Violations of the proportional hazards assumption

[Huitzilihuitl Saucedo]

Hello everyone

Perform a COX proportional hazard regression model. However, the final multivariate model violates the risk proportionality assumption by all methods (stphtest, tvc () texp (ln (_t))), residuals, etc.

Later I evaluate the model, concluding that it is nested, and I decided to use a Weibull model.



My question to you once you have used a parametric model and violated the proportional hazards assumption, it is okay to go ahead and use a Weibull type model, and what does it imply?

[Maarten Buis]

The Weibull model also assumes a proportional hazard assumption, so using a Weibull model cannot be a solution to a violation of the proportional hazard assumption in a Cox model. So my advise is to go back to a Cox model and try again with the tvc() and texp() options. Especially the texp() option allows for a great flexibility of possible interaction terms with time. So if you can't find an appropriate model with that, you have a pretty much hopeless problem (or you have misunderstood or misapplied the criteria for chosing your model)

[Huitzilihuitl Saucedo]

Thank you very much.