The shared weighted Lindley frailty model for cluster failure time data

TL;DR

This paper introduces a new frailty model based on the weighted Lindley distribution for clustered survival data, providing a flexible way to model unobserved heterogeneity with an EM algorithm for estimation.

Contribution

The paper proposes the first weighted Lindley frailty model for clustered survival data, with parametric and semiparametric forms and an R package implementation.

Findings

The WL frailty model outperforms classical models in real data analysis.

Simulation studies show good finite sample performance.

The model effectively captures unobserved heterogeneity in survival data.

Abstract

The primary goal of this paper is to introduce a novel frailty model based on the weighted Lindley (WL) distribution for modeling clustered survival data. We study the statistical properties of the proposed model. In particular, the amount of unobserved heterogeneity is directly parameterized on the variance of the frailty distribution such as gamma and inverse Gaussian frailty models. Parametric and semiparametric versions of the WL frailty model are studied. A simple expectation-maximization (EM) algorithm is proposed for parameter estimation. Simulation studies are conducted to evaluate its finite sample performance. Finally, we apply the proposed model to a real data set to analyze times after surgery in patients diagnosed with colorectal cancer and compare our results with classical frailty models carried out in this application, which shows the superiority of the proposed model. We implement an R package that includes estimation for fitting the proposed model based on the EM-algorithm.