Dear all,
I have a question concerning validation of a prognostic survival model. We
determined a prognostic score for metastatic breast cancer patients (outcome
: time to death). The prognostic model has been estimated from a training
sample of 343 patients. The prognostic index, which was a weighted sum of
several covariates (10 parameters), was derived from Cox regression, CARTS,
medical experts know-how.
After the calculation of the prognostic index, a cut off was chosen to
divide the groups in three prognostic strata. Then we did an evaluation of
the prognostic performance of this 3 group classification rule, calculating
different separation indices (SEP, PSEP) and also we calculated 1 and 2 year
survival misclassification rates (Brier Score, Rē measures, etc.).
Knowing that these measures are too optimistic, because we used the training
sample as evaluation sample, we want to adjust them - maybe using a
resampling approach. Here are my problems, since I have no experience in
resampling methods. In fact we want to validate the 3 group classification
rule, not the prognostic index, is this possible by applying a resampling
approach?
If yes,
- Would it make sense to do a cross-validation, how to divide the
sample, 50:50 or 70:30, are there any recommendations about this in the
literature.
- Would it make sense to do this maybe more than once (repeated
cross-validation) and then take the mean of this.
- What about bootstraping, would it make sense to apply the
classification rule on 100, 500 or 1000 bootstrap samples derived from the
training sample. What information would we gain doing this?
Thanks
Bernd
----------------------------------------------------------------------------
---------------------------
Dr Bernd Genser
MSc, PhD
Statistician
Federal University of Bahia
Salvador, Brazil
www.isc.ufba.br
----------------------------------------------------------------------------
----------------------------
---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.654 / Virus Database: 419 - Release Date: 06-04-2004
