Need help regarding univariate survival analysis

[Hasan Raza]

Dear all,

I have a thesis in which I have to do survival analysis. In which I have to find multiple prognostic factors.
Usually it is done by applying Univariate analysis in which many factors are analyzed individually and if they get significant value then they are assessed in multivariate analysis. For example we want to assess age, gender, tumor size, socio-economic status.
For this we have to analyze all of them in univariate analysis individually and suppose three of them get significant result,we will then analyze those three in multivarate analysis.

My question is how to use Univariate analysis in SPSS using 1 or 2

1) Analyze>survival>Kaplan-Meier
2) Analyze>Survival>Cox Regression and using only one covariate at a time for Univariate analysis

Thanking you

Regards

 

[Karabiner]

Usually it is done by applying Univariate analysis in which many factors are analyzed individually and if they get significant value then they are assessed in multivariate analysis.

This strategy leads to overfitting and to distorted p-values in the multple predictor model.
The final analysis doesn't "know" that you combed through the dataset to pick some statistically
significant predictors out of a bunch of candidate variables. It is therefore not considered as an
acceptable way of doing statistcial analyses, AFAIK.

With kind regards

Karabiner

 

[Hasan Raza]

Please see the article
The same has been done in this. There are other similar articles as well.

This strategy leads to overfitting and to distorted p-values in the multple predictor model.
The final analysis doesn't "know" that you combed through the dataset to pick some statistically
significant predictors out of a bunch of candidate variables. It is therefore not considered as an
acceptable way of doing statistcial analyses, AFAIK.

With kind regards

Karabiner

 

[GretaGarbo]

Usually it is done by applying Univariate analysis in which many factors are analyzed individually and if they get significant value then they are assessed in multivariate analysis.

I think that this is an incorrect method.

This strategy leads to overfitting and to distorted p-values in the multple predictor model.

Yes.

Or underfitting.

Please see the article
The same has been done in this. There are other similar articles as well.

Why should we read this? If someone else has made a mistake does not make it acceptable for you to do a mistake.