観察者間の診断の一致性を評価する頑健な統計量AC₁について

抄録

Understanding inter-observer variability in clinical diagnosis is crucial for reliability studies. As the statistical measurements of reliability, the kappa statistic and its extensions have been widely adopted in medical research, but it has been discussed that kappa is vulnerable to prevalence and presence of bias. As an alternative robust statistic, AC₁ has attracted recent statistical attentions. This article describes fundamental ideas and quantitative features of AC₁. The reliability of infrared thermoscanner as an application in detecting febrile patients of pandemic influenza is discussed by means of Monte Carlo simulation. AC₁ adjusts chance agreement more appropriately than kappa and is regarded as a more useful measurement for assessing inter-observer agreement, especially when prevalence is small.