Inference for changepoint survival models

TL;DR

This paper develops new methods for estimating changepoints in non-proportional hazards survival models, using score processes and quadratic residuals, with simulations and real data illustration.

Contribution

It introduces a novel estimation approach for multiple changepoints in survival models based on standardized score processes and residual minimization.

Findings

Confidence regions for changepoints are derived.

The new method performs well in simulations.

Application to real data demonstrates practical utility.

Abstract

We consider a non-proportional hazards model where the regression coefficient is not constant but piecewise constant. Following Andersen and Gill (1982), we know that a knowledge of the changepoint leads to a relatively straightforward estimation of the regression coefficients on either side of the changepoint. Between adjacent changepoints, we place ourselves under the proportional hazards model. We can then maximize the partial likelihood to obtain a consistent estimation of the regression coefficients. Difficulties occur when we want to estimate these changepoints. We obtain a confidence region for the changepoint, under a two-step regression model (Anderson and Senthilselvan, 1982), based on the work of Davies (1977). Then we introduce a new estimation method using the standardized score process (Chauvel and O'Quigley, 2014), under a model with multiple changepoints. In this context, rather than make use of the partial likelihood, we base inference on minimization of quadratic residuals. Simulations and an example are provided.