DATA Step, Macro, Functions and more

Eliminate Non-Overlapping Terms

Accepted Solution Solved
Reply
Regular Contributor
Posts: 247
Accepted Solution

Eliminate Non-Overlapping Terms

I have a list of adverse events (AEs), and I am supposed to find if a subject has any overlapping terms and output them into a report.  I have been able to do so, but on rare occasions, I have a term that is not overlapping in some places but is in others and is outputting all records.  I would appreciate some help finding a way to eliminate the records that are not truly overlapping.

 

For example, Subject 101-002 is showing that "Urinary tract infection" is overlapping, which is true for AESPID 2 & 4; however, AESPID=1 is not overlapping.  I want to get rid of this.

 

SUBJECTNUMBERSTRAESPIDLLT_NAMEAESTDATAEENDAT
101-0021Urinary tract infection03AUG201715AUG2017
101-0022Urinary tract infection07SEP2017 
101-0024Urinary tract infection03OCT2017 

 

 My method for getting to this point is:

  1. Get a list of all Subject/LLT records where there is a repeat of LLT_NAME
  2. Compare all AESTDAT & AEENDAT values to see if there is an overlap and flag
  3. Keep distinct Subject/LLT records with a flag
  4. Merge the overlapping records with the original data and only keep overlapping terms

 Here is the code I have.  Please feel free to modify to help me remove AESPID=1

 

data AEALL;
	input @1 SUBJECTNUMBERSTR $7. @9 AESPID @13 LLT_NAME $30. @45 AESTDAT date9. @57 AEENDAT date9.;
	format AESTDAT AEENDAT date9.;
	cards;
101-001	1	Acute respiratory failure		05AUG2017	.
101-001	2	Pleural effusion				05AUG2017	.
101-001	3	Pneumonia						07AUG2017	14AUG2017
101-001	4	Septic shock					16AUG2017	22AUG2017
101-001	5	Pneumothorax traumatic			19AUG2017	06SEP2017
101-001	6	Anemia							20AUG2017	.
101-001	7	Pneumothorax traumatic			28AUG2017	09SEP2017
101-001	8	Pneumonia						29AUG2017	08SEP2017
101-001	9	Hypotension						04SEP2017	07SEP2017
101-001	10	Hypernatremia					12AUG2017	23AUG2017
101-001	11	Thrombocytosis					15AUG2017	20AUG2017
101-001	12	Paroxysmal atrial fibrillation	13SEP2017	.
101-001	13	Respiratory infection			15SEP2017	.
101-002	1	Urinary tract infection			03AUG2017	15AUG2017
101-002	2	Urinary tract infection			07SEP2017	.
101-002	3	Pressure ulcer					11SEP2017	.
101-002	4	Urinary tract infection			03OCT2017	.
;
run;


proc sort data=aeall;
	by subjectnumberstr llt_name aestdat aeendat;
	where llt_name^='';
run;

data duplicate;
	set aeall;
	by subjectnumberstr llt_name;
	if first.llt_name & last.llt_name then delete;
run;

proc sort data=duplicate;
	by subjectnumberstr llt_name aespid;
run;


data overlap;
	set duplicate;
	by subjectnumberstr llt_name aespid;
	format prev_stdat prev_endat date9.;
	prev_stdat=lag(aestdat);
	prev_endat=lag(aeendat);
	if first.llt_name then do;
		prev_stdat=.;
		prev_endat=.;
	end;
	else do;
		if (prev_endat^=. & aestdat^=. & prev_stdat<=aestdat & aestdat<=prev_endat) or
			(prev_endat=. & aestdat^=. & prev_stdat<=aestdat) or
			(prev_endat^=. & aestdat^=. & aestdat=prev_endat+1) or
			(aestdat=.) then flag=1;
	end;
run;

proc sort data=overlap out=overlap_distinct (keep=subjectnumberstr llt_name) nodupkey;
	by subjectnumberstr llt_name;
	where flag=1;
run;


data final;
	merge aeall overlap_distinct (in=a);
	by subjectnumberstr llt_name;
	if a;
run;
proc sort data=final nodupkey;
	by subjectnumberstr LLT_NAME AESPID AESTDAT AEENDAT;
run;

 


Accepted Solutions
Solution
‎01-11-2018 04:54 PM
Super User
Posts: 6,629

Re: Eliminate Non-Overlapping Terms

[ Edited ]
Posted in reply to djbateman

I hope I'm not simplifying this too much, to determine when there is an overlap.  You might need to fiddle with the conditions.

 

The idea is not to delete anything until the end.  Instead, assign a new variable (GROUP) that will be the same for all instances that overlap with one another.

 

proc sort data=AEALL;

by SubjectNumberStr llt_name aestdat aeendat;

run;

 

data with_groups;

set AEALL;

by SubjectNumberStr llt_name aestdat;

prev_enddate = lag(aeendat);

if first.llt_name

or (first.llt_name=0 and  (prev_enddate=. or aestdat <= prev_enddate))

then group + 1;

run;

 

If GROUP can be accurately defined, the records are easy to subset:

 

data overlapping;

set with_groups;

by group;

if first.group and last.group then delete;

run;

 

Note that you don't have to sort by GROUP ... the data set is already sorted.  But the trick is to increment GROUP properly.  See how close this comes to the result  you want.

 

View solution in original post


All Replies
Solution
‎01-11-2018 04:54 PM
Super User
Posts: 6,629

Re: Eliminate Non-Overlapping Terms

[ Edited ]
Posted in reply to djbateman

I hope I'm not simplifying this too much, to determine when there is an overlap.  You might need to fiddle with the conditions.

 

The idea is not to delete anything until the end.  Instead, assign a new variable (GROUP) that will be the same for all instances that overlap with one another.

 

proc sort data=AEALL;

by SubjectNumberStr llt_name aestdat aeendat;

run;

 

data with_groups;

set AEALL;

by SubjectNumberStr llt_name aestdat;

prev_enddate = lag(aeendat);

if first.llt_name

or (first.llt_name=0 and  (prev_enddate=. or aestdat <= prev_enddate))

then group + 1;

run;

 

If GROUP can be accurately defined, the records are easy to subset:

 

data overlapping;

set with_groups;

by group;

if first.group and last.group then delete;

run;

 

Note that you don't have to sort by GROUP ... the data set is already sorted.  But the trick is to increment GROUP properly.  See how close this comes to the result  you want.

 

Regular Contributor
Posts: 247

Re: Eliminate Non-Overlapping Terms

Posted in reply to Astounding
From my initial glance, this seems to be doing what I want. I'll have to dig into this more tomorrow. Thank you for your reply!
☑ This topic is solved.

Need further help from the community? Please ask a new question.

Discussion stats
  • 2 replies
  • 106 views
  • 0 likes
  • 2 in conversation