The best suggested method to internally validate a Cox regression model bases on bootstrap resampling technique with subsequent assessment of discrimination and calibration performance of generated models.
Could you please recommend a statistical software that easily generate bootstrap resamples, applies generated models to the test sample, calculate suitable statistics and draw suitable graphs?
Thank you!
Hi Michal.
SPSS already has a "bootstrap" module for the all the techniques incorporated in the software, and the survival analysis is one of those. It's just that simple like click on the bootstrap button and provide the number of samples that you need. Nevertheless, assumption checking is a little bit difficult with SPSS (they only provide the Schoenfeld residuals, but you can generate the martingale residuals by "hand"). But I recommend you the Landau & Everitt book "A Handbook of Statistical Analysis using SPSS". They provide a very simple explanation of the survival analysis in SPSS including the assumption checking.
Finally, if you can't use SPSS for any reason, maybe you can do a very simple technique called "Split half sample cross-validation". It just takes two random samples of the 50% of your sample, and check the consistencies and inconsistencies in the coefficients estimation.
I hope than my explanation was useful.
Using R can be an alternative solution. It has a package called bootstrap. The survival package also works with Cox regression and bootstraping
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software