Is there a way to select covariates in survival analysis?
Created:
| Last Modified:
Survival analysis (Cox proportional hazards model) does not have a stepwise function for automatic covariate selection ( *1 ).
Therefore, you must manually select the covariates.
Several methods and selection criteria for covariates have been proposed, and there are no clear standards for their use. However, we introduce the following simple method as a reference example.
How to reduce covariates
- Once you have decided on the covariates (effects) to include in the Cox proportional hazards model, launch the Fit Model or Fit Proportional Hazards platform.
- Enter all columns that you have previously determined to be covariates into the Model Effects Configuration box, set other options such as Censoring as necessary, and click Run.
- In the report that appears, check the p-value for each factor (effect) in the Effect Summary, select the factor with the largest p-value and click Remove.
- Repeat step 3 until the p-values of all factors meet the predetermined criterion ( *2 ).
*1 " What kind of analyses can be performed using the stepwise method in JMP? "
JMP can perform stepwise methods for least squares methods (regression and ANOVA) and logistic regression. In JMP Pro, you can also use Generalized Regression under Fit Model to perform model selection, including forward selection.
*2 Effect Summary is a feature from JMP 12. In JMP 11, you need to carry out step 2 above to reduce the effects each time based on the p-value of the Effect Likelihood Ratio Test.
FAQ #3801
This post originally written in Japanese and has been translated for your convenience. When you reply, it will also be translated back to Japanese.