cox proportional hazard model

Question

am doing my dissertation involving cox model and i would like to understand how you interpret the survival table at mean of covariates. how i do u i determine the survival function from the output

Survival Table  

At mean of covariates
Time    Baseline SurvivalSE Cum Hazard
0   .023    .991    .006    .009
3   .033    .987    .007    .013
5   .044    .982    .009    .018
8   .058    .977    .011    .024
9   .088    .965    .014    .036
11  .107    .957    .017    .044
12  .128    .949    .019    .052
14  .173    .932    .024    .071
24  .232    .910    .033    .095
34  .326    .875    .048    .133
45  .481    .821    .068    .197
46  .668    .761    .088    .273
49  .946    .679    .109    .387
56  1.769   .485    .178    .723

Correlation Matrix of Regression Coefficients
          Sex      Birthweight  MartenaAgeBreatsfeeding
Birthweight -.151           
MartenalAge -.108   -.077       
Breatsfeeding   .030    .222    .057    
Immunisation    .240    -.085   .016    -.168

Covariate Means
    Mean
Sex        .552
Birthweight .129
MartenalAge 3.067
Breatsfeeding   .238
Immunisation    .124

my output from my cox model and it contains the above covariates and immunisation and breatsfeeding are siginificant but i want to intepret the survival and hazard functions for young children from 0-60 months

Answer 1

The survival table at the mean of the covariates shows the estimated survival as a function of time (i.e. probability of no event by a given time point) for a hypothetical subject whose covariates equal to covariate means. In this case, it is a subject with Sex=0.552, Birthweight=0.129, MaternalAge=3.067 (those must be some strange units), etc. It is quite obvious that for binary predictors this does not make much sense. Sex can be (probably) 0 or 1, but not 0.552. So, really, this table is quite meaningless.

There are several ways around this problem, depending what are you trying to achieve. One option is to predict survival for each group, such as non-immunized breastfed boys with mean birth weight and maternal age.

A word of caution unrelated to your question: breastfeeding and immunization are probably time-dependent covariates, so you have to be careful incorporating them into the model.