CoxPhLb: An R Package for Analyzing Length Biased Data under Cox Model

Chi,Hyun Lee,Heng Zhou,Yu Shen,Diane,D Liu,Jing Ning

doi:10.32614/rj-2020-024

Abstract

Data subject to length-biased sampling are frequently encountered in various applications including prevalent cohort studies and are considered as a special case of left-truncated data under the stationarity assumption. Many semiparametric regression methods have been proposed for length-biased data to model the association between covariates and the survival outcome of interest. In this paper, we present a brief review of the statistical methodologies established for the analysis of length-biased data under the Cox model, which is the most commonly adopted semiparametric model, and introduce an R package CoxPhLb that implements these methods. Specifically, the package includes features such as fitting the Cox model to explore covariate effects on survival times and checking the proportional hazards model assumptions and the stationarity assumption. We illustrate usage of the package with a simulated data example and a real dataset, the Channing House data, which are publicly available.

Highlights

In prevalent cohort studies, subjects who have experienced an initiating event but have not yet experienced a failure event are sampled from the target population and followed until a failure or censoring event occurs
Given that the data are subject to length bias, we evaluate the covariate effects on the failure time under the Cox model using the estimation method for length-biased data
While statistical methodologies have been well established for analyzing such types of failure time data, the lack of readily available software has been a barrier to the implementation of proper methods

Summary

Introduction

Subjects who have experienced an initiating event (e.g., disease diagnosis) but have not yet experienced a failure event (e.g., death) are sampled from the target population and followed until a failure or censoring event occurs. We outline how the Cox proportional hazards model assumptions can be checked both graphically and analytically, and describe two approaches to test the stationarity of the underlying incidence process. Among many estimation methods established for length-biased data under the Cox model, we provide the estimation function based on the inversely weighted estimating equation of Qin and Shen (2010).

Results

Conclusion