Previously, I started discussing the flexsurv package. I used it to fit a Weibull model. This is implemented as an accelerated failure time model. It is also a proportional hazards model (although, as I found previously, converting between the two is not so straightforward, but it can be done by SurvRegCensCov).
Now let’s compare Weibull regression with Cox regression. Firstly, Weibull regression:
assumes proportional hazards; the number of parameters is equal to \(k + 2\), where \(k\) is the number of covariates; we can estimate things like the median, \(P(S>s^*)\), etc.
I’m going to write about some of my favourite R packages. I’ll start with flexsurv (https://github.com/chjackson/flexsurv-dev) by Chris Jackson, which can be used to fit all kinds of parametric models to survival data. It can really do a lot, but I’ll pick out just 2 cool things I like about it:
Fit a standard survival model, but where it’s slightly easier to work out what the parameters mean. Fit a proportional hazards model, which is a lot like a Cox model, but where you also model the baseline hazard using a spline.