matching start and end dates

[axescot78]

I have 2 groups of patients that I am matching. One of the matching criteria is medication history range (rx start and end dates) within 5 years. I'm at a loss on how to code this because of all the possible combinations. I want to assign them a score so that the closer the range the lower the score. The farther the range the higher the score. Or, I could also do if the duration is within 5 years then assign 1, otherwise assign 0. However, I prefer the first method.

data groupA;
input pat_id $ rx_start_date:MMDDYY10. rx_end_date:MMDDYY10. ;
format rx_start_date MMDDYY10. rx_end_date MMDDYY10.;
datalines;
A1  1/17/2018  12/25/2018
A2  5/31/2018  11/6/2018
A3  3/7/2018  8/27/2018
A4  2/26/2014  11/22/2014
A5  5/4/2019  7/29/2020
A6  2/7/2016  5/4/2018
A7  4/26/2019  5/27/2019
A8  4/19/2007  5/11/2017
A9  10/8/2008  3/14/2009
A10  9/11/2007  1/26/2015
;

data groupB;
input pat_id $ rx_start_date:MMDDYY10. rx_end_date:MMDDYY10. ;
format rx_start_date MMDDYY10. rx_end_date MMDDYY10.;
datalines;
B1  4/18/2018  6/19/2021
B2  2/15/2008  8/2/2019
B3  2/7/2020  7/12/2020
B4  10/24/2004  5/7/2010
B5  11/20/2003  6/5/2008
B6  4/6/2016  5/17/2016
B7  11/10/2015  2/10/2021
B8  1/9/2015  8/22/2016
B9  8/1/2007  9/23/2009
B10  6/18/2006  8/29/2017
B11  8/28/2002  5/15/2021
B12  7/2/2009  8/6/2009
B13  8/23/2015  12/7/2019
B14  12/12/2011  7/8/2013
B15  6/11/2013  5/14/2016
B16  6/2/2018  7/13/2019
B17  6/6/2013  10/24/2016
B18  9/22/2004  6/16/2020
B19  1/9/2007  8/23/2018
B20  4/23/2013  6/29/2018
;

[SASKiwi]

You haven't explained your matching criteria fully. Is it group A matching with group B or is the reverse (B matching with A) or is it both?

Are you after a one-to-one match, maybe the closest, or are one-to-many matches allowed?

One method that would be useful in deciding the closest match would be the number of days the date ranges actually overlap, then if they don't overlap how close the non-overlapping dates are.

[axescot78]

Thank you for the reply. Group B is the control and the larger group, so I believe that would be matching group A with group B. I am doing a 1:5 match. I am matching on sex, ethnicity, and age within 3 years in addition to Rx duration. However, I can figure out the code for the other 3. It's the date ranges that I haven't gotten yet. I do like your idea about number of days that overlap.

[rrothwell]

It sounds like you are looking to quantify the "closeness" of the start and end dates of a given record in one set to all the records in the other, correct? If so, you could create a score using the DATDIF function to get the difference, and use the absolute value to determine how close they are. The sum of the absolute value of the differences could be the overall score for closeness.