Cox.zph after survSplit aregImpute and fit.mult.impute

Hello guros,

I’m doing follow-up analysis of operated abdominal aneurysms. Time to death is the main outcome. Each patient is given a variable (what proportion of EVAR technique is used at the hospital) and the main question is if more use of EVAR affect mortality. There are 29 hospitals and there are alsom some 9 potential confounders and 8 other variables that might help in imputation. There are 4868 patients.

First I use survSplit to establish a counting process notation and divide patient observation in up to three lines (timeperiod 0 -1/2year, >1/2 -3 years, > 3 years - max 10 years).

Then imputation is done with aregImpute

Finally fit mult.impute with coxph is used nested in strata(timeperiod).

All this gives a reasonable result.

Now I want to do cox.zph on the result from fit mult.impute to evaluate the ‘prortional hazards assumption’ within each timeperiod

Unfortunately I get
Error in order(istrat, y[, 1]) : argument lengths differ.

A few years ago I managed to do this with success and reasonable results
in a simular analysis

Unfortunatle a cup of tea in the keyboard of that computers killed it so I can not study how I made the analysis.

Has the used program changed or is it impossible to test for ‘proportiomal hazards’ in this situation?

Any help highly appreciated

Yours sincerely Fredrik L

I would do goodness of fit assessment either in complete cases or after stacking all the filled-in datasets.