Longitudinal data - programming

[enigmaticepi]

Hi am have a clinical dataset of multiple observations per patient, with the number of visits varying widely. I want to find whether the baseline levels for each patient are abnormal (above a certain value), however, there are a few patients where the first value is missing for one of my variables on the first visit date, so I am struggling with this. At first I used first.id, but realized I wasn't putting everyone in the right boxes because of the missings Any advice how to approach this? 

                                                            id         datechem        CREAT  AST       ALKP    TBILI   N_VISITS

4	15MAR2018	0.8	19	58	0.5	6
4	16MAR2018	0.8	19	58	0.5	6
4	20MAR2018	0.7	18	.	.	6
4	23MAR2018	0.6	23	.	.	6
4	03APR2018	0.6	33	.	.	6
4	06APR2018	0.7	29	.	.	6
5	13JUL2018	.	36	135	.	7
5	18JUL2018	1.1	24	123	0.3	7
5	28JUL2018	0.9	23	92	0.3	7
2.01	07AUG1989	.	.	.	.	7

[Astounding]

Are you asking for the earliest non-missing value for each test? 

Here's an easy way (although it may leave a few stray variables hanging around):

proc sort data=have;

by id descending datechem;

run;

data want;

update have (obs=0) have;

by id;

run;

[PGStats]

@Astounding, I think adding if last.id; would eliminate non meaningful data from the result set.

[Astounding]

@PGStats,

Actually, it's automatic.  UPDATE outputs just the final observation for each BY group.

I might add dropping a variable or two from the result, such as the date variable.

[PGStats]

Had forgotten about that. Thanks @Astounding.

[enigmaticepi]

Hi, thanks for your reply!

So I ran this and it looks like it just printed the last chemdate for each patient.

Not really sure if I just did something wrong. 

I was wanting to print the second observation for ID because it is the first non-missing observation or just get this date. Is there a way to do this? 

I was thinking of using something like this? I already have code for getting the date of the first abnormal value, I just don't have a way to tell if this occurred on the first test or not, so thought if I got a first test date (creatdate below) I could just compare them. 

retain inint creatdate enddate;
keep ptindex creatdate enddate interval ;

if first.ptindex then inint=0;

if (inint=0) and (creat GT .) then do;
creatdate=datechem;
inint=1;
end;

if last.ptindex=1 and inint=1 then do;
interval=(enddate-begdate);
output;
end;

format begdate date9. enddate date9.

Thanks again for your help!

                                                           ID         datechem   CREAT      AST       ALKP  TBILI    N_VISITS

3	15MAR2000	.	19	58	0.5	6
3	16MAR2000	1.6	19	58	0.5	6
3	20MAR2000	0.7	18	.	.	6
3	23MAR2000	0.6	23	.	.	6
3	03APR2000	0.6	33	.	.	6
3	06APR2000	0.7	29	.	.	6
4	13JUL2000	0.8	36	135	0.4	7
4	18JUL2000	0.8	24	123	0.3	7
4	28JUL2000	0.9	23	92	0.3	7
4	07AUG2000	.	.	.	.	7
4	14AUG2000	.	65	.	.	7
4	21AUG2000	.	68	.	.	7
4	28AUG2000	.	66	.	.	7

[mkeintz]

The where filter is your friend;

data want;
  set have (where=(nmiss(ast,alkp)=0));
  by id;
  if first.id;
run;

The nice thing about WHERE filters is that they are exported to the data engine.  So such observations are never seen in the data step.  As a result the "first.id" test will be the first instance of nonmissing AST and ALKP for each id.   BTW, you could just as well use a where STATEMENT as opposed to a where dataset name parameter.

[Astounding]

It should have printed the first DATECHEM for each, not the last.  Make sure you sorted properly.  Or better, yet, just drop DATECHEM.  If you spot check the numbers, they should be 100% correct.

This approach isn't intended to select a single observation.  It will select the first nonmissing CREAT, the first nonmissing AST, the first nonmissing ALKP, etc. for each ID.  They could easily be selected from different incoming observations.