R help - Feb 2010 - Censored outcomes - repeated measures and mediators

Hello,

In a study exploring transgenerational transmission of anxiety  
disorder we investigate whether infants react to experimentally  
induced mood changes of their mothers. We measured the time that an  
infant needed to cross a cliff (=crossing time) depending on whether  
his mother had previously undergone a mood induction (treatment) or  
not (control). The treatment is thus a within-subjects factor with two  
levels. The outcome, crossing time, is censored as some infants did  
not cross the cliff within the predefined time span of 180 seconds.

My first question is which kind of proportional hazard model would be  
suitable here. Based on what I’ve found in the literature 1) a  
marginal model or 2) a frailty (mixed-effects) model would be suitable  
here.

If ct=crossing time, event=crossed/not crossed, tr=treatment,  
ptid=person ID, then the two models could be implemented like this:

1) coxp(surv(ct,event)~treat+cluster(code)

2) coxme(surv(ct,event)~treat+(1|code)),  assuming a random intercept  
for each mother/infant

Am I on the right track here?

A second question relates to mediator analyses. We have variables  
regarding mothers’ negative affectivity which, according to our model,  
are caused by mood induction and at the same time affect infants’  
crossing time. I don’t know of any mediator model that directly tests  
censored outcomes in a repeated measures setting given one (or  
several) mediator variable(s).

Any help to either question is greatly appreciated.

Andrea







	[[alternative HTML version deleted]]