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
- /
- Repeated Measures Mixed Model ""Error: Stopped be...

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
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-04-2017 05:57 PM

Hello,

I am trying to measure the difference in change of eGFR (postindex_egfr) between treated (cohort_num=1) and untreated (cohort_num=0) groups. I am using a repeated measures mixed model to study this association and eGFR values have been divided by quarter (over 5 years). A lot of quarters will have missing eGFR values.

There are no duplicate values by quarter for any given patient. The code is written below and I'm getting an infinite likelihood error. Advice would be appreciated.

proc mixed data=hcvrenal.analysis_set_3 method=reml covtest;

class patid cohort_num (ref='0') gender bus_num dm(ref='0') htn(ref='0') ccf(ref='0') mi(ref='0') pvd(ref='0') cva(ref='0') can(ref='0') abuse(ref='0')

hiv hbv ftg(ref='0') ckd_stage(ref='1') quarter (ref='1');

model postindex_egfr = cohort_num ageatindex gender bus_num

dm htn ccf mi pvd cva can abuse hiv hbv ftg

fib4 ckd_stage

QUARTER indexdate /s ddfm=satterth COVB ;

WHERE QUARTER IN (1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20);

RANDOM patid/cl;

repeated quarter/type=ar(1) subject=patid*cohort_num ;

lsmeans cohort_num/cl;

estimate 'Change from baseline to quarter 4' quarter

-1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0/e cl;

estimate 'Change from baseline to final eGFR' quarter

-1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1/e cl;

run;

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

Posted in reply to sanketshah07

08-05-2017 03:27 PM

I'm not familiar with this procedure but you missed posting the ERROR you got.

Please post, too, the log and any results you got, marking the errors.

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

Posted in reply to Shmuel

08-05-2017 06:00 PM

Here is the log:

WARNING: Class levels for patid are not printed because of excessive size.

NOTE: 20364 observations are not included because of missing values.

WARNING: ODS graphics with more than 5000 points have been suppressed. Use the PLOTS(MAXPOINTS= )

option in the PROC MIXED statement to change or override the cutoff.

**WARNING: Stopped because of infinite likelihood.**

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

Posted in reply to sanketshah07

08-05-2017 06:55 PM - edited 08-05-2017 07:01 PM

I have googled for: **sas proc mixed infinite likelihood**

it seems to me that next link may help:

__Next lines were copied from that link page:__

An infinite likelihood during the iteration process means that the Newton-Raphson algorithm has stepped into a region where either the or matrix is nonpositive definite. This is usually no cause for concern as long as iterations continue. If PROC MIXED stops because of an infinite likelihood, recheck your model to make sure that no observations from the same subject are producing identical rows in or and that you have enough data to estimate the particular covariance structure you have selected. Any time that the final estimated likelihood is infinite, subsequent results should be interpreted with caution.

You may understand it better than me.

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

Posted in reply to sanketshah07

08-08-2017 03:56 PM

There could be some helpful tips here

http://support.sas.com/resources/papers/proceedings12/332-2012.pdf

and

http://support.sas.com/resources/papers/proceedings15/SAS1919-2015.pdf

I have found that rescaling the response can be miraculous (sometimes anyway).

You may have collinearity problems with the predictor variables (both continuous and categorical).

You could start with a basic model that includes only a minimal number of predictors (i.e., cohort_num and quarter). If that converges, then you have something to work with. If it doesn't then that may point you towards a potential problem with patid.