How do I select rows where 2 numbers are closest?

Question

Hi,
I need to select one row from each person where the days are the closest between the rows.
In this example, I have rows of order data, and when a company first set up an account and then a column with the number of days it took for a certain order to be made after they joined.
I need to select order lines where the number of days between joining and ordering are closest.
e.g.
customer date_joined order_date days_between_joining_ordering
1 1/1/2007 10/1/2007 9
1 1/1/2007 12/1/2007 11
1 1/1/2007 12/1/2007 11
1 1/1/2007 20/1/2007 19
2 1/1/2007 1/2/2007 0
2 1/1/2007 11/2/2007 10
2 1/1/2007 30/2/2007 30
So what I am looking to end up with in the results is:
customer date_joined order_date days_between_joining_ordering
1 1/1/2007 12/1/2007 11
2 1/1/2007 11/2/2007 10
So I need one row per customer but where the customers days_between_joining_ordering are the closest together.
Any ideas on how I can get this through SQL or PL/SQL??
Thanks
Daniel

Answer 1

I'm not sure what criteria you want used to uniquely identify each record, so I just used Patient_ID and FU_Days.
Anyway with the supplied data this provides the requested output, though I expect you have more than 2 patients in your full dataset. Are you wanting the two closest records for each possible pairing of patients? Or the lowest ranked records for any given patient relative to all the other patients?
For the first you could probably use this query:
set echo on
with INT_MATCH_TCES as
(select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('10/01/2007','dd/mm/yyyy') post_vl_date, 291 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('12/01/2007','dd/mm/yyyy') post_vl_date, 239 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('12/01/2007','dd/mm/yyyy') post_vl_date, 160 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('20/01/2007','dd/mm/yyyy') post_vl_date, 89 fu_days from dual union all
select 2 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 301 fu_days from dual union all
select 2 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 320 fu_days from dual union all
select 3 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 321 fu_days from dual union all
select 3 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 325 fu_days from dual)
, ranking as (
  select l.patient_id lid, r.patient_id rid, l.fu_days ldays, r.fu_days rdays,
         abs(l.fu_days - r.fu_days) delta,
         rank() over (partition by l.patient_id, r.patient_id order by abs(l.fu_days - r.fu_days)) rnk,
         least(l.patient_id,r.patient_id) o1,
         greatest(l.patient_id,r.patient_id) o2
  from INT_MATCH_TCES l, INT_MATCH_TCES r
  where l.patient_id <> r.patient_id
select a.*, delta
from  INT_MATCH_TCES a, ranking
where a.patient_id = lid
  and a.fu_days = ldays
  and rnk = 1
order by o1,o2, a.patient_id
PATIENT_ID DRUG_LIST        PR_MUTATIONS RT_MUTATIONS POST_VL_DATE FU_DAYS DELTA
1          3TC+ABACAVIR+AZT *            101E         10-JAN-2007  291     10  
2          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  301     10  
1          3TC+ABACAVIR+AZT *            101E         10-JAN-2007  291     30  
3          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  321     30  
2          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  320     1   
3          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  321     1   
6 rows selectedfor the second just rank patient_id from the previous results by the delta e.g.:
set echo on
with INT_MATCH_TCES as
(select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('10/01/2007','dd/mm/yyyy') post_vl_date, 291 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('12/01/2007','dd/mm/yyyy') post_vl_date, 239 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('12/01/2007','dd/mm/yyyy') post_vl_date, 160 fu_days from dual union all
select 1 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('20/01/2007','dd/mm/yyyy') post_vl_date, 89 fu_days from dual union all
select 2 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 301 fu_days from dual union all
select 2 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 320 fu_days from dual union all
select 3 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 321 fu_days from dual union all
select 3 patient_id, '3TC+ABACAVIR+AZT' drug_list, '*' pr_mutations, '101E' rt_mutations, to_date('01/02/2007','dd/mm/yyyy') post_vl_date, 325 fu_days from dual)
, ranking as (
  select l.patient_id lid, r.patient_id rid, l.fu_days ldays, r.fu_days rdays,
         abs(l.fu_days - r.fu_days) delta,
         rank() over (partition by l.patient_id, r.patient_id order by abs(l.fu_days - r.fu_days)) rnk,
         least(l.patient_id,r.patient_id) o1,
         greatest(l.patient_id,r.patient_id) o2
  from INT_MATCH_TCES l, INT_MATCH_TCES r
  where l.patient_id <> r.patient_id
, ranking2 as (
  select a.*, delta,
  rank() over (partition by a.patient_id order by delta) rnk2
  from  INT_MATCH_TCES a, ranking
  where a.patient_id = lid
    and a.fu_days = ldays
    and rnk = 1
  order by o1,o2, a.patient_id
select * from ranking2 where rnk2=1
PATIENT_ID DRUG_LIST        PR_MUTATIONS RT_MUTATIONS POST_VL_DATE FU_DAYS DELTA RNK2
1          3TC+ABACAVIR+AZT *            101E         10-JAN-2007  291     10    1  
2          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  320     1     1  
3          3TC+ABACAVIR+AZT *            101E         01-FEB-2007  321     1     1  
3 rows selected