Beyond first-order asymptotics for Cox regression

TL;DR

This paper develops advanced statistical methods, including a parametric bootstrap and second-order techniques, to improve inference accuracy in Cox regression beyond standard first-order asymptotics, addressing issues with censoring mechanisms.

Contribution

It introduces a reference censoring model-based parametric bootstrap method and a second-order approach for Cox regression, enhancing inference accuracy beyond first-order methods.

Findings

Bootstrap method improves p-value accuracy

Second-order method offers computational efficiency

Methods are robust to censoring model assumptions

Abstract

To go beyond standard first-order asymptotics for Cox regression, we develop parametric bootstrap and second-order methods. In general, computation of $P$-values beyond first order requires more model specification than is required for the likelihood function. It is problematic to specify a censoring mechanism to be taken very seriously in detail, and it appears that conditioning on censoring is not a viable alternative to that. We circumvent this matter by employing a reference censoring model, matching the extent and timing of observed censoring. Our primary proposal is a parametric bootstrap method utilizing this reference censoring model to simulate inferential repetitions of the experiment. It is shown that the most important part of improvement on first-order methods - that pertaining to fitting nuisance parameters - is insensitive to the assumed censoring model. This is supported by numerical comparisons of our proposal to parametric bootstrap methods based on usual random censoring models, which are far more unattractive to implement. As an alternative to our primary proposal, we provide a second-order method requiring less computing effort while providing more insight into the nature of improvement on first-order methods. However, the parametric bootstrap method is more transparent, and hence is our primary proposal. Indications are that first-order partial likelihood methods are usually adequate in practice, so we are not advocating routine use of the proposed methods. It is however useful to see how best to check on first-order approximations, or improve on them, when this is expressly desired.