Article
On the incorporation of model selection uncertainty into prognostic models for survival data
Search Medline for
Authors
-
Anika Buchholz
- Universitätsklinikum Freiburg, Freiburg
-
Norbert Holländer
- Universitätsklinikum Freiburg, Freiburg
-
Willi Sauerbrei
- Universitätsklinikum Freiburg, Freiburg
| Published: | September 8, 2005 |
|---|
Outline
Text
Introduction
Predictions of disease outcome in prognostic models are usually based on a single selected model. However, often several models fit data equally well, but these models might differ substantially in terms of included explanatory variables and might lead to different predictions for individual patients. Ignoring the uncertainty caused by the model selection process results in the underestimation of the variance of a predictor, which is obtained by a single model.
Subject and methods
For survival data, we discuss two approaches to account for model selection uncertainty, with the main emphasis on variable selection in a Cox model. The approaches are based on Bayesian model averaging [Ref. 1] and bootstrap model averaging [Ref. 2], respectively. In data examples, we compare predictions of the approaches and discuss issues of interpretation and practical utility.
Results
The approach based on the bootstrap eliminates variables hardly supported by the data. It gives results which are easier to interpret and which have a higher practical utility.
Discussion
The bootstrap model averaging approach has several advantages for practical use. For both model averaging approaches simulation studies are required to get further insight into the properties.
References
- 1.
- Volinsky CT, Madigan D, Raftery AE, Kronmal RA. Bayesian model averaging in proportional hazard models: Assessing the risk of a stroke. JRSS Series C 1997; 46: 433-48
- 2.
- Augustin N, Sauerbrei W, Schumacher M. The practical utility of incorporating model selection uncertainty into prognostic models for survival data. Statistical Modelling 2005, in press.
