Kaplan meier graph with survival curves crossing over and standard Cox model

[eric some]

Dear Stata Forumists,
I described mothers'breastfeeding behavior using Kaplan Meier survival curves with country stratification. I realised that the curves are crossing over. I wanted to use a Cox standard proportional hazard ratio to identify the determinants of the mothers'practice. Can I still use the Cox analysis with my Kaplan Meier curves crossing over? Otherwise,what is the alternative?
Thanks for your help!
Regards

[Carlo Lazzaro]

Eric:
you may want to take a look at -stcox PHassumption tests- entry in Stata .pdf manual.

[Marcos Almeida]

Carlo gave very good advice. I also wish to add that the log-rank test is not at all a conservative test: quite on the contrary, even when you get a significant p-value, you may found lack of significance after adjusting for covariates by applying a Cox model. On second thoughts, shall the Kaplan-Meier curves cross only once, for example, in the end, you may select a weighted test instead of log-rank and see what happens. That said, perhaps you'll need to face it: there are high chances the null hypothesis cannot be refuted.

Best,

Marcos

[Steve Samuels]

You can incorporate groups with crossing curves in an stcox analysis, by specifying them as strata(). You can also add interactions of the stratum variables with others.T There are other options for directly modeling the crossing curves, including adding an interactions with time, using the tvc() and/or texp() options. You can get flexible parametric models for the curves with Patrick Royston's stpm and Paul Lambert's stpm2, both available via SSC. Please when you post, show commands and results with CODE delimiters, as the FAQ (Sec 12) request.