Kaplan-Meier(K-M) vs. Cox Regression I am very new to survival analysis. I am looking for differences between these two methods - Kaplan-Meier(K-M) vs. Cox Regression.


*

*KM Survival Analysis cannot use multiple predictors, whereas Cox Regression can.

*KM Survival Analysis can run only on a single binary predictor, whereas Cox Regression can use both continuous and binary predictors.

*KM is a non-parametric procedure, whereas Cox Regression is a semi-parametric procedure.


Please check the above points. If incomplete or incorrect, please suggest the changes.
 A: I generally try to use KM as a descriptive statistic and Cox regressions for anything related to my hypothesis. Regarding your questions:


*

*You could in theory split your KM analysis into any number of subgroups, the only limit is the size of your data. It is a rather inefficient method but you can show KM Survival curves for women with diabetes, women wihtout diabetes, men with diabetes and men without diabetes.

*Correct. You can though categorize your continuous variables and thus have a categorical variable that you can use for splitting. Note that binary is not necessary, you need a categorical variable.

*Correct. I was in the beginning scared of parametric procedures as they rely on assumptions that I found hard to test and I liked the clean non-parametric tests. Unfortunately it turns out that the non-parametric tests, while being mathematically robust, are often hard to interpret. E.g. a low p-value often only indicates that the groups are not the same but it conveys little about in what way they aren't the same. 

