A2078
Title: Deep nonparametric inference for conditional mean models with panel count data
Authors: Anyin Feng - The Hong Kong Polytechnic University Shenzhen Research Institute (China) [presenting]
Qiang Wu - The Hong Kong Polytechnic University (Hong Kong)
Xiangbin Hu - Beijing Institute of Technology (China)
Xingqiu Zhao - The Hong Kong Polytechnic University (Hong Kong)
Abstract: In the statistical analysis of panel count data, semiparametric models such as the proportional mean model and the accelerated mean model are often used to investigate the occurrence rate of recurrent events. Misspecification of the prespecified model may lead to biased estimation and poor predictive performance. Therefore, it is essential to develop hypothesis testing methods to assess the adequacy of the assumed model. However, most existing diagnostic methods rely on graphical tools and lack theoretical justification. To address this issue, a hypothesis testing method to assess model adequacy with theoretical support is developed. Specifically, a deep extended mean model is proposed that integrates commonly used nonparametric and semiparametric models within a general framework. Treating the proposed model as the encompassing model, the hypothesis test is constructed based on the discrepancy between the parameter estimates obtained under the encompassing and prespecified models. Extensive simulation studies and an application to data from a skin cancer chemoprevention trial demonstrate the superior performance of the proposed method.
