混合分布モデルによる正常値域(参考値域)の推定法とその実際

Abstract

In this paper we propose a method of estimating the normal(or reference)ranges of blood constituents, assuming that each distribution of their measurements is a mixture of normal and abnormal components. Values of the parameters of the mixture distribution are determined so as to minimize ?O2 called as the Cramer-von Mises criterion. Because this method of estimation is less disturbed by outliers than the maximum likelihood method, it is suitable for medical data analysis and is applicable both to data of apparently healthy subjects and to those of patients. The estimation results based on the hospital data are illustrated on seventeen items of blood constituents by their histogram and estimated curves.