Research Planning Launch/Methods Conference (February 18-20, 2004)
The research planning conference series began with a Launch/Methods Conference that was held at the Natcher Center on the campus of the National Institutes of Health on February 18-20, 2004. This initial conference involved more than 65 investigators representing the Core Working Groups for the ten upcoming research conferences. The overall aims of the Launch conference were to initiate the process of planning for the subsequent research conferences (i.e., selection of specific topics and participants) and to provide attendees with an overview of cutting edge statistical approaches to apply to analyses of existing and emerging data from diverse fields.
Early versions of the Diagnostic and Statistical Manuals (DSMs) were based largely on clinical consensus. DSM-III and DSM–III-R placed emphasis on observable/measurable signs and symptoms and, in turn, on the reliability of psychiatric diagnoses. Although DSM-IV maintained the focus on reliability through its increased attention to empirical evidence, the fact that work on DSM-IV started so soon after the publication of DSM-III-R limited opportunities for sophisticated statistical analyses of new and existing data. In order to avoid similar constraints in the development of DSM-V and to encourage more effective involvement of statistical methodologists in the DSM-V development process, a series of presentations about innovative analytic methods was incorporated into the Launch conference.
The Methodology component of the launch conference was chaired by Dr. Helena Kraemer, Professor of Biostatistics in Psychiatry at Stanford University, and Dr. Patrick Shrout, Professor of Psychology at New York University, in collaboration with Dr. Maritza Rubio-Stipec, consultant to APIRE. Unlike the ten successive conferences that will primarily focus on a specific diagnostic topic, the Methods session was designed to examine a broad array of statistical methods and techniques that can be utilized to capture, examine, interpret, and, when useful, synthesize disparate datasets. Overall, the conference afforded statisticians with long experience in psychiatric research the opportunity to propose and debate the types of problems they foresee could arise across the various diagnostic categories and to discuss, in non-mathematical language, some analytic approaches to such problems. The longer term objective was to initiate a dialogue among statistical experts, methodologists, and psychiatric researchers that will continue to enrich future diagnosis-related basic and clinical research. During breakout sessions, the Core Working Group members developed initial lists of topics to be covered during their conferences and also had the opportunity to meet with the methodologists, who have expressed willingness to be available as consultants on an as-needed basis to each of the research conferences held over the next four years.
Several methodological issues were addressed. A forthcoming paper from Kraemer and Shrout, to be published in a peer-reviewed journal, will provide an in-depth discussion. Some of the more important issues that were discussed include the following:
- On the controversy between categorical vs. dimensional approaches to diagnostic classifications, the take home message was that, fundamentally, both are needed, and both can be developed When highly reliable and valid diagnoses are developed, the issue of categorical as opposed to dimensional will be moot; both approaches can be applied to any diagnosis. The overriding methodological issues are how to demonstrate high reliability and validity, and how best to identify when utilization of a categorical or dimensional approach to a diagnosis is most appropriate. For reimbursement purposes, for example, a categorical diagnosis likely will be optimal, whereas in assessing treatment outcome, dimensional diagnoses may yield more precise information.
- Comorbidity among psychiatric disorders is the rule rather than the exception. The expert statistical consultants underscored the importance of distinguishing between types of co-morbidities and understanding their implications for the definition of a psychiatric disorder. Because the boundaries of disorders can be rather “fuzzy,” different types of co-morbidity bring different information to the revision of the nosology. Epidemiological comorbidity (non-random co-occurrence) may indicate two different expressions of one disease process, or one disease that leads to another; alternatively, it could indicate two distinct disorders that have a common etiology. Clinical comorbidity reflects instances wherein the expression, course, or response to treatment of one disorder may vary as a function of the presence of another disorder. Familial comorbidity, in which patterns of familial transmission may vary depending on whether another distinct disorder is present or not, could be indicative of familial etiology of the disorder (genetic or not).
- Identifying the true underlying disease presents many methodological challenges. There is no “gold standard” against which diagnostic criteria can be assessed; rather, there are only multiple external validators that reflect what is known about a disorder from past and ongoing research (such as course of illness, family history, treatment response), and multiple internal factors that suggest the presence of a latent construct (such as the presence of irritability, sleep disturbances, cognitive slowing, etc.). Conference participants discussed different statistical approaches that can be used to address the issue of validity using existing data.
- When designing studies to generate data putatively relevant to the revision of a classification system, the study design must explore the trade-offs associated with variations in sample size, number of informants, categorical, and dimensional variables. The statistical experts presented and discussed various approaches that can be used to identify the most efficient study design for a given diagnostic question.
- In light of the fact that diverse types of study designs can be used in revising a classification system, it is critically important to begin with a conceptual model for addressing a particular nosologic question that is built on and integrates valid psychiatric assumptions as well as appropriate statistical analyses. Mental health researchers and statisticians must therefore collaborate in any revision of a nosology.