以事先合併方法估計雙二元變數相符度

Abstract

觀察者間的相符度(agreement)研究，在實務上經常遇到超過一個以上的二元變數且兩變數具有相關性的情境，如病人健康問卷(Patient Health Questionnaire, PHQ)與專科醫師的結構式問診(Structured Clinical Interview, SCI)即在篩選具憂鬱、焦慮、恐慌症等共病性的精神科疾患。傳統估計兩個具相關的二元變數相符度的統計方法是分別估計之，而Shoukri and Donner(2009)提出一個考慮相關性下以概似函數為基礎的事後合併估計方式。本研究提出一個事先合併方法，且以共同相關模式及Dirichlet-trinomial 模式建立概似函數推論過程。經由模擬討論本方法的正確度與精確度。

In many practical studies of interobserver agreement yield outcome measures on more than one binary variable, which are correlative. For example, Becker(2001) evaluated the agreement between the Structured Clinical Interview and a Patient Health Questionnaire, which are two common screening instruments of several mental diseases with high comorbidity, such as depression, anxiety, and panic disorder. Couventionally, inference on each variable are usually used. Shoukri and Donner (2009) reported a likelihood-based post-combination procedure by taking into account the correlation between diseases. Here, alternatively, we propose a pre-combination method which proceedes the likelihood inference procedure under the common correlation model and Dirichlet-trinomial model. Simulations are performed to evaluate the accuracy and precision of our method.