CODEWITHSUNDEEP
rsurvival-analysis
Frederick
Frederick

Reputation: 850

Estimating survival with a generalized gamma function in flexsurv fails

I have survival data for patients. Some of them are censored. I want to fit a generalized gamma function to the empirical data with the flexsruv package in R. While all other functions that I want to fit (such as exponential, weibull etc.) are working just fine, I get the following error code for dist = "gengamma":

Error in optim(method = "BFGS", par = c(5.02274354115438, -0.0670900421918298,  : 
  non-finite finite-difference value [2]

Here is the data that I have:

db.survival <- data.frame(time =  c(101, 111, 185, 707, 85, 58, 427, 672, 90,
                                1452, 608, 99, 556, 62, 60, 1445, 563, 246,
                                163, 276, 216, 64, 61, 66, 67, 68, 81, 83,
                                99, 100, 100, 100, 100, 100, 100, 100, 100),
                      status = c(1, 0, 1, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1,
                                 1, 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1,
                                 0, 0, 0, 0, 0, 0, 0, 0, 0))

I receive an error code for:

library(flexsurv)
flexsurvreg(Surv(time, status) ~ 1, data = db.survival, dist="gengamma")

However, when I choose the methods "CG" or "SANN", I get results (but they differ).

My questions is: why do I get an error in the first place? Do I code wrongly?

Thanks a lot in advance for any kind of help!

Upvotes: 1

Views: 1235

Answers (1)

Benjamin Christoffersen
Benjamin Christoffersen

Reputation: 4841

I do not think there is something wrong with your code. The error comes from the optim that flexsurvreg uses. By default, optim uses a finite-difference approximation for the gradient as stated in help("optim")

gr A function to return the gradient for the "BFGS", "CG" and "L-BFGS-B" methods. If it is NULL, a finite-difference approximation will be used. For the "SANN" method it specifies a function to generate a new candidate point. If it is NULL a default Gaussian Markov kernel is used.

You can see where the code fails with the following calls

> options(error = recover)
> flexsurvreg(Surv(time, status) ~ 1, data = db.survival, dist="gengamma", control = list(trace = 100))
initial  value 144.298116 
iter  10 value 131.067771
iter  20 value 128.013812
iter  30 value 127.761461
iter  40 value 127.753269
Error in optim(control = list(trace = 100), method = "BFGS", par = c(5.02274354115438,  : 
  non-finite finite-difference value [2]

Enter a frame number, or 0 to exit   

1: flexsurvreg(Surv(time, status) ~ 1, data = db.survival, dist = "gengamma", control = list(trace = 100))
2: do.call("optim", optim.args)
3: optim(control = list(trace = 100), method = "BFGS", par = c(5.02274354115438, -0.0670900421918298, 0), fn = function (optpars, .

Selection: 3

After which you see that the optim does run for few iterations but than fails in some C code. Further, you can see that the flexsurv does not supply the lower or upper arguments to optim which could be used with the method "L-BFGS-B". You can see in this vignette that one of the parameters have restricted support so that may help you.

Upvotes: 1

Related Questions