Calling PROC PHREG multiple times

[themanoj20080]

I am trying to use phreg procedure and extract the likelihood value. I can do this for one value of c however I need to repeat this for c=1,...,50. For each value of c, I need to store the likelihood value. So far, I have the following which works just fine one value of c.

proc iml;  
   use test; read all;
proc phreg data = test;
    ods output FitStatistics=Output noprint;
    model times * dead(0) = sexE sexL;
    c = 1;
    if times <= c and sex = 1 then sexE=1; else sexE=0;
    if times > c and sex = 1 then sexL=1; else sexL=0;
 run; 

proc print data=Output noobs;
run;
data _null_;
    set Output;
    if Criterion="-2 LOG L" then call symputx("value",WithCovariates);
run;

%put value= &value;

Any help would be appreciated. Thanks.

[Reeza]

Instead of running the proc multiple times, expand your data set and then run the PROC PHREG once with a BY statement. 

data test_large;
set have;
 do c=1 to 50;
   if times <= c and sex = 1 then sexE=1; else sexE=0;
    if times > c and sex = 1 then sexL=1; else sexL=0;
   output;
end;
run;

proc sort data=test_large;
by c;
run;

proc phreg data=........
BY C;
*everything else here is the same minus the manual coding of the variable;
....
run;

[themanoj20080]

Thank you for the reply. Actually, the manual coding of the variable needs to be inside the proc phreg. Otherwise, the results are not the same.

[Reeza · Accepted solution]

Then you'll need a macro or CALL EXECUTE.

There's an example in this tutorial that's fairly close to what you're trying to do UCLA introductory tutorial on macro variables and macros

https://stats.idre.ucla.edu/sas/seminars/sas-macros-introduction/

Tutorial on converting a working program to a macro

[SteveDenham]

Our shop does this a lot.  If you don't have an indicator variable for each dataset, you will need to create one.  Then you can do something like the following (NOTE: This is a different PROC, so look at the various ODS table names for PHREG to get what you want..):

This part generates the by variables and merges them back onto the original dataset:

/*Process anlsdat byvar*/
proc sort data=anlsdatin out=byvars (keep = paramid parameter) nodupkey;
    by paramid parameter;
run;

data byvars;
    set byvars end=eof;
    by paramid parameter;

    byvar = _N_;

    if eof then call symputx('LastByVar', _N_);
run;

proc sort data=anlsdatin;
    by paramid parameter;
run;

data anlsdat;
    merge anlsdatin byvars;
    by paramid parameter;
run;

This extracts the fit information from the PROC (GLIMMIX here):

%macro runtimes();
%do b=2 %to &LastByVar;
proc sort data=anlsdat (where=(byvar=&b.)) out=anlsdat_&b.;
by byvar paramid parameter grp_no sex animal;
run;

%macro runtests(covtype=, it=);

proc glimmix data=anlsdat_&b. method = laplace;
nloptions maxiter = 5000;
by byvar paramid parameter;
class grp_no sex animal;
model value = grp_no sex grp_no*sex / dist = gamma ddfm=bw;
random grp_no / type = &covtype. subject=animal;
ods output fitstatistics = _rm_fit_&b._&it._1 convergencestatus=_convdat1_&b._&it._1;
run;

data _rm_fit_&b._&it._1;
length covtype $10 value 8 descr $100;
%if %sysfunc(exist(_rm_fit_&b._&it._1)) %then %do; set _rm_fit_&b._&it._1; %end;
covtype = "&covtype.";
Distribution="Gamma";
run;

%mend runtests;/*
%runtests(covtype=CSH, it=1);
%runtests(covtype=CHOL, it=2);

data fit&b.;
set _rm_fit_&b.:;
run;

proc datasets nolist lib=work;
delete _rm_fit_&b.:;
run;

proc sort data=fit&b. out=checkcov&b. (where=(lowcase(descr) = "aicc (smaller is better)" or lowcase(descr) = "aicc (smaller is better)"));
by value;
run;

%let cs = 0;
%let dist = ;
/*
data checkcov&b.;
%if %sysfunc(exist(checkcov&b.)) %then %do; 
set checkcov&b.;
%end;

if _N_ = 1 then do;
call symputx('cs', covtype);
call symputx('dist', Distribution);
selected = 1;
end;
run;
*/
%if "&cs" = "0" %then %do;
%let cs = UN(1);
%let dist = Gamma;
%end;
%mend runtimes

If you use this, be sure that all the %do loops are closed, as I have edited out some things we would rather keep proprietary.

SteveDenham