Notes, suggestions, most common errors
- Important: time must always be in days (follow-up time, age to be used in ratetables).
- Before fitting any models, always first plot the relative survival curve.
This should give you the idea whether your data are properly organised. If the curve looks weird (above 1, etc),
your ratetable object is most probably wrongly organised.
- If something seems to be wrong with results, try using the enclosed population tables (slopop) on your data. The difference should not be huge, if
it is, your problems might lie in the ratetable argument organisation.
- Note that the age and year you have in the data must be evaluated at time 0 - age and year at the time of diagnosis
Questions and answers
Q: I am trying to fit the additive model. While it converges with the EM algorithm, it seems to find colinearities with
the max.lik method?
A: The most probable reason is that there are too few events (with respect also to the number of covariates) in some
subintervals. Try enlarging the intervals in which the baseline excess hazard is constant. Say you have 10 years of
follow-up. First try fitting with
int=c(0,10), i.e. a constant baseline excess hazard. If it converges, try breaking into
more subintervals, e.g.