Test Proportional Hazards Assumption in Recurrent Event Model

[kc]

Hello,

The assess statement in proc phreg can be used to test proportional hazards assumption in a cox regression model like in model A. But, how do you assess proportionality of the explanatory variables (both categorical) in model B which is a recurrent event model?

Any reference with sample code/example is appreciated.

/*Model A*/
proc phreg data=test;
class trt(ref="A") strata(ref="B");
model time*status(0 2) = trt strata / risklimits ties=exact type3(score);
hazardratio trt / cl=pl diff=ref;
assess ph resample;
run;

/*Model B - Recurrent Event Model*/
proc phreg data=test2 covs(aggregate);
class trt(ref="A") strata(ref="B");
model (tstart, tstop)*status(0 2) = trt strata /  risklimits ties=exact type3(score);
hazardratio trt / cl=pl diff=ref;
baseline covariates=covds out=out cmf=_all_ / nomean;
id subjid;
run;

[Mike_N]

To begin, have a look at this page: SAS Help Center: Analysis of Multivariate Failure Time Data . Which model do you intend to fit?

[kc]

The code I have provided in model B above corresponds to fitting a proportional rates/mean model using Lin et al. 2000, abbreviated as LWYY model.

[Mike_N · Accepted solution]

I don't think there is a built-in way to get model diagnostics for the proportional means model. As you noted, the ASSESS statement is used to examine proportional hazards, not proportional means/rates. The LWYY paper (Semiparametric Regression for the Mean and Rate Functions of Recurrent Events | Journal of the Royal...) does provide a section that specifies how to carry out model checks based on residuals, but those methods are not implemented in SAS (to my knowledge).

The process is similar to that implemented in the ASSESS statement for proportional hazards models. You would need to use the OUTPUT statement from PROC PHREG to store the martingale residuals and then use those in the model-checking methods described in LWYY.  

[kc]

OK - thank you!