The response-aided clustering, multi-scale Bayesian change point detection, and error contaminated functional quantile regression with the applications to the BOSS data.

Professor Yin, Guosheng   (Principal Investigator (PI))

Dr Jiang Fei   (Co-Investigator)

36

2018-07-15

456452

The response-aided clustering, multi-scale Bayesian change point detection, and error contaminated functional quantile regression with the applications to the BOSS data.

Bayesian change point, Haar wavelet, measurement error, non-local prior, response-aided

Probability & Statistics,Blood/Hematology

Physical Sciences (P)

17307218

General Research Fund (GRF)

2018

Completed

1 To develop estimation and clustering procedures for grouping the continuously monitored blood pressure curves whiling taking into account the outcome information. The semiparametric efficient estimating equation will be used for the parameter estimation. The clustering step will apply the k-means centering algorithm introduced in Chiou and Li (2007). The goal is to develop a method that will yield clustering labels that are consistent with the patients’ outcomes. Further, we will study whether including the patients’ outcomes improves the clustering efficiency. 2 To develop a multi-scale change point detection method to identify the change points in the blood pressure curves. We will use the Haar wavelet to transform the blood pressure sequence in different scales. The multi-scale procedure allows us to evaluate the blood pressure changes in different resolutions. Further, the non-local priors help to reduce the false discovery of the change points. We will introduce a complete multi-scale Bayesian framework for the change point detection, including identification, post-processing, and selection of the optimal number of change points. The procedure can be used in the early prevention of the stroke onset. 3 To develop a censored regression model between the stroke recurrent survival time and the continuous monitored blood pressure. The blood pressure curves are measured with missing segments. We propose to impute the missing segments through the B-spline method. After the imputation, the resulting curves are contaminated by the imputation errors. Hence, we propose an estimation under the measurement error model. The procedure will be used to identify the patients with high risk of stroke recurrence and adjust the medicare plan according.