Inference for probability of selection with dependently truncated data using a Cox model

ABSTRACT

A truncated sample consists of realizations of two variables L and T subject to the constraint L < T. One simple solution to dependently truncated data is to take L as a covariate of T in the Cox model. We aimed at studying the probability of selection, P(L < T), in this framework. We proposed the point estimator and derived its asymptotic distribution. Both truncated-only data and censored and truncated data were generated in the simulation study. The proposed point and variance estimators showed good performance in various simulated settings. The bone marrow transplant registry data were analyzed as the illustrative example.

KEYWORDS:

• Dependently truncated data
• Inverse probability weighting
• Quasi-independence

MATHEMATICS SUBJECT CLASSIFICATION:

• 62N99