Construction of an efficient multivariate dynamic screening system

Abstract

Recently, multivariate dynamic screening system (MDySS) has been proposed in the literature for monitoring processes whose in‐control distributions change over time. Multivariate dynamic screening system has broad applications, ranging from monitoring of dynamic engineering systems, such as nuclear reactors, airplanes, and other durable goods, to early disease detection. Conventional construction of MDySS is appropriate only in cases when sampling rate of observations is the same in Phase I and Phase II process monitoring, and the process observations are independent. In practice, these assumptions are rarely valid. In this paper, we propose a new construction of the MDySS method, which is shown to be more reliable than the conventional construction in various cases. The new construction is demonstrated using a real data from the SHARe Framingham Heart Study of the National Heart, Lung and Blood Institute.

Early detection of high disease activity in juvenile idiopathic arthritis by sequential monitoring of patients' health‐related quality of life scores

“…To accommodate serial data correlation, Li and Qiu (2016, 2017) proposed to decorrelate serial observations by Cholesky decomposition before monitoring the standardized observations. The one by Li and Qiu (2016) is on univariate cases, which is briefly described below.…”

“…The one by Li and Qiu (2016) is on univariate cases, which is briefly described below. For multivariate cases, see Li and Qiu (2017). In univariate cases when , let be the vector of all random errors in model (1) by the current time point .…”

“…This is reasonable because Table 2 presents the results of the backward elimination variable selection procedure, the variables are deleted sequentially according to their association with the response, and the variables less associated with the response would be deleted earlier. The multivariate DySS methods considered here include the chart (8)‐(9) (denoted as QX‐2S), the chart (10) (denoted as QX‐1S), their respective version based on data decorrelation by Li and Qiu (2017) (denoted as LQ‐2S and LQ‐1S), the method by simultaneously monitoring individual variables using the univariate EWMA chart (2) (denoted as SIMUL), the method by simultaneously monitoring individual variables using the nonparametric EWMA chart (7) (denoted as NP‐SIMUL), and the one‐sided multivariate CUSUM chart (11) (denoted as MCUSUM). In each of these charts, the weighting parameter λ or the allowance constant k is chosen to be , 0.2, and 0.5.…”

“…Therefore, a natural research question is whether JIA patients at the HDA stage can be detected early based on the observed HRQOL scores. Recent research in a new statistical area, called dynamic screening system (DySS), makes it possible for us to answer this question (Qiu & Xiang, 2014, 2015; Li & Qiu, 2016, 2017; Qiu, Zi, & Zou, 2018). By a DySS approach, the longitudinal pattern of the HRQOL scores for JIA patients at the “low disease activity (LDA)” stage can be first estimated properly from a dataset containing observed HRQOL scores of JIA patients whose LDA status has been confirmed by the traditional JADAS scores.…”

Dynamic Disease Screening by Joint Modelling of Survival and Longitudinal Data

“…Then, to detect the disease for a given person, his/her longitudinal pattern of the observed disease predictors is compared with the estimated regular longitudinal pattern, and a signal of disease occurrence is triggered if their cumulative difference exceeds a certain level, facilitated by a built‐in statistical process control (SPC) chart. Some modified versions of this approach can be found in Li and Qiu (2016, 2017), Qiu et al (2018) and You and Qiu (2019). However, these methods only use the information in the observed longitudinal data of the disease predictors for disease early detection, and the association between the longitudinal data and the survival outcomes related to the occurrence of the disease in question is ignored completely.…”