of our etiologic and process-based understanding should become possible.

Received January 22, 2003; accepted January 24, 2003

REFERENCES

Angold, A., & Costello, E. J. (1996). The relative diagnostic utility of child and parent reports of oppositional defiant behaviors. International Journal of Methods in Psychiatric Research, 6, 253-259.

Beyers, J. M., & Loeber, R. (2003). Untangling developmental relations between depressed mood and delinquency in male adolescents. Journal of Abnormal Child Psychology, 31, 247-266.

Biederman, J., Baldessarini, R. J., Wright, V., Keenan, K., & Faraone, S. (1993). A double-blind placebo controlled study of desipramine in the treatment of ADD: III. Lack of impact of comorbidity and family history factors on clinical response. Journal of the American Academy of Child and Adolescent Psychiatry, 32, 199-204.

Biederman, J., Milberger, S., Faraone, S. V., Guite, J., & Wharburton, R. (1994). Associations between childhood asthma and ADHD: Issues of psychiatric comorbidity and familiality. Journal of the American Academy of Child and Adolescent Psychiatry, 33, 842-848.

Biederman, J., Newcorn, J., & Sprich, S. (1991). Comorbidity of attention deficit hyperactivity disorder with conduct, depressive, anxiety, and other disorders. American Journal of Psychiatry, 148, 564-577.

Bird, H., Gould, M., & Staghezza, B. (1992). Aggregating data from multiple informants in child psychiatry epidemiological research. Journal of the American Academy of Child and Adolescent Psychiatry, 31, 78-85.

Brown, T. E. (2000). Attention-deficit disorders and comorbidities in children, adolescents, and adults. Washington, DC: American Psychiatric Publishing.

Cantwell, D. P. (1995). Child psychiatry: Introduction and overview. In H. I. Kaplan, & B. J. Sadock (Eds.), Comprehensive textbook of psychiatry/VI (pp. 2151-2154). Baltimore, MD: Williams and Wilkins.

Clarkin, J. F., & Kendall, P. C. (1992). Comorbidity and treatment planning: Summary and future directions. Journal of Consulting and Clinical Psychology, 60, 904-908.

Conners, C. K., March, J. S., Erhardt, D., & Butcher, T. (1995). Assessment of attention deficit disorders (ADHD): Conceptual issues and future trends [Monograph]. Journal of Psychoeducational Assessment, 28, 185-204.

Hart, E. L., Lahey, B. B., Loeber, R., & Hanson, K. S. (1994). Criterion validity of informants in the diagnosis of disruptive behavior disorders in children: A preliminary study. Journal of Consulting and Clinical Psychology, 62, 410-414.

Jensen, P.S. (2001a). ADHD comorbidity and treatment outcomes in the MTA. Journal of the American Academy of Child and Adolescent Psychiatry, 40, 134-136.

Jensen, P. (2001b). Clinical equivalence: A step, a misstep, or just a misnomer? Clinical Psychology: Science and Practice, 8,436-440.

Jensen, P. S., Hinshaw, S. P., Kraemer, H. C., Lenora, N., Newcom, J. H., Abikoff, H. B., et al. (2001). ADHD comorbidity findings from the MTA study: Comparing comorbid subgroups. Journal of the American Academy of Child and Adolescent Psychiatry, 40, 147-158.

Jensen, P., Martin, D., & Cantwell, D. (1997). Comorbidity in ADHD: Implications for research, practice, and classification. Journal of the American Academy of Child and Adolescent Psychiatry, 36, 1065-1079.

Jensen, P., Rubio-Stipec, M., Canino, G., Bird, H., Dulcan, M., Schwab-Stone, M., et al. (1999). Parent and child contributions to child psychiatric diagnosis: Are both informants always needed? Journal of the American Academy of Child and Adolescent Psychiatry, 38, 1569-1579.

Jensen, P. S., Shervette, R. E., Xenakis, S. N., & Richters, J. (1993). Anxiety and depressive disorders in attention deficit disorder with hyperactivity: New findings. American Journal of Psychiatry, 150, 1203-1209.

Keefe, R. S. E. (1995). The contribution of neuropsychology to psychiatry. American Journal of Psychiatry, 152, 6-15.

Keiley, M. K., Lofthouse, N., Bates, J. E., Dodge, K. A., & Pettit, G. S. (2003). Differential risks of covarying and pure components in mother and teacher reports of externalizing and internalizing behavior across ages 5 to 14. Journal of Abnormal Child Psychology, 31, 267-283.

Keller, M. B., Ryan, N. D., Strober, M., Klein, R. G., Kutcher, S. P., Birmaher, B., et al. (2001). Efficacy of paroxetine in the treatment of adolescent major depression: A randomized control trial. Journal of the American Academy of Child and Adolescent Psychiatry, 40, 762-772.

Loeber, R., Green, S., Lahey, B. B., & Stouthamer-Loeber, M. (1989). Optimal informants on childhood disruptive behaviors. Developmental Psychopathology, 1, 317-337.

MTA Cooperative Group. (1999a). A 14-month randomized clinical trial of treatment strategies for attention-deficit hyperactivity disorder. Archives of General Psychiatry, 56, 1073-1086.

MTA Cooperative Group. (1999b). Moderators and mediators of treatment response for children with attention-deficit/hyperactivity disorder. Archives of General Psychiatry, 56, 1088-1096.

Nottelmann, E., & Jensen, P. (1995). Comorbidity of disorders in children and adolescents: Developmental perspectives. In T. Ollendick & R. Prinz (Eds.), Advances in clinical child psychology (vol. 17, pp. 109-155). New York: Plenum.

Piacentini, J. C., Shaffer, D., Fisher, P., Schwab-Stone, M., Davies, M., & Gioia, P. (1993). The diagnostic interview for children-revised version (DISC-R): Ill. Criterion validity. Journal of the American Academy of Child and Adolescent Psychiatry, 32, 658-665.

Research Unit on Pediatric Psychopharmacology (RUPP) Anxiety Study Group. (2001). Fluvoxamine for the treatment of anxiety disorders in children and adolescents. New England Journal of Medicine, 344, 1279-1285.

Richters, J. E. (1992). Depressed mothers as informants about their children. A critical review of the evidence for distortion. Psychological Bulletin, 112, 485-499.

Robins, E., & Guze, S. B. (1970). Establishment of diagnostic validity in psychiatric illness: Its application to schizophrenia. American Journal of Psychiatry, 126, 983-987.

Singer, H. S., Giuliano, J. D., Zimmerman, A. M., & Walkup, J. T. (2000). Infection: A stimulus for tic disorders. Pediatric Neurology, 22, 380-383.

Spitzer, R. L. (1983). Psychiatric diagnosis: Are clinicians still necessary? Comprehensive Psychiatry, 24, 399-411.

Sprich-Buckminster, S., Biederman, J., Milberger, S., Farnone, S. V., & Lehman, B. K. (1993). Are perinatal complications relevant to the manifestation of ADD? Issues of comorbidity and familiality. Journal of the American Academy of Child and Adolescent Psychiatry, 32, 1032-1037.

Swedo, S. E., & Kiessling, L. S. (1994). Speculations on antineuronal antibody-mediated neuropsychiatric disorders of childhood. Pediatrics, 93, 323-326.

Wozniak, J., Biederman, J., Kiely, K., Ablon, J. S., Farnone, S. V., Mundy, E., et al. (1995). Mania-like symptoms suggestive of childhood-onset bipolar disorder in clinically referred children. Journal of the American Academy of Child and Adolescent Psychiatry, 34, 867-876.

Youngstrom, E. A., Findling, R. L., & Calabrese, J. R. (2003). Who are the co-morbid adolescents? Agreement between psychiatric diagnosis, youth, parent, and teacher report. Journal of Abnormal Child Psychology, 31, 231-245.

Peter S. Jensen (1,2)

(1.) Center for Advancing Children's Mental Health, Department of Child Psychiatry, Columbia University/NYSPI, Riverside Drive, New York.

(2.) Address all correspondence to Peter S. Jensen, Center for Advancing Children's Mental Health, Department of Child Psychiatry, Columbia University/NYSPI, 1051 Riverside Drive, Unit #78, New York, New York 10032; e-mail: pj131@columbia.edu.

COPYRIGHT 2003 Plenum Publishing Corporation
COPYRIGHT 2003 Gale Group