Nonparametric dynamic screening system for monitoring correlated longitudinal data

Abstract

In many applications, including the early detection and prevention of diseases and performance evaluation of airplanes and other durable products, we need to sequentially monitor the longitudinal pattern of certain performance variables of a subject. A signal should be given as soon as possible after the pattern has become abnormal. Recently, a new statistical method, called a dynamic screening system (DySS), was proposed to solve this problem. It is a combination of longitudinal data analysis and statistical process control. However, the current DySS method can only handle cases where the observations are normally distributed and within-subject observations are independent or follow a specific time series model (e.g., AR(1) model). In this article, we propose a new nonparametric DySS method that can handle cases where the observation distribution and the correlation among within-subject observations are arbitrary. Therefore, it significantly broadens the application area of the DySS method. Numerical studies show that the new method works well in practice.