More option for cure model

Looks like the cure model can also be estimated using another R package GAMLSS. 

de Castro, M., V. G. Cancho, and J. Rodrigues. 2009. “A hands-on approach for fitting long-term survival models under the GAMLSS framework.” Computer Methods and Programs in Biomedicine.  

Cure model using R

The package "nltm" seems to be able to estimate proportional hazard and proportional odds cure models. I will do some experiments and see how it goes.

There is still no cure model facitility in R

I remember a while ago I posted a message to R mailing list asking about the cure model (long-term survivor model) using R, somebody pointed me to: http://post.queensu.ca/~pengp/software.html

But the author does not provide source code and there is no Linux binary. These are the same options when I was in graduate school in the early 2000s. By contrast, now Stata has "strsmix" and "strsnmix" to do both mixture version and non mixture version of the cure model, both are very well documented.

Cure models are tricky

I am trying to estimate cure model (long-term survivor model) using "strsmix" and "strsnmix" in Stata. It took me quite a while to realize that good starting values are crucial for quick convergence, something I only thought of when estimating aML or Mplus models.

Split-population model (cure model, long-term survivor model)

When there are a portion of respondents who will never experience the event (immortal), ordinary survival modeling techniques are not adequate. Special models designed to handle this kind of situations are called split-population model, cure model, or long-term survivor model.

aML does not handle split-population model; Mplus handles it by imposing constraints on a two-class mixture model; Stata has the following some facilities:

1. lncure: log-normal model with split-population;
2. spsurv: discrete time split-population model;
3. cureregr: split-population model with weibull, lognormal, logistic, gamma, and exponential distribution;
4. strxmix and strsnmix: split-population model with weibull, lognormal, gamma, and some mixture distribution.

Among the above, 1-3 are not well documented, while 4 is described in the most recent issue of Stata Journal (7-3).

For discrete-time models, there are only two alternatives: Mplus or spsurv.