turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- Analytics
- /
- Stat Procs
- /
- Cumulative incidence estimates at time points usin...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago - last edited 2 weeks ago

Hi there,

I'm trying to fit a Fine and Gray competing risks model and then estimate the cumulative incidence at specific time points (eg. 2 years follow-up). Currently, I'm using the proc phreg function with the baseline options, but this will output all the estimate cumulative incidence at each failure time. I've tried using the timelist option but that can only works in bayesian Cox models.

My dataset is relatively big so even when running this model with one covariate, it is taking hours and hours. Ideally, I'd like to had about 8 variables in the model so currently this doesn't seem like a viable option for estimating cumulative incidence. However, producing the graph of the cumulative incidence doesn't take that long, surely there is some way to get the estimate around a certain time point rather than at all failure times.

Does anyone know whether you can specifiy time points to estimate cumulative incidence outputs in the proc phreg command?

ods graphics on;

proc phreg data=test plots(cl overlay stratum range=0,10)=cif;

class male2 (order=internal ref=first);

model time_yr*outcome(0)=male2 / eventcode=1;

baseline covariatesrisk out=risk2 cif=_all_;

run;

ods graphics off;

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

2 weeks ago

I dont think you can speed up the procedure even if you find a way only to get the estimate at a specific timepoint. It is most likely the estimation of the parameter that takes time. But you can try taking graphics off, and then just pick out the observation at 2 years as in a datastep like this:

**data** _NULL_;

set risk2(where=(exit<=**2*****365**)) end=end;

if end;

put _all_;

**run**;

Or, if you know you have a failuretime around 2 years you can make the restriction on the riskcurve already when phreg create it:

baseline out=risk2(where=(**1.5**<=t<=**2**)) cif=_all_;

An alternative solution is to use the Aalen Johansen estimater, which can be computed with PROC LIFETEST, using the failcode option. This is a non-parametric method that is based on a Markov chain theory, and it runs therefore in computation time linear to the size of the dataset.

I

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

a week ago

Thanks for your suggestions Jacob. I thought as much that it wouldn't be able to provide cumulative incidence estimates at certain time points.