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
- /
- Proc Phreg problem

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

11-05-2014 10:45 AM

Dear All,

I am doing COX analysis, here is my syntax:

proc phreg data=a;

class HR;

model time*status(0)=sex HR age HTN / rl;

run;

Sex , HR, and HTN are binary variables, age is a continuous variable. the output of Analysis of Maximum Likelihood Estimates showed below:

Parameter DF Hazard 95% Hazard Ratio Confidence

Ratio Limits

HR low 1 16.143 5.8 79.678

age 1 0.753 0.643 0.882

sex 1 0.099 0.015 0.637

HTN 1 5100 0 .

Why HTN has this kind of 95% hazard ratio confidence limits: 0, . ?

is this related to small number of no HTN in HR low group? below is a 2X2 table:

NO HTN HTN

HRLOW 1 17

HRhigh 4 51

How can I solve this problem?

Thanks for your help.

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

11-05-2014 03:33 PM

I think your own answer is correct, namely that the samplesize is too low in some group. Though, since you did not include the interaction between HR and HTN I rather think it is the total "NO HTN" group that are too small.

Also, remember that you need to see some events in all groups, therefore it is relevant to count the number of events at each level for all variables. If there are some level with 0 events, then you can not estimate the hazard-ratio for that level.

Sometimes estimating becomes impossible even if there are events at all levels. Forexample if there are just one event at some level and that event happens at some time where there are no other at risk in any other levels - or if that event is the only event in some level for some other variable (a kind of coliniarity). Try therefore to make the 2X2 table you did, but weighted with number of events.

Jacob