Institut für Medizinische Biometrie
und Statistik (IMBI)

Inhalt:

Multivariable regression models are widely used in all areas of science in which empirical data are analyzed. In this lecture, we will discuss key issues of building various types of regression models such as linear regression, logistic regression, and models for survival data (Cox proportional hazards model).

We will concentrate on two components: variable selection to identify the subset of “important” variables, and identification of possible non-linearity in continuous variables. Many researchers assume a linear function for continuous variables, which may be problematic if the assumption is incorrect. This may prevent the detection of stronger effects or cause the effects to be mismodeled.

 Ad hoc ‘traditional’ approaches to variable selection have been in use for over 5 decades. Similarly, methods for determining functional forms for continuous variables were proposed many years ago. Meanwhile, many alternative approaches to address these two challenges have been developed, but knowledge of their properties and meaningful comparisons between them are scarce. We will provide an overview of variable selection procedures and discuss some open issues (Sauerbrei et al., 2020).

The multivariable fractional polynomial (MFP) approach (Royston and Sauerbrei, 2008; https://mfp.imbi.uni-freiburg.de/node/13) combines variable and functional form selection simultaneously. It is a relatively simple approach which can be understood without advanced training in statistical modeling. We will discuss key issues of MFP in details. Recently, a new R package, mfp2, was published, which implements the MFP approach. This package will be used to demonstrate MFP modelling in practice (Kipruto et al., 2023). At the end of day 3, we will provide a dataset to all participants to apply what they have learned. On day 4, we will discuss the analyses of this dataset.

In addition, we will briefly discuss the importance of good reporting, which helps in understanding the relevant steps of an analysis (Sauerbrei et al. 2023).  Participants should have a basic knowledge of linear regression models.

References

• Kipruto, E., Kammer, M., Royston, P., & Sauerbrei, W (2023). mfp2: Multivariable Fractional Polynomial Models with Extensions. R package version 1.0.0, <https://CRAN.R-project.org/package=mfp2>.
• Royston, P., & Sauerbrei, W. (2008). Multivariable model-building: a pragmatic approach to regression anaylsis based on fractional polynomials for modelling continuous variables. John Wiley & Sons
• Sauerbrei, W., Kipruto, E., & Balmford, J. (2023). Effects of influential points and sample size on the selection and replicability of multivariable fractional polynomial models. Diagnostic and Prognostic Research, 7(1), 7.
• Sauerbrei, W., Perperoglou, A., Schmid, M., Abrahamowicz, M., Becher, H., Binder, H., ... & TG2 of the STRATOS initiative (2020). State of the art in selection of variables and functional forms in multivariable analysis—outstanding issues. Diagnostic and prognostic research, 4, 1-18.