When analyzing the survival of a given cohort, it is often the case that a significant number of observed deaths do not occur due to the treated disease; but rather, due to other causes. It may be of interest to estimate the proportion of these deaths and to interpret them accordingly. Whenever the cause of death information is not available in the data, this estimation is not straightforward. The subfield of survival analysis that deals with this issue is relative survival, its main idea is to incorporate the population mortality information using external population mortality tables. In this doctoral thesis, we develop two extensions of the currently available methodology: we consider population mortality in multi-state models and life years difference analysis.
Multi-state models provide an extension of the usual survival analysis setting. In the medical domain, multi-state models give the possibility of further investigating intermediate events such as relapse and remission. In this work, a further extension is proposed using relative survival, where mortality due to population causes (i.e. non-disease-related mortality) is evaluated. The objective is to split the total mortality in disease and non-disease-related mortality, with and without intermediate events. Precise definitions and suitable non-parametric estimators are provided for both transition hazards and probabilities. Variance estimating techniques and confidence intervals are introduced and the behaviour of the new methodology is investigated through simulations. The newly developed work is implemented in the R package mstate.
Our second goal is to consider the number of years lost/saved for a given cohort with long-term follow-up. This measure has a simple and appealing interpretation which is also intuitive for non-statisticians. However, the theoretical aspects of this measure have not yet been fully studied. In this work, we consider three possible definitions of the measure, their differences, assumptions, characteristics and the corresponding non-parametric estimators. We also study variance estimation and consider some of the challenges that might occur in practice. Finally, an efficient R implementation in the package relsurv is developed for all of the three measures which makes this work easily available to subsequent users.
We believe that the two proposed approaches can provide better understanding of the survival data whenever population mortality is not negligible.
|