Relative survival analysis in R

https://doi.org/10.1016/j.cmpb.2006.01.004Get rights and content

Abstract

Relative survival techniques are used to compare the survival experience in a study cohort with the one expected should they follow the background population mortality rates. The techniques are especially useful when the cause-specific death information is not accurate or not available since they provide a measure of excess mortality in a group of patients with a certain disease. There are several approaches to modeling relative survival, but there is no widely used statistical package that would incorporate the relevant techniques. The existing software was mostly written by the authors of different methods, in different computer languages and with different requirements for the data input, which makes it almost impossible for a user to choose between available models. We describe our R package relsurv that provides functions for easy and flexible fitting of several relative survival regression models.

Introduction

Survival analysis is concerned with studying the random variable T, representing the time between entry to a study and some event of interest (e.g. death, recurrence of disease …). Instead of the cumulative distribution function F(t), we usually estimate the cumulative survival function S(t), which is defined as S(t)=P(T>t)=1F(t). In the case of the final event being death, S(t) is the probability of being alive at time t. When the final event of interest is death due to a certain disease, but the causes of death are not known, it is not possible to directly estimate the proportion of dead due to the disease in question. We then resort to the methods of relative survival. The cumulative relative survival function r(t) is defined [1] asr(t):=SO(t)SP(t)where SO(t) denotes observed survival and SP(t) stands for population or expected survival, which is estimated on the basis of population life tables. Obviously, r(t) can be any non-negative number, although the methods are most often applied to data where r(t) is less than 1.

Several approaches to modelling relative survival exist, but all of the existing programmes (for example: Surv3 [2], SAS macros and Stata functions [3], RSurvR function [4]) focus on only one of the models and use specifically organised general population tables, making it difficult for the users to compare different methods.

We present three R [5] functions organised as a package called relsurv that largely simplify the usage of relative survival regression models. All the functions require the same basic organisation of the data and can be used with any format of the population mortality data.

Section 2 briefly describes the three most commonly used regression approaches and gives an outline of the theory for the five fitting methods used in the relsurv functions. Section 3 describes the functions, their arguments, the preparation of the data and the returned values. The usage is further explained through an example in Section 4. In case the user does not yet have the relevant population mortality tables organised in the required ratetableformat, the appendix provides some quick help.

Section snippets

Relative survival regression models

The relative survival literature most frequently refers to the additive models, dominating especially in cancer research. The main assumption is that the hazard of every individual can be split into two additive components:

λO(t)=λP(t)+λE(t),where λP(t) is the hazard every patient takes because of his age, sex and cohort year (or any other combination of covariates included in the population mortality data). λE(t) denotes the excess hazard, specific for the disease in question and λO(t) is used

The relsurv package

The core of the package are three functions that fit the models described in the previous section:

  • rsadd fits an additive model. As described in Section 2, different methods of estimation exist, and the user can choose among them through the methodargument. The default is the Estève method (iii) which is specified by ”max.lik”, the other two options are method=”glm.bin”and method=”glm.poi” for models described in (i) and (ii) of Section 2. When using one of the glm methods, the observed and

Example

To illustrate the usage of the program, we use a subset of data from the study of survival of patients after acute myocardial infarction that is included in the package as the file rdata. The data were collected in the study carried out at the University Clinical Center in Ljubljana and contain 1040 patients diagnosed between 1982 and 1986 and followed up until 1997. During this time 547 deaths occurred and as the causes of death are not given, this is a good example of the need of the relative

To conclude

The usage of any statistical method depends mainly on the availability of adequate software, and relative survival techniques are no exception. For now, the choice of the model, and the method to fit it, were mostly influenced by software at hand. Different methods were programmed in different languages, and their usage was hindered by considerable effort needed to learn the special requirements of such programs. We believe that by incorporating all the different methods into an R package, now

References (14)

  • R. Giorgi et al.

    Rsurv: a function to perform relative survival analysis with S-plus or R

    Comput. Meth. Progr. Biomed.

    (2005)
  • F. Ederer et al.

    The relative survival rate: a statistical methodology

    Natl. Cancer Inst. Monogr.

    (1961)
  • Available at: http://www....
  • Available at: http://www. pauldickman....
  • R Development Core Team, R. A Language and Environment for Statistical Computing, R Foundation for Statistical...
  • P.W. Dickman et al.

    Regression models for relative survival

    J. R. Stat. Soc. Ser. C

    (2003)
  • J. Stare et al.

    An individual measure of relative survival

    J. R. Stat. Soc. Ser. C

    (2005)
There are more references available in the full text version of this article.

Cited by (153)

  • What is relative survival and what is its role in haematology?

    2023, Best Practice and Research: Clinical Haematology
  • Outcomes of surgical aortic valve replacement over three decades

    2022, Journal of Thoracic and Cardiovascular Surgery
View all citing articles on Scopus
View full text