It appears that your persistence_3m_12M is meant only to identify which case among duplicates to choose. It does not add new information. If so, then:
data want;
set sum;
if _n_=1 then do;
declare hash h (dataset:'persistence_3m_12m');
h.definekey(all:'Y');
h.definedone();
end;
if h.check()=0;
run;
The data do not need to be sorted for this to work, but this program does assume that the dataset persistence_3m_12m is complete - i.e. it identifies every case from sum that you want.
The program create a hash object h, with all the content from persistence_3m_12m. The object is keyed (i.e. it is "indexed" on each combination encountered of all the variable found in persistence_3m_12m). The subsetting if statement merely tests whether the record in sum is found in the hash object. (success here means the h.check() method returns a zero).
Editted note, per your revision of the problem.
If the persistence_3m_12m has other variables to be added to the result, then a minor modification to the hash object definition, and changing h.check() to h.find() will do:
data want;
set sum persistence_3m_12m (obs=0);
if _n_=1 then do;
declare hash h (dataset:'persistence_3m_12m');
h.definekey('enrolid','ndcnum','svcdate','daysupp','copay');
h.definedata(all:'Y');
h.definedone();
end;
if h.find()=0;
run;
In this version, only the variables in common between persistence_3m_12m and sum are used in the definekey statement, while all the variables are in the definedata statement. Unlke the check() method, which only confirms the existence of an appropriate match in hash object h, the find method actually retrieves all the data upon successful matching. So this will bring in the other variables from persistence_3m_12m.
The purpose of the (obs=0) applied to the persistence_3m_12m dataset is to obligate the sas compiler to make provision for any variables in the datasets that are not already in the sum dataset, but without reading in any actual data. The data from that file are read in the declare hash statement.
