ITEM RESPONSE THEORY AND THE MOS SF-36

在成本控制和不断变化的医疗保健提供系统的时代，对价值购买的需求大大增加了自我报告的健康状况和生活质量信息在临床和政策决策中的使用。虽然有许多工具来测量这些结构，测量水平及其对科学推理和分数解释的影响通常被忽视。项目反应理论（IRT）等测量模型在教育心理学、社会学和政治学等领域得到了成功的应用，健康状况研究者也开始倡导将其用于健康状况评估。然而，只有有限的实证研究，他们的健康状况测量的适用性进行了。本研究将调查IRT模型在产生精确的健康估计的有用性，通过将这些模型应用于特定条件（哮喘，终末期肾病）和正常（一般人群）研究的独立样本。本研究将重点关注广泛使用的健康状况测量的身体功能量表，即医学结局研究简表36（SF-36）。主要的IRT假设的单维性将通过因素分析和项目参数的稳定性检查进行测试。 还将通过拟合优度检验和残差分析来检查双参数和多分类IRT模型的贡献。 将通过比较测量误差、测试信息、组间差异和随时间变化的分数大小和变异性以及预测有效性来检查不同测量模型（总结性、IRT）对总分的影响。这项研究将产生经验证据的项目反应模型估计健康状况的有用性。 此外，研究结果可用于指导研究人员选择最有用的模型来对其措施进行评分，并从数据中做出有效的科学推断和解释。 实际上，通过减少测量误差，更高的测量精度可以转化为更低的研究成本。最终，更好的测量特性将使更精确的理论问题得到测试，并做出更准确的临床和政策预测。

The demand for value purchasing in an era of cost-containment and changing health care delivery systems has dramatically increased the use of self-reported health status and quality of life information in clinical and policy decision-making. Although numerous instruments for measuring these constructs are available, measurement level and its influence on scientific inference and score interpretation have generally been overlooked. Measurement models, such a item response theory (IRT), have been used successfully in fields such as education psychology, sociology and political science, and health status researchers are beginning to advocate their use in health status assessment. However, only limited empirical investigation of their applicability in health status measurement have been performed. This study will investigate the usefulness of IRT models in generating precise estimates of health by applying these models on independent samples from condition-specific (asthma, end-stage renal disease) and norming (general population) studies. This study will focus on the physical functioning scale of the widely used health status measure, the Medical Outcomes Study Short-Form 36 (SF-36). The major IRT assumption of unidimensionality will be tested through factor analyses and an examination of item parameter stability. The contribution of two parameter and polytomous IRT models will also be examined through goodness-of-fit tests an analyses of residuals. The influence of different measurement models (summative, IRT) on aggregate scores will be examined through comparisons of errors of measurement, test information, magnitude and variability of scores for group differences and change over time, and predictive validity. This study will generate empirical evidence of the usefulness of item response models estimating health status. In addition, study results may be used to guide researchers selecting the most useful model for scoring their measures and in making valid scientific inferences and interpretations from their data. Practically, greater measurement precision could translate into lower costs of conducting research, by reducing measurement error. Ultimately, better measurement properties will enable more precise theoretical questions to be tested and more accurate clinical and policy predictions to be made.