B1626
Title: Copula based quantile modelling under dependent censoring
Authors: Myrthe D Haen - Hasselt University (Belgium)
Ingrid Van Keilegom - KU Leuven (Belgium)
Anneleen Verhasselt - Hasselt University (Belgium) [presenting]
Abstract: In survival analysis under random right censoring, one may observe a censoring time C for some values rather than the survival time T. Often, such censoring is dealt with under an independence assumption on T and C, given the covariate X. However, in some cases, this may not be a very realistic assumption; by taking care of the possible dependency, inference on the survival time could be handled more accurately. In this research, this inference for T is done with a focus on quantile regression, but some broader regression results are obtained as a by-product. In order to capture any dependence, the quantile model for T is derived from a bivariate copula model for (T, C). For this copula model, a flexible copula parameter is taken to deal with the practice of often unknown association. It comes at the cost of marginals that are necessarily fully parametric, but this can be overcome by considering the family of so-called enriched asymmetric Laplace (EAL) distributions for T. While preserving the parametric character, they enable introducing sufficient modelling flexibility by means of Laguerre orthogonal polynomials. The identifiability of the bivariate model is shown, comprising all parameters for T rather than only its quantiles. In this sense, the scope of distributional regression is also captured.
