Medication Switches

[SRoubanis]

Hi everyone, 

I have a dataset with a similar structure as the one shown below:

data have;
input PAT_ID RX_DT : ddmmyy10. RX_Name $;
format RX_DT ddmmyy10.;
datalines;;
201 18/07/2013 DrugA
201 27/12/2013 DrugA
201 13/06/2013 DrugC
201 19/08/2013 DrugC
201 01/05/2014 DrugC
201 14/10/2013 DrugC
201 20/02/2014 DrugC
201 27/03/2014 DrugA
201 16/09/2013 DrugB
201 18/12/2013 DrugC
202 16/04/2014 DrugC
202 16/01/2014 DrugC
202 02/10/2013 DrugC
202 15/07/2013 DrugC
202 13/05/2013 DrugC
202 11/03/2014 DrugB
202 11/11/2013 DrugB
202 23/09/2013 DrugA
202 05/08/2013 DrugA
202 24/06/2013 DrugA
;

I would like to count the number of switches per patient from DrugA -> DrugB or from DrugB -> DrugA.

For example, both patients have one swhitch.

Thank you very much for your help in advance!

[ballardw]

This seems to work for your example data:

data have;
   input PAT_ID RX_DT : ddmmyy10. RX_Name $;
   format RX_DT ddmmyy10.;
   ld = lag(rx_name);
   switch = (ld='DrugA' and Rx_name='DrugB') or (ld='DrugB' and Rx_name='DrugA') ;
   drop ld;
datalines;
201 18/07/2013 DrugA
201 27/12/2013 DrugA
201 13/06/2013 DrugC
201 19/08/2013 DrugC
201 01/05/2014 DrugC
201 14/10/2013 DrugC
201 20/02/2014 DrugC
201 27/03/2014 DrugA
201 16/09/2013 DrugB
201 18/12/2013 DrugC
202 16/04/2014 DrugC
202 16/01/2014 DrugC
202 02/10/2013 DrugC
202 15/07/2013 DrugC
202 13/05/2013 DrugC
202 11/03/2014 DrugB
202 11/11/2013 DrugB
202 23/09/2013 DrugA
202 05/08/2013 DrugA
202 24/06/2013 DrugA
;
run;

 

May get cumbersome to use this approach for more specific comparisons though.

And I hope your real data has very consistent spelling.

[Ksharp]

data have;
input PAT_ID RX_DT : ddmmyy10. RX_Name $;
format RX_DT ddmmyy10.;
datalines;;
201 18/07/2013 DrugA
201 27/12/2013 DrugA
201 13/06/2013 DrugC
201 19/08/2013 DrugC
201 01/05/2014 DrugC
201 14/10/2013 DrugC
201 20/02/2014 DrugC
201 27/03/2014 DrugA
201 16/09/2013 DrugB
201 18/12/2013 DrugC
202 16/04/2014 DrugC
202 16/01/2014 DrugC
202 02/10/2013 DrugC
202 15/07/2013 DrugC
202 13/05/2013 DrugC
202 11/03/2014 DrugB
202 11/11/2013 DrugB
202 23/09/2013 DrugA
202 05/08/2013 DrugA
202 24/06/2013 DrugA
;
run;
data temp;
 set have(where=(RX_Name in ('DrugA' 'DrugB')));
 by PAT_ID RX_Name notsorted;
 if first.RX_Name;
run;
proc sql;
select PAT_ID,int(count(*)/2) as n_switch
 from temp
  group by PAT_ID ;
quit;

[mkeintz · Accepted solution]

I like the use of the where and first.id techniques. In the interest of compactness, I think it makes sense to reduce the amount of code to one data step:

data want (keep=pat_id nswitch);

  set have;

  where drug in ('A','B');

  by pat_id drug notsorted;

  if first.pat_id then nswitch=0;

  else if first.drug then nswitch+1;

  if last.pat_id;

run;

m