Solved – Comparison of Cox models

cox-modelkaplan-meierspss

I am new to statistics and SPSS v 20. I am doing survival analysis – KM and Cox regression with SPSS. I am trying to make Cox models with different variables – clinical, genetic etc. I am interested to know how well each of these models are able to predict the occurrence of metastases. My questions are:

Is it possible to do this with SPSS?
Is there a way to save the model predicted outcome while doing Cox regression? I saw that when I do multinomial logistic regression in SPSS, I can save the predicted category and compare it with the actual category. I don't have that option in Cox.
Is it possible to make ROC curves for different Cox models and then compare the area under the curve for different models?

I am not sure I am going about this the right way.

Best Answer

Yes it is possible! please see this: http://www.youtube.com/watch?v=wTLsw-Ckfvw
Yes from the menu or from the syntax you can save these. Depending on you programming skills R might be more versatile for this...(maybe I'm biased)
Yes, please look in survival analysis from SPSS documentation.

Related Solutions

Solved – Multivariate analysis for factors affecting survival

You should be able simply to specify one of the levels of the histology factor as the reference. For example, choose as reference the level that is expected to have the best outcome based on clinical expectations. I don't use SPSS but I expect, like in the R survival package that I use, that won't pose a problem, with hazard ratios for the other levels of the histology factor then expressed with respect to your reference level.

Before you do this, though, think hard about what you're trying to accomplish with the multivariate analysis. In my experience with similar analyses, many clinical variables having significant univariate relations to outcome are highly correlated with each other (like tumor size or spread to lymph nodes and the patient's disease stage, which is based on tumor size and spread to nodes). Even your histology subtypes may be correlated with other clinical variables. The particular variables that you find "significantly" related to outcome in the multivariate analysis then may be highly idiosyncratic to your particular data set, and not very useful for generalization to other data sets. And you have to resist the temptation to include too many variables in the multivariate analyis, lest you overfit the data. Some early consultation with a statistician experienced in survival analysis may save you a lot of trouble down the road if you want to publish your results.

Solved – Cox PH model selection and validation

Your first method of bivariate screening and then backwards elimination is not recommended. These methods have numerous problems and have been extensively discussed here and elsewhere (search on "Stepwise" to find some).

Taking your questions one at a time:

Is it statistically correct to use an a priori defined model and then add my statistically significant variables, comparing the LLR of
the different models (using the hierarchical Cox analysis)?

This is not a bad method. It is certainly an improvement.

If this is correct, how do I initially select a model to which I then add another variable?

Based on theory, substantive knowledge, research questions etc.

How do I solve the problem of overfitting? Do I have to use bootstrapping for my final model?

If you have a lot of cases, then overfitting shouldn't be too much of an issue, but you can do things like use a training and test set if you are concerned.

Your latter questions seem to be more SPSS specific; I don't use SPSS.

Best Answer

Related Solutions

Solved – Multivariate analysis for factors affecting survival

Solved – Cox PH model selection and validation

Related Question