Hi,

I am wondering whether anyone can help with a problem I am having synthesizing data for a meta-analysis.

The primary outcome is time to patient recovery (status=1) in the presence of death as a competing risk (status=2) and all trials compare treatment (group=1) to control (group=0). I have the patient-level data for 5 trials (study) & am aiming to plot the survival curve for the pooled analysis (cumulative incidence function).

In a two-step meta-analysis I assess both the cause-specific hazard & the subdistribution hazard for recovery using the following for each trial individually:

Code:
stset days, failure(status==1)
stcox group
stcrreg group, compete(status==2)
& pool the study-specific estimates using standard meta-analysis techniques.

In a one-step meta-analysis, I need to account for the clustering of trials. As far as I have got with this would be:

Code:
stcrreg group i.study, compete(status==2) vce(cluster study)
stcurve, cif at1(group=1) at2(group=0)
However, I am not sure this adjustment is sufficient - do you agree? I have read about different methods to estimate separate baseline hazards for different studies here (stratification, frailty models, marginal models). Is there a way to implement these in Stata in the context of competing risks? & plot the resultant cumulative incidence function?

I have looked at stmixed (ssc install stmixed), but this does not account for competing risks as far as I can tell.

Thank you for reading,
Megan