Article
The influence of categorising survival time on selection and estimation of time-varying effects in an extended Cox model
Search Medline for
Authors
Published: | September 20, 2011 |
---|
Outline
Text
Background: With longer follow-up times, the proportional hazards assumption is questionable in the Cox model. Cox already suggested to include an interaction between a covariate and a function of time, i.e. a time-varying effect. Since then, several approaches have been proposed for modelling time-varying, some of which require a substantial enlargement of the data. When investigating large data sets, e.g. large studies or registry data, this may cause severe computational problems.
Methods: Categorisation of survival times is an easy way to handle such problems. However, it raises issues as to the number of cutpoints, their position, the increased number of ties and the loss of information. One specific approach requiring such an enlargement of data is the MFPT approach [1], which combines variable selection, selection of functional forms of (continuous) covariates and selection of time-varying effects.
Results/Discussion: Based on the example of Whitehall I, a large (N=17,260) epidemiological cohort study, we will demonstrate that computational difficulties can be suitably handled by categorisation to about 50 to 100 distinct event times without severely influencing the results. In addition, results of a simulation study on the selection and estimation of time-varying effects will be presented.
References
- 1.
- Sauerbrei W, Royston P, Look M. A new proposal for multivariable modelling of time-varying effects in survival data based on fractional polynomial time-transformation. Biometrical Journal. 2007;49:453-73.
- 2.
- Buchholz A, Sauerbrei W, Royston P. Modelling time-varying effects in large survival data sets may require categorisation of time: Does it influence the results of the MFPT approach? [submitted]. 2011.