gms | German Medical Science

MAINZ//2011: 56. GMDS-Jahrestagung und 6. DGEpi-Jahrestagung

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V.
Deutsche Gesellschaft für Epidemiologie e. V.

26. - 29.09.2011 in Mainz

The reliability of penalized models in survival analysis. Using the value first deritative parametrization in the penalized likelihood framework

Meeting Abstract

Search Medline for

  • Ronny Westerman - Philipps-Universität Marburg, Marburg
  • Katharina Belting - Philipps-Universität, Marburg
  • Ulrich Mueller - Philipps-Universität, Marburg

Mainz//2011. 56. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (gmds), 6. Jahrestagung der Deutschen Gesellschaft für Epidemiologie (DGEpi). Mainz, 26.-29.09.2011. Düsseldorf: German Medical Science GMS Publishing House; 2011. Doc11gmds096

DOI: 10.3205/11gmds096, URN: urn:nbn:de:0183-11gmds0961

Published: September 20, 2011

© 2011 Westerman et al.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.en). You are free: to Share – to copy, distribute and transmit the work, provided the original author and source are credited.


Outline

Text

Introduction: In Survival analysis the parametric assumption seems to be too restrictive for many practical applications. Nonparametric techniques will be often seen as an alternative to overcome such problems in limitations. The motivation using penalized splines (P-splines) yields on the compromise between smoothing and regression splines. One problem involving penalized splines is considered with use of single penalties. With this abstract one want to apply the value-first derivative parametrization (VFDP) introduced by Costa et al. 2008 for sperm count data. The convenience of VFDP takes account for the right parametrization in the penalties and the optional use for double penalties.

Data and methods: The appropriated dataset enfold more than 2.000 sperm count records from the department of andrology at Marburg University Hospital Germany were sampled between 1949 and 1998. For analyzing the statistical association between male sperm count and survival one conduct the Mortality-follow-up until December 31th 2006. The main purpose of this abstract is aimed on the comparison of the performance for certain penalty parametrization (single or double) as nonparametric fashion in proportional hazard context. Analysis and simulation will be also done with R package gam.models for the generalized additive model (GAM) in regarding to multiple quadratic penalties.

Results: Refer to results from simulation study one could not determine significant differences on survival for fertile men and males with lower sperm counts. The smoothness for the double penalty condition yield on slightly better convergence in comparision to single penalty but the disparities between the two options should be quiet small.


References

1.
Brown D, Kauermann G, Ford I. A partial likelihood approach to smooth estimation of dynamic covariates effects using penalized splines. Biometrical Journal. 2007;49(3):441-52.
2.
Costa MJ, Shaw JEH. Parametrization and penalties in spline models with an application to survival analysis. Computational Statistics and Data Analysis. 2009;53(3):657-70.
3.
Hofner B, Hothorn T, Kneib T. Variable selection and model choice in structured survival models. Technical Report. 2008;43.
4.
http://www.stat.uni-muenchen.de External link
5.
Kauermann G. Penalized spline smoothing in multivariate survival models with varying coefficients. Computational Statistics and Data Analysis. 2005;49:169-84.
6.
Strasak AM, Umlauf N, Pfeiffer RM, Lang S. Comparing penalized splines and fractional polynominals for flexible modells of the effects of continous predictor variables. Computational Statistics and Data Analysis. 2011;55:1540-51.
7.
Wood SN. Generalized Additive Models: An Introduction with R. Chapman and Hall. CRC Press; 2006.