Semiparametric Multivariate Accelerated Failure Time Model with Generalized Estimating Equations

TL;DR

This paper introduces a generalized estimating equation approach for semiparametric multivariate accelerated failure time models, improving computational efficiency and handling complex dependence structures, with proven consistency and efficiency gains.

Contribution

It proposes a novel GEE-based method for multivariate AFT models that accommodates various dependence structures and simplifies computation using induced smoothing.

Findings

Estimator is up to three times more efficient than initial rank-based estimator.

Method performs well with stronger dependence and heavier censoring.

Real data examples demonstrate practical utility.

Abstract

The semiparametric accelerated failure time model is not as widely used as the Cox relative risk model mainly due to computational difficulties. Recent developments in least squares estimation and induced smoothing estimating equations provide promising tools to make the accelerate failure time models more attractive in practice. For semiparametric multivariate accelerated failure time models, we propose a generalized estimating equation approach to account for the multivariate dependence through working correlation structures. The marginal error distributions can be either identical as in sequential event settings or different as in parallel event settings. Some regression coefficients can be shared across margins as needed. The initial estimator is a rank-based estimator with Gehan's weight, but obtained from an induced smoothing approach with computation ease. The resulting estimator is consistent and asymptotically normal, with a variance estimated through a multiplier resampling method. In a simulation study, our estimator was up to three times as efficient as the initial estimator, especially with stronger multivariate dependence and heavier censoring percentage. Two real examples demonstrate the utility of the proposed method.