search

giabaio / survHE

Survival analysis in health economic evaluation Contains a suite of functions to systematise the workflow involving survival analysis in health economic evaluation. survHE can fit a large range of survival models using both a frequentist approach (by calling the R package flexsurv) and a Bayesian perspective.
https://gianluca.statistica.it/software/survhe/
41 stars 19 forks source link

plot model #21

Closed Geoff-Holmes closed 6 years ago

Geoff-Holmes commented 6 years ago

Hi Gianluca

When I run: m1<-fit.models(formula, data=regdata, distr="weibull") plot(m1) everything looks fine. However, m2<-fit.models(formula, data=regdata, distr="weibull", method="hmc") plot(m2) The model survival curves don't coincide at all with the KM as below. Is this a glitch or do I need to do something different?

plot1

plot2

giabaio commented 6 years ago

Hi, This looks weird and I've not experienced problems like this with the Weibull distribution. (In a previous version of survHE there was an error in the code for the Gompertz distribution, which would make the calculation of the survival curves all wrong --- but this has been long fixed). In all my tests with both parameterisations of the Weibull, things have worked out just fine...

Can you share the data so I can reproduce the error and debug?

Geoff-Holmes commented 6 years ago

Unfortunately, I can't share the data that I used for this. However, as soon as I can I will see if I can adapt it or reproduce the problem in some way that I can share.

Geoff-Holmes commented 6 years ago

Thanks for the email correspondence on this Gianluca.

So, if I understand correctly, this is down to the way survHE parses the formula passed to it.

I noticed previously that whilst in flexsurv, I can do: survObj<-with(dat, Surv(Survival, event)) formula<-survObj~group m<-flexsurvreg(formula, data=dat, dist="weibull")

In survHE, this doesn't work (Error in attributes(terms(formula))$variables[[2]][1] : object of type 'symbol' is not subsettable) and the call to Surv has to be included explicitly in the passed formula. This is not a problem, but I think coming from familiarity with flexsurv, it tripped me up somewhat.

In the current issue, the formula I specified was: formula<-with(dat, Surv(Survival, Status==1)~Group)

This formula works okay in flexsurv, but for survHE, the event needs to first defined as a binary: dat$Event<-dat$Status==1 And then, formula<-with(dat, Surv(Survival, Event)~Group) m2<-fit.models(formula, data=regdata, distr="weibull", method="hmc") plot(m2) all works fine and agrees with m1 in the original post.

giabaio commented 6 years ago

Yes. In fact, you can check this by comparing the two version of the model (MLE or HMC) when using a binary indicator Event (taking value 1 if fully observed and 0 if right censored --- irrespective of how the right censoring comes about) with the version of the MLE (flexsurv-based) model when you simply subset the original Status indicator to the value 1 (ie fully observed) or its survHE rendition. image

Now all the models are basically in alignment (as they should!).

I think, more interestingly, the point is that flexsurv can handle the Status indicator when you subset it to be =1 while still considering the fully observed as well as the censored cases. In that respect, flexsurv is more general (and probably better) than survHE. But I think the specificity of survHE should be maintained --- even if at the expense of asking the user to format the data in the "correct" way...