Mental Health Computerized Adaptive Testing

描述(申请人提供)：心理健康研究在很大程度上依赖于过时的测量系统。传统心理健康量表的构建在很大程度上是基于主观判断，充其量也就是应用经典测试理论中的方法来确定量表的心理测量学特性。在这一应用中，我们借鉴了在教育测量和现代心理测量理论领域中发展起来的考试建设和管理方面的主要进展。特别是，我们建议使用项目反应理论(IRT)来校准由626个情绪障碍项目组成的大型项目库，然后自适应地管理它们，以便可以在项目的一个小子集上对给定的主题进行评估，以达到任何实际的准确度。IRT模型的使用使我们能够评估不同受试者情绪障碍的强度，这些受试者从项目库中选择了可能不同数量的项目。使用计算机化自适应测试(CAT)，我们可以根据每个受试者对先前项目的反应，从表征低到高损害程度的一小组项目开始，自适应地为每个受试者选择最合适的项目集。最终的结果是，可以开发出能够彻底表征一种特定疾病的大型“题库”。虽然通常不可能对任何一个受试者对所有项目进行评估，但CAT允许对每个受试者对整个项目池中的一小部分进行评估，而信息损失最小且可控。 将IRT应用于心理健康测量问题的一个复杂之处在于，与传统的能力测试(如数学成绩)不同，心理健康测量量表本质上是多维的。虽然多维IRT模型已经存在，但在适应性测试的背景下，它们还没有得到很好的研究。值得注意的是，心理健康测量量表具有多维性的主要原因之一是，项目通常是从多个领域(如各种情绪障碍)抽样的，因此违反了一维IRT模型的假设。为此，Gibbons和Hedeker(1992)发展了一个“项目双因素”模型，该模型允许每个项目加载到一个主要维度(例如，抑郁)和一个子域(例如，睡眠障碍)。在心理健康测量的背景下，高因素模型的优势在于它产生了一种可以作为适应性测试重点的总体损害测量。在这项申请中，我们建议对CAT和IRT在心理健康测量工具的校准和管理中的使用进行深入的可行性研究。具体地说，我们建议开发必要的统计理论和软件，然后将其应用于情绪焦虑谱系障碍量表。

DESCRIPTION (provided by applicant): Mental health research relies heavily on antiquated systems of measurement. The construction of traditional mental health scales is based largely on subjective judgment, and at best, application of methods from classical test theory to determine a scale's psychometric properties. In this application we borrow strength from major advances in test construction and administration that have been developed in the fields of educational measurement and modem psychometric theory. In particular, we propose to use Item Response Theory (IRT) to calibrate a large item pool of 626 mood disorder items and then adaptively administer them such that a given subject can be evaluated on a small subset of the items to any practical degree of accuracy. The use of the IRT model allows us to evaluate the intensity of the mood disorder for different subjects who have taken potentially different numbers of items selected from the item pool. Using computerized adaptive testing (CAT) we can then adaptively select the most appropriate set of items for each subject based on his/her responses to previous items, beginning from a small screening set of items that characterize low to high levels of impairment. The net result is that large "item banks" can be developed that thoroughly characterize a particular disorder. Although it is not routinely possible for any one subject to be evaluated on all of the items, CAT permits each subject to be evaluated on a small subset of the total item pool, with minimal and controllable loss of information. A complication of applying IRT to mental health measurement problems is that unlike traditional ability testing (e.g., mathematics achievement) which are inherently unidimensional, mental health measurement scales are inherently multidimensional. Although multidimensional IRT models are available they have not been well studied in the Context of adaptive testing. Note that one of the primary reasons for the multidimensionality of mental health measurement scales is that the items are often sampled from multiple domains (e.g., various mood disorders), thereby violating the assumption of a unidimensional IRT model. To this end, Gibbons and Hedeker (1992) developed an "item bi-factor" model which allows each item to load on a primary dimension (e.g., depression) and one subdomain (e.g., sleep disturbance). In the context of mental health measurement, the advantage of the hi-factor model is that it yields a measure of overall impairment that can be the focus of adaptive testing. In this application, we propose to provide an in depth feasibility study of the use of CAT and IRT in the calibration and administration of mental health measurement instruments. Specifically, we propose to develop the statistical theory and software that is necessary, and to then apply it to the Mood-Anxiety Spectrum Disorders Scale.