Much better! Runs slightly slower than the DOW approach (.06 seconds vs .05 seconds).
FWIW: Here is the code I ran to compare the three methods:
data have;
input id event $ start end;
informat start end ddmmyy10.;
format start end date10.;
if id=1 then do id=1 to 10000;
output;
end;
else if id=4 then do id=10001 to 20000;
output;
end;
cards;
1 Headache 30-01-2017 08-02-2017
1 Headache 25-01-2017 03-02-2017
1 vomiting 04-02-2017 05-02-2017
4 Nausea 20-11-2017 31-12-2017
4 Nausea 03-11-2017 05-11-2017
;
proc sort data=have;
by id event;
run;
data every_combination;
do until (last.event);
set have;
by id event;
i+1;
if first.event then j=i;
if last.event then n=i;
end;
do until (last.event);
set have;
by id event;
output;
end;
run;
data every_combination (drop=i j n);
set every_combination;
do k=j to n;
set have (rename=(start=start_date end=end_date)) point=k;
output;
end;
run;
data want;
set have;
do _n_=1 to nobs;
set have(rename=(id=_id event=_event start=start_date end=end_date)) nobs=nobs point=_n_;
if id=_id and event=_event then output;
end;
drop _:;
run;
data want;
if _n_=1 then do;
if 0 then set have(rename=(start=start_date end=end_date));
dcl hash h(dataset:'have(rename=(start=start_date end=end_date))', multidata: 'y', ordered: 'y');
h.definekey('id','event');
h.definedata(all:'y');
h.definedone();
end;
set have;
by id event;
do while(h.do_over(key:id,key:event) eq 0);
output;
end;
run;
Art, CEO, AnalystFinder.com
