I wrote a program to "look ahead" a little while back and modified it to work with your data as it appears here.  This will not work well if the lines are not separated by a blank line.  Anyway, the technique of reading the buffer and examining it before deciding to output the text may be useful. %let linesize = 80; data input_text (keep = text);     infile datalines eof=eof;  * use eof=label    with datalines,                                              use end=variable with an input file;     length buffer text  $  &linesize;     retain text;     input buffer $ 1-&linesize; * Read the next input line;     if (strip(buffer) = '') then do;   eof:    put 'buffer empty or eof so output "' text '"';         if (strip(text) ^= '') then              output;         text = '';     end;     else do;         put 'buffer not empty so append "' buffer '" to text';         text = catx(' ', text, buffer);     end; datalines; Feb 15, 2012. Testing how data is working. www.test.com Mar 16, 2013  Fake book entry. www.book.com Jan 02, 2011  Test title. Frank Testing www.titlestore.com ; *run; * RUN not needed when reading datalines; /* From log: buffer not empty so append "Feb 15, 2012. Testing how data is working. www.test.com " to text buffer empty or eof so output "Feb 15, 2012. Testing how data is working. www.test.com " buffer not empty so append "Mar 16, 2013  Fake book entry. " to text buffer not empty so append "www.book.com " to text buffer empty or eof so output "Mar 16, 2013  Fake book entry. www.book.com " buffer not empty so append "Jan 02, 2011  Test title. Frank Testing " to text buffer not empty so append "www.titlestore.com " to text buffer empty or eof so output "Jan 02, 2011  Test title. Frank Testing www.titlestore.com " */ proc sql; select * from input_text; /* text -------------------------------------------------------------------------------- Feb 15, 2012. Testing how data is working. www.test.com                          Mar 16, 2013  Fake book entry. www.book.com                                      Jan 02, 2011  Test title. Frank Testing www.titlestore.com                       */ ... View more

One way is to create a table of the possible graphvar values and use a left join with the output table from proc freq.  You loose the cumulative count and frequency, but hopefully that is not important.   data graphvar_tbl; input graphvar cost_range $20.; datalines; 1 $27,350 and Below 2 $27,350-$29,350 3 $29,350-$31,350 4 $31,350-$33,350 5 $33,350-$35,350 6 $35,350-$37,350 7 Above $37,350 ; data test; srvc = 0; graphvar = 1; do i=1 to 6; output; end; graphvar = 3; do i=1 to 21; output; end; graphvar = 4; do i=1 to 83; output; end; graphvar = 5; do i=1 to 16; output; end; graphvar = 6; do i=1 to 2; output; end; graphvar = 7; do i=1 to 4; output; end; run;   PROC FREQ DATA=TEST NLEVELS;   BY srvc;   TABLES GRAPHVAR/LIST MISSING out=test_freq; RUN; proc contents varnum data=test_freq; /* # Variable  Type Len Label 1 srvc      Num  8 2 graphvar  Num  8 3 COUNT     Num  8   Frequency Count 4 PERCENT   Num  8   Percent of Total Frequency */ proc sql; create table test_results as select coalesce(b.srvc,0)               as srvc,        a.graphvar, a.cost_range,        coalesce(b.count,0)              as count,        coalesce(b.percent,0) format 5.2 as percent from      graphvar_tbl A left join test_freq    B on b.graphvar = a.graphvar ; select * from test_results; /* srvc graphvar cost_range count percent ----------------------------------------------------------- 0 1 $27,350 and Below 6  4.55 0 2 $27,350-$29,350   0  0.00 0 3 $29,350-$31,350  21 15.91 0 4 $31,350-$33,350  83 62.88 0 5 $33,350-$35,350  16 12.12 0 6 $35,350-$37,350   2  1.52 0 7 Above $37,350     4  3.03     */   ... View more

Once you have the id_list table, you can easily do left joins to create flags for the sources.  I prefer the flag(0,1) implementation because it makes it easy to sum them to get counts:   proc sql; select count(*) as cnt        from id_list; create table id_summary as select a.id,        case when missing(b1.id) then 0 else 1 end as tbl1_flag 'tbl1 flag',        case when missing(b2.id) then 0 else 1 end as tbl2_flag 'tbl2 flag',        case when missing(b3.id) then 0 else 1 end as tbl3_flag 'tbl3 flag',        calculated tbl1_flag +           calculated tbl2_flag +              calculated tbl3_flag                 as tbl_cnt   'tbl cnt' from   id_list A left join tbl1 B1 on b1.id = a.id left join tbl2 B2 on b2.id = a.id left join tbl3 B3 on b3.id = a.id ; select tbl_cnt, count(*) as cnt from   id_summary group  by tbl_cnt ; select tbl1_flag, tbl2_flag, tbl3_flag, count(*) as cnt from   id_summary group  by tbl1_flag, tbl2_flag, tbl3_flag ; select  sum(tbl1_flag) as tbl1_cnt,         sum(tbl2_flag) as tbl2_cnt,         sum(tbl3_flag) as tbl3_cnt from    id_summary; /*      cnt --------  4400000  tbl cnt       cnt ------------------        1   2800000        2   1400000        3    200000     tbl1      tbl2      tbl3     flag      flag      flag       cnt --------------------------------------        0         0         1    400000        0         1         0    800000        0         1         1    200000        1         0         0   1600000        1         0         1    400000        1         1         0    800000        1         1         1    200000 tbl1_cnt  tbl2_cnt  tbl3_cnt ----------------------------  3000000   2000000   1200000 */   ... View more

I usually create a "discard" table before deleting from the data.  I'm assuming that 'A' is for Alive and 'D' is for Dead, but what is 'R'?  Anyway, the discard table below is populated by any rows where the subject is not 'A', so we can delete any ID from the original table that appears in the discard table.   proc sql; create table discard as select * from   tbl where  status ^= 'A' ; delete from tbl where  id in (select id from discard); ... View more