ACTUARIAL STRATEGIES IN CHILD DIAGNOSTIC ASSESSMENT

University Of Hawaii At Manoa, Honolulu HI

Investigators

Linked publications & trials

Abstract

The proposed project involves a preliminary investigation of potentially significant methodological advance in diagnostic assessment. The current state-of-the-art in diagnostic assessment involves the use of a structured interview. Typically, structured interviews involve a static skip structure, i.e., some portions of the interview are administered conditional on particular responses to prior questions. For example, if there is a negative response to a question about depression and anhedonia, most structured interviews require the clinical to skip the remaining questions about associated symptoms (e.g., sleep disturbance, impaired concentration, etc.). Although structured interviews represent an enormous advantage over earlier diagnostic procedures, their inflexible structure is often incompatible with the heterogeneity of most child and adolescent populations, and can result in superfluous questioning about uncommon disorders and insufficient follow-up about more common ones. Many interviews do not make exceptions for individual characteristics. For example, 1) a 17 year old boy might need to answer "no" to 5 or 6 questions about separation anxiety before the interviewer may move on to another set of questions; or 2) an underweight 16 year old girl might not be asked important follow-up questions when replying "no" to the initial question about eating disorders. One might conclude that introducing more clinician flexibility would be the solution; however, the literature on clinical judgment suggests that increasing clinician involvement in determination of interview structure would likely degrade classification accuracy and introduce unwanted sources of error and bias. To address this issue in another manner, the principal investigator has developed a data-driven, actuarial expert system to guide a flexible interview structure. Thus, interview structure is dynamically responsive to individual characteristics, without introducing error associated with qualitative clinical judgments. Pilot modeling revealed that his system offers advantages in classification accuracy over state-of-the-art diagnostic approaches, with the additional benefit of reducing administration time for particular disorders. The current project is planned to generate requisite data to develop a formal expert system and to forecast its relative accuracy and efficiency in a child and adolescent population. It is predicted that this system will demonstrate improvements in classification accuracy over a static structured interview approach, with reduced administration time. If the data are supportive, these developments have the potential to significantly advance the manner in which future diagnostic interviews are conducted with mental health populations.