One way to do this if you don't have too many duplicates, is to put each of duplicate observations into a different data set, then compare the two datasets to each other.

For instance, below I've got up to two observations per my ID. I take only those that have duplicates, then put them into two different data sets, then do a compare on the two different data sets. Note I'm assuming that "visitid" is unique across all patients (if yours is not, create a new ID by concatenating the patient ID and the visit ID together).

proc sql;

create table duplicates as

select *

from data_check

where visitid in(

select visitid

from data_master

group by visitid

having count(visitid) =2)

order by visitid;

quit;

run;

data duplicates1 duplicates2;

set duplicates;

by visitid;

if first.visitid then output duplicates1;

if last.visitid then output duplicates2;

run;

ods trace on;

ods select CompareDifferences;

ods output CompareDifferences=CompareDifferences_set;

proc compare base=duplicates1

compare=duplicates2 NOMISSBASE NOMISSCOMP;

by visitid;

run;

Things would get a lot trickier if you could have more than two of a particular visitid, but perhaps you could pull two at a time and resolve those first.

-Mary

----- Original Message ----- From: idostatistics@GMAIL.COM To: SAS-L@LISTSERV.UGA.EDU Sent: Friday, May 30, 2008 1:33 PM Subject: Using PROC COMPARE , comparing by sub group ( patient visit )

How do you use PROC COMPARE for , not row-by-row comparison, but compare by individuals, say by PATIENT or PATIENT-VISIT, to produce an output dataset that has the rows for all patients or patient-visits for which the data is in any way different?