Estimation of Extreme Survival Probabilities with Cox Model

TL;DR

This paper extends the Cox proportional hazards model to better estimate the probabilities of rare, extreme survival events, especially in heavily censored data, by incorporating tail modeling with Pareto distributions.

Contribution

It introduces a novel method combining Cox models with tail estimation via Pareto distribution, allowing more reliable prediction of rare event probabilities beyond observed data.

Findings

Effective tail estimation in heavily censored data

Automatic threshold selection procedure

Application to real datasets demonstrates improved estimates

Abstract

We propose an extension of the regular Cox's proportional hazards model which allows the estimation of the probabilities of rare events. It is known that when the data are heavily censored at the upper end of the survival distribution, the estimation of the tail of the survival distribution is not reliable. To estimate the distribution beyond the last observed data, we suppose that the survival data are in the domain of attraction of the Fr\'echet distribution conditionally to covariates. Under this condition, by the Fisher-Tippett-Gnedenko theorem, the tail of the baseline distribution can be adjusted by a Pareto distribution with parameter $\theta$ beyond a threshold $\tau$. The survival distributions conditioned to the covariates are easily computed from the baseline. We also propose an aggregated estimate of the survival probabilities. A procedure allowing an automatic choice of the threshold and an application on two data sets are given.