Dear statalisters,
I am wondering how to present the results of nested models in survival analyses (using piecewise exponential models with streg). If I have understood correctly, comparison of HRs of an explanatory covariate between nested models is not meaningful as HRs of this covariate can change even if it is not related to the new covariate (cf Mood 2010). As far as I understand, in logistic regression analyses or in discrete-time survival analyses margins-command and produced predicted probabilities would solve this issue. However, I am somewhat confused of the margins-command after streg. Using margins C1(covariate) –command after streg produces (as default) predicted median time to event in categories of the covariate, which in my case is not as useful as would be, for example, adjusted relative risks, or risk ratios. What I don’t know is how to produce these, or solve the problem otherwise.
I am currently doing the analyses in the following way:
(first stset, stsplit; follow-up time in 3-months ‘pieces’)
streg ibn.ftime3 i.COV1 i.COV2 i.COV3, dist(exp) nolog nocons
On the basis of HRs from the nested models, the effect of COV2 on dependent variable (first births) diminishes markedly after introducing covariate COV3 in the model.
Apologies if my question is very trivial, but any help or reference to previous Q-A would be highly appreciated!
I am wondering how to present the results of nested models in survival analyses (using piecewise exponential models with streg). If I have understood correctly, comparison of HRs of an explanatory covariate between nested models is not meaningful as HRs of this covariate can change even if it is not related to the new covariate (cf Mood 2010). As far as I understand, in logistic regression analyses or in discrete-time survival analyses margins-command and produced predicted probabilities would solve this issue. However, I am somewhat confused of the margins-command after streg. Using margins C1(covariate) –command after streg produces (as default) predicted median time to event in categories of the covariate, which in my case is not as useful as would be, for example, adjusted relative risks, or risk ratios. What I don’t know is how to produce these, or solve the problem otherwise.
I am currently doing the analyses in the following way:
(first stset, stsplit; follow-up time in 3-months ‘pieces’)
streg ibn.ftime3 i.COV1 i.COV2 i.COV3, dist(exp) nolog nocons
On the basis of HRs from the nested models, the effect of COV2 on dependent variable (first births) diminishes markedly after introducing covariate COV3 in the model.
Apologies if my question is very trivial, but any help or reference to previous Q-A would be highly appreciated!
