topic Re: Multiple imputation using chained equations in Statistical Procedures

Multiple imputation using chained equations

Alib — Tue, 05 Sep 2017 16:22:18 GMT

Hello,

I'm a beginner on SAS. I need to use multiple imputation using chained equations (FCS) to deal with missing values.

I want to use fitted linear regression models for a continuous variable X (fourth root) and fitted multinomial regression for nominal variable Y (4 categories) to create 10 imputed datasets; analyses are run separately and their results would be combined by use of Rubin's method.

Can someone help me for syntax?

Thanks

Re: Multiple imputation using chained equations

SAS_Rob — Tue, 05 Sep 2017 18:44:39 GMT

The approach from a syntax standpoint would be similar to this example

http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_mi_examples07.htm

The only difference in your case would be that rather than using the DISCRIM method, you would use the LOGISTIC method.

fcs logistic (species/link=glogit);

Re: Multiple imputation using chained equations

Alib — Wed, 06 Sep 2017 16:23:42 GMT

Thanks,

I use this syntax

proc mi data=table1 seed=1305417 nimpute=10 out=imputedtable1;
   class slevel sex period city death;
   fcs logistic(slevel/link=glogit) logistic(period/link=glogit) reg(rate/details);
   var slevel sex period city death rate time_log ;
run;

It worked. Can I use proc mianalyze to combine results from the 10 imputations? If not, what can i use to combine their results?

Re: Multiple imputation using chained equations

SAS_Rob — Wed, 06 Sep 2017 16:26:08 GMT

The way it actually works is that you next run whatever analysis you were planning to run (regression for example) and then use MIANALYZE to combine the estimates from that analysis.

There is a good description of the three step process here in the documentation.

http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_mianalyze_overview.htm

Re: Multiple imputation using chained equations

Alib — Thu, 07 Sep 2017 12:43:16 GMT

I perform univariate competing risk with this syntax but i have many errors from sas

data slevel;
	slevel=1;output;
	slevel=2;output;
	slevel=3;output;
	slevel=4;output;
	format slevel slevel.;
	run;

Proc phreg data=pop1_imput1 plots (overlay=stratum)=cif;
	format slevel slevel.;
	class slevel (param=ref ref='Primary');
	model delriskc1*riskc1(0)=slevel/eventcode=1 risklimits;
	baseline covariates=slevel out=outslevel cif=_all_;
by _Imputation_; 
ods output parameterEstimates  = a_mvn CovB=mxcovb;
	run;

proc mianalyze parms(classvar=full)=a_mvn COVB(effectvar=rowcol)=mxcovb;
class slevel;
modeleffects slevel;
RUN;

ERROR: Event Stack Underflow. This is probably caused by mis-matched begin and end event calls

ERROR: An exception has been encountered.
Please contact technical support and provide them with the following traceback information:

The SAS task name is [PHREG]
ERROR: Read Access Violation PHREG
Exception occurred at (06497DDF)
Task Traceback
Address Frame (DBGHELP API Version 4.0 rev 5)
0000000006497DDF 000000007FC24270 sasods:tkvercn1+0x126D9F
000000000CA383B9 000000007FC24278 sasstgr:tkvercn1+0x57379
000000000CA6465A 000000007FC24BC0 sasstgr:tkvercn1+0x8361A
000000000CA54506 000000007FC24E80 sasstgr:tkvercn1+0x734C6
000000000739677A 000000007FC24E88 sasoda:tkvercn1+0x2573A
000000000738FA18 000000007FC24FC0 sasoda:tkvercn1+0x1E9D8
000000000739677A 000000007FC25000 sasoda:tkvercn1+0x2573A
00000000073724F8 000000007FC25610 sasoda:tkvercn1+0x14B8
000000000F4C6CFA 000000007FC25618 sasmrk:tkvercn1+0x105CBA
000000000F453FD1 000000007FC25B90 sasmrk:tkvercn1+0x92F91
000000000F388A6A 000000007FC25B98 sastags:tkvercn1+0x47A2A
000000000F36C343 000000007FC26180 sastags:tkvercn1+0x2B303
000000000E9CDC6A 000000007FC26188 sasxml:tkvercn1+0x4CC2A
000000000E982548 000000007FC26260 sasxml:tkvercn1+0x1508
000000000653F63A 000000007FC262A0 sasods:tkvercn1+0x1CE5FA
0000000006433576 000000007FC265C0 sasods:tkvercn1+0xC2536
00000000063CEECA 000000007FC26CA0 sasods:tkvercn1+0x5DE8A
000000000653F63A 000000007FC26CE0 sasods:tkvercn1+0x1CE5FA
00000000063E4690 000000007FC2AC70 sasods:tkvercn1+0x73650
000000000653F63A 000000007FC2ACB0 sasods:tkvercn1+0x1CE5FA
00000000063D4C99 000000007FC2CF40 sasods:tkvercn1+0x63C59
000000000F156FBA 000000007FC2CF48 sasphreg:tkvercn1+0x295F7A
000000000F0B0AD1 000000007FC2D600 sasphreg:tkvercn1+0x1EFA91
000000000F0AE1C8 000000007FC2E080 sasphreg:tkvercn1+0x1ED188
000000000EEC15F8 000000007FC2FBF0 sasphreg:tkvercn1+0x5B8
00000000031A8A2B 000000007FC2FBF8 sashost:Main+0x10F0B
00000000031AE66D 000000007FC2FF50 sashost:Main+0x16B4D
00007FFC4BC42774 000000007FC2FF58 KERNEL32:BaseThreadInitThunk+0x14
00007FFC4DE30D51 000000007FC2FF88 ntdll:RtlUserThreadStart+0x21

ERROR: An exception has been encountered.
Please contact technical support and provide them with the following traceback information:

The SAS task name is [PHREG]
ERROR: Read Access Violation PHREG
Exception occurred at (0CA645F9)

ERROR: File WORK.MXCOVB.DATA does not exist.

Thanks

Re: Multiple imputation using chained equations

SAS_Rob — Thu, 07 Sep 2017 18:52:10 GMT

That first ERROR usually has something to do with changes made to the output template. I would contact technical support to ask about the second.

In the mean time you might also try simplifying your PHREG code so that only what you need for MIANALYZE will be produced.

Proc phreg data=pop1_imput1;
	format slevel slevel.;
	class slevel (param=ref ref='Primary');
	model delriskc1*riskc1(0)=slevel/eventcode=1;
by _Imputation_;

ods output parameterEstimates  = a_mvn;
	run;

proc mianalyze parms(classvar=full)=a_mvn;
class slevel;
modeleffects slevel;
RUN;


run;

Re: Multiple imputation using chained equations

Alib — Wed, 13 Sep 2017 09:20:37 GMT

Thanks Rob,

After use of proc mianalyze statement,i used this syntax

data u_slevel2; set u_slevel1;
HR_est=EXP(ESTIMATE);
LCI_HR=HR_est*EXP(-1.96*STDERR);
UCI_HR=HR_est*EXP(+1.96*STDERR);
RUN;

proc print data=u_slevel2;
var Parm slevel Probt HR_est LCI_HR UCI_HR;
run;

and I've had p-value and hazards ratio by class but i need p-value for all categories (global p-value of slevel), how i can get it?