Cox proportional hazard model for survival analysis on multilevel data

#1
I am trying to run a survival analysis on a multilevel data (two levels) with five units nested within each person (each case has data from five different years). I have a number of covariates to enter into the model as I am interested in their interaction.
I am considering using a Cox proportional hazard model but I found this is not a good fit for multilevel data unless random effect is incorporated.
My questions are: first, is Cox regression model with mixed effects the best model for analyzing my data given its nature. Secondly, does anyone know how to incorporate random effect into a conventional Cox proportional hazards model on SPSS? I have found a description on how to do this from an article by Austin, P. C. (2017) (A tutorial on multilevel survival analysis: methods, models, and applications. International Statistical Review) but he has demonstrated this using STATA and R.
 

hlsmith

Omega Contributor
#2
Can you provide an example of what your data look like? So you have repeated measures clustered in subjects. Are these repeat values for covariates, outcome or both? Why is this survival data in your mind, I didn't follow that part in your description.


Lastly, the purpose of running multilevel analyses is to control for random effects, random intercepts, or both. So do you want to do this.


P.S., Austin usually has quality publications. Are you able to try a new program (e.g., STATA or R)?
 
#3
Can you provide an example of what your data look like? So you have repeated measures clustered in subjects. Are these repeat values for covariates, outcome or both? Why is this survival data in your mind, I didn't follow that part in your description.


Lastly, the purpose of running multilevel analyses is to control for random effects, random intercepts, or both. So do you want to do this.


P.S., Austin usually has quality publications. Are you able to try a new program (e.g., STATA or R)?

Thanks for your reply hlsmith,
i have repeated measures for both outcome and covariates clustered in subjects. my outcome is the age at which the subjects initiate a lifestyle behaviour as such I am trying to analyse the sruvival time as well as the effect of the covariates (including their interaction with one another) on survival.
Also,I am interested in controlling for control for both random effects and random intercepts.
 

hlsmith

Omega Contributor
#4
Given the repeated IVs and DVs, I wonder if a marginal structural model may work. I have ran one before but they address time dependent confounding and across time measures.