Question: Why use LASSO / elastic net in survival regression?
1
gravatar for CY
15 days ago by
CY270
United States
CY270 wrote:

We usually use cox model for survival analysis. However, some papers (PMID: 29628290) used LASSO / elastic net regularized survival regression. What is the reason using that? Is it because researchers think some of variables are not indenpendent (highly correlated)?

survival analysis lasso • 106 views
ADD COMMENTlink modified 15 days ago by Jean-Karim Heriche17k • written 15 days ago by CY270

You will probably get a better answer to your stats question on Cross Validated

ADD REPLYlink written 15 days ago by Philipp Bayer5.8k
0
gravatar for Jean-Karim Heriche
15 days ago by
EMBL Heidelberg, Germany
Jean-Karim Heriche17k wrote:

The LASSO and elastic net are regularization methods that can and have been applied to Cox models. They are useful because they lead to sparser models, in particular when dealing with high-dimensional data.

ADD COMMENTlink written 15 days ago by Jean-Karim Heriche17k

Yes. LASSO can get rid of some highly correlated variables and only keep one (sparser solution). But why do they make this judgement of using LASSO instead of regular Cox? Perhaps they think that their variables are redundant (highly correlated)?

ADD REPLYlink written 15 days ago by CY270

What about doing a PCA first, then cox or other regressions on the components? The PCA would deal with correlation among the variables. The lasso regression would penalize correlated variables and possibly remove them from the model (as they will have lower betas), but is that what you really want?

ADD REPLYlink modified 15 days ago • written 15 days ago by Giovanni M Dall'Olio26k
1

The lasso regression would penalize correlated variables

Not necessarily, it depends also on many other factors such as the effect size, how many variables, how much data you have...

ADD REPLYlink modified 15 days ago • written 15 days ago by Jean-Karim Heriche17k

LASSO regularization forces some coefficients to 0 whereas in a regular regression, coefficients can be low but rarely 0.

ADD REPLYlink written 15 days ago by Jean-Karim Heriche17k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 837 users visited in the last hour