Help with flexible parametric cure model

[Paul Sanfilippo]

I think I have a reasonable understanding of flexible parametric survival models and how to use them (not to mention that I like using them because I think they make survival analyses much easier on many levels).

We came up with a model that we thought was ok, but it was then pointed out that with about 50% survival (at the end of years of observation) in the two primary exposure groups, we should really consider a cure model.

These are data where the primary exposure is a time-varying covariate - initially unexposed and then once exposed, always exposed. The data are in counting process form, so hopefully that's not an issue for estimating a cure model.

I didn't know anything about cure models, but have done some reading and tried to use such a model but I'm falling flat. I wonder if anyone can help?

stpm3 is great but currently doesn't seem to support cure models, so I went back to stpm2 which does.

The stpm2 equivalent that runs perfectly as a best-fitting model in stpm3 is:

stpm2 live_birth age_dur disease_years age_start base_EDSS High_eff_DMT ARR sym_period_dum2 sym_period_dum3 sym_period_dum4 country_dum2 country_dum3, scale(hazard) tvc(live_birth age_start base_EDSS ARR) df(1) dftvc(1) eform cure

however, I get:

DF must be between 3 and 11

which is the first thing I don't understand - why can we not have a simpler cumulative baseline function? If I then attempt the model with df = 3, i.e.

stpm2 live_birth age_dur disease_years age_start base_EDSS High_eff_DMT ARR sym_period_dum2 sym_period_dum3 sym_period_dum4 country_dum2 country_dum3, scale(hazard) tvc(live_birth age_start base_EDSS ARR) df(3) dftvc(1) eform cure

I get another error:

varlist required
               st_data():  3598  Stata returned error
                orthgs():     -  function returned error
                 <istmt>:     -  function returned error

Removing the time-dependent effects on the 4 variables fixes that, but then gives me another error:

stpm2 live_birth age_dur disease_years age_start base_EDSS High_eff_DMT ARR sym_period_dum2 sym_period_dum3 sym_period_dum4 country_dum2 country_dum3, scale(hazard) df(3) eform cure

initial values not feasible

Can anyone tell me what could be going wrong?

Thanks.