SQL query for date intervals comparing non-adjacent rows? - sql

Question

I want to flag the first date in every window of at least 31 days for each ID unit in my data.
ROW ID INDEX_DATE
1 ABC 1/1/2019
2 ABC 1/7/2019
3 ABC 1/21/2019
4 ABC 2/2/2019
5 ABC 2/9/2019
6 ABC 3/6/2019
7 DEF 1/5/2019
8 DEF 2/1/2019
9 DEF 2/8/2019
The desired rows are 1, 4, 6, 7 and 9; these are either the first INDEX_DATE for the given ID, or they occur at least 31 days after the previously flagged INDEX_DATE. Every suggestion I have found uses LAG() or LEAD with window functions, but I could only get these to compare adjacent rows. Row 4, for example, needs to be compared to Row 1 in order to be identified as the first after a 31-day window has completed.
I tried the following:
Data
DROP TABLE tTest IF EXISTS;
CREATE TEMP TABLE tTest
(
ROWN INT,
ID VARCHAR(3),
INDEX_DATE DATE
) ;
GO
INSERT INTO tTEST VALUES (1, 'ABC', '1/1/2019');
INSERT INTO tTEST VALUES (2, 'ABC', '1/7/2019');
INSERT INTO tTEST VALUES (3, 'ABC', '1/21/2019');
INSERT INTO tTEST VALUES (4, 'ABC', '2/2/2019');
INSERT INTO tTEST VALUES (5, 'ABC', '2/9/2019');
INSERT INTO tTEST VALUES (6, 'ABC', '3/6/2019');
INSERT INTO tTEST VALUES (7, 'DEF', '1/5/2019');
INSERT INTO tTEST VALUES (8, 'DEF', '2/1/2019');
INSERT INTO tTEST VALUES (9, 'DEF', '2/8/2019');
GO
Query:
DROP TABLE TTEST2 IF EXISTS;
CREATE TEMP TABLE TTEST2 AS (
WITH
RN_CTE(ROWN, ID, INDEX_DATE, RN) AS (
SELECT *, ROW_NUMBER() OVER (PARTITION BY ID ORDER BY INDEX_DATE)
FROM tTEST),
MIN_CTE(ROWN, ID, INDEX_DATE, RN) AS (SELECT * FROM RN_CTE WHERE RN=1),
DIFF_CTE(ROWN,ID, INDEX_DATE, RN, DAY_DIFF) AS (
SELECT RN.*, DATE(RN.INDEX_DATE + INTERVAL '30 DAYS')
FROM RN_CTE AS RN
JOIN MIN_CTE AS MC ON RN.ID=MC.ID
WHERE RN.RN=1
OR RN.INDEX_DATE > MC.INDEX_DATE + INTERVAL '30 DAYS' ),
MIN_DIFF_CTE AS (
SELECT ID, DAY_DIFF, MIN(ROWN) AS MIN_ROW
FROM DIFF_CTE
GROUP BY ID, DAY_DIFF)
SELECT T.*
FROM MIN_DIFF_CTE AS MDC
JOIN tTEST AS T ON MDC.MIN_ROW = T.ROWN
ORDER BY ID, INDEX_DATE
);
Result:
SELECT * FROM TTEST2 ORDER BY ID, INDEX_DATE;
ROWN ID INDEX_DATE
1 ABC 2019-01-01
4 ABC 2019-02-02
5 ABC 2019-02-09
6 ABC 2019-03-06
7 DEF 2019-01-05
9 DEF 2019-02-08
Row 5 with INDEX_DATE = 2019-02-09 should not be in the output because it is less than 31 days after Row 4's INDEX_DATE.

Answer 1

Something like this. The CTE's locate the unique window of the minimum ROW value for each ID transition and 31 days rolling too.
Data
drop table if exists #tTEST;
go
select * INTO #tTEST from (values
(1, 'abc', '1/1/2019'),
(2, 'abc', '1/7/2019'),
(3, 'abc', '1/21/2019'),
(4, 'abc', '2/2/2019'),
(5, 'abc', '2/9/2019'),
(6, 'abc', '3/6/2019'),
(7, 'def', '1/5/2019'),
(8, 'def', '2/1/2019'),
(9, 'def', '2/8/2019')) V([ROW], ID, INDEX_DATE);
Query
;with
rn_cte([ROW], ID, INDEX_DATE, rn) as (
select *, row_number() over (partition by ID order by INDEX_DATE)
from #tTEST),
min_cte([ROW], ID, INDEX_DATE, rn) as (select * from rn_cte where rn=1),
diff_cte([ROW], ID, INDEX_DATE, rn, day_diff) as (
select rn.*, datediff(d, mc.INDEX_DATE, rn.INDEX_DATE)/31
from rn_cte rn
join min_cte mc on rn.ID=mc.ID
where rn.rn=1
or datediff(d, mc.INDEX_DATE, rn.INDEX_DATE)/31>0),
min_diff_cte as (
select ID, day_diff, min([ROW]) min_row
from diff_cte
group by ID, day_diff)
select t.*
from min_diff_cte mdc
join #tTEST t on mdc.min_row=t.ROW
order by 1;
Output
ROW ID INDEX_DATE
1 abc 1/1/2019
4 abc 2/2/2019
6 abc 3/6/2019
7 def 1/5/2019
9 def 2/8/2019