creating unique window periods

[dr2014]

Hi ,

below is a table 'have'.  I created the 'duration' column in it that looks at the duration between the 'enddate' of the previous observation and 'startdate' of the next record using the 'lag' function. I flagged any duration <32 as Y. For consecutive observation where the  duration  < 32 , thos have to be grouped together to create a unique period i.e. the dates of obs 1 and 2 have to be combined to get the new_startdate and new_enddate i.e 02/30/2017 to 8/1/2017.

It is possible that more than three obs with duration <32 could be combined to get  the new_startdate and new_enddate. 

Similarly for observation 3 and 4 I had the new_startdate and new_enddate i.e 10/30/2017 to 4/15/2018. I have thousands of such ids for each of which I need to do the steps to get the dataset 'want'

How can I accomplish the dataset 'want'? Please let me know. I need advice urgently.

have:

obs	ptid	index date	start	end	duration	flag
1	1	2/18/2018	02/30/17	5/30/2017
2	1	2/18/2018	6/1/2017	8/1/2017	2.00	y
3	1	2/18/2018	10/30/2017	11/30/2017	90.00	n
4	1	2/18/2018	12/1/2017	4/15/2018	1.00	y

I need to have this dataset 'want' below with the new_start and new_end columns:

obs	ptid	index date	start	end	duration	flag	New_start	New_end
1	1	2/18/2018	02/30/17	5/30/2017			02/30/2017	8/1/2017
2	1	2/18/2018	6/1/2017	8/1/2017	2.00	y	02/30/17	8/1/2017
3	1	2/18/2018	10/30/2017	11/30/2017	90.00	n	10/30/2017	4/15/2018
4	1	2/18/2018	12/1/2017	4/15/2018	1.00	y	10/30/2017	4/15/2018

[novinosrin]

Even a leap year has only 29 days in feb if i m not wrong 🙂

02/30/17

[dr2014]

Please excuse the example. I hope its still understandable. 

[novinosrin]

See if this helps

a lazy solution though, my apologies as caffeine didn't kick in today yet

data have;
input obs	ptid	(indexdate	start	end) (:mmddyy10.)	duration	flag $;
format indexdate	start	end mmddyy10.;
cards;
1	1	2/18/2018	02/28/17	5/30/2017	.	.
2	1	2/18/2018	6/1/2017	8/1/2017	2	y
3	1	2/18/2018	10/30/2017	11/30/2017	90	n
4	1	2/18/2018	12/1/2017	4/15/2018	1	y
;

data w;
set have;
by ptid;
if flag ne 'y' then grp +1;
run;

proc sql;
create table want(drop=grp) as
select *,min(start) as New_start format=mmddyy10.,max(end) as new_end format=mmddyy10.
from w
group by ptid, grp;
quit;

[dr2014]

Thanks for your reply @novinosrinand apologies for the delay. I had to stop working on this concern as data was insufficient and had to look into more urgent tasks. Nonetheless I will test it on available data and reply.