A0789
Title: Bayesian scalar on image regression with non-ignorable non-response
Authors: Xiangnan Feng - The Chinese University of Hong Kong (Hong Kong)
Tengfei Li - University of Texas MD Anderson Cancer Center (United States)
Xinyuan Song - Chinese University of Hong Kong (Hong Kong) [presenting]
Hongtu Zhu - University of Texas MD Anderson Cancer Center (United States)
Abstract: Medical imaging data have been widely used in modern health care, particularly in the prognosis, screening, diagnosis, and treatment of various diseases. We consider a scalar-on-image regression model that uses ultrahigh dimensional imaging data as explanatory covariates. The model is used to investigate important risk factors for the scalar response of interest, which is subject to non-ignorable missingness. We propose the use of an efficient functional principle component analysis method to reduce the dimensions of the imaging observations. Given that non-ignorable non-response distorts the accuracy of statistical inference and generates misleading results, we propose an imaging exponential tilting model for the examination of the potential influence of imaging observations along with scalar variables on the probability of missingness. An instrumental variable, such as a covariate associated with the response but conditionally independent of the probability of missingness, is introduced to facilitate model identifiability. Statistical inference is conducted in a Bayesian framework with Markov chain Monte Carlo algorithms. Simulation studies show that the proposed method exhibits satisfactory finite sample performance. The methodology is applied to a study on the Alzheimer's Disease Neuroimaging Initiative dataset.
