How can I flag the earliest of overlapping records? - sql

Question

I have a dataset where I need to pick out and keep records that have no overlapping time frames, and for those that do overlap, keep the earliest record.
I have been able to successfully picked out the records that have no overlapping time frames with the below code:
IF OBJECT_ID('tempdb..#overlaps') IS NOT NULL DROP Table #overlaps
SELECT
CASE WHEN EXISTS(SELECT 1 FROM #service r2
WHERE r2.client_ID = r1.client_ID
AND r2.service_ID <> r1.service_ID
AND r1.service_start_date <= r2.service_end_date
AND r2.service_start_date <= r1.service_end_date)
THEN 1
ELSE 0
END AS Overlap
,*
into #overlaps
FROM #services r1
This produces the below for an example client:
Overlap client_ID service_ID service_start_date service_end_date
1 12345 123 27-Oct-2009 03-Jan-2013
1 12345 124 27-Dec-2012 19-Mar-2013
1 12345 125 18-Mar-2013 04-Jun-2014
1 12345 126 29-Jun-2014 28-Apr-2017
1 12345 127 23-Jun-2014 14-Aug-2014
1 12345 128 27-Apr-2015 07-Nov-2015
1 12345 129 01-Aug-2015 01-Dec-2015
0 12345 132 01-Jul-2017 09-Dec-2017
0 12345 133 02-Jan-2018 20-Jan-2018
0 12345 134 03-May-2018 05-Jun-2018
What I want to do, is for where overlap = 1, add a column to flag if that record is the first record of an overlapping "set", first in terms of the start date. The service_ID is not actually sequential, I just replaced it to be as dummy data.
So in the above case, record #1 should be flagged a 1 because it has the earliest start of the service compared to its overlapped service record #2 which started later, so record #2 would be flagged a 0, the same for record #3 (ie. flagged as a 0). Going on, record #4 should be flagged as a 1, as it overlaps the ones records below it.
In terms of the final product, I eventually want to just show any non-overlapping periods, and the earliest/first record for the records that do overlap So in the above scenario, records #1,4, 8,9,10 would remain and the rest would be removed. Each record should remain it's own record though, they should not be "pivoted" up into a continuous record.
In other words, what I need to flag are the earliest record that started where there is more than one active service occurring in parallel.
EDIT:
So for example, client has 4 services: Service A started Jan 1 - July 31, Service B Started Feb 1 ended August 1, Service C started September 1 ended Oct 1, Service D started Nov 1 ended Dec 1...Service A should be flagged as 1, Service B which started while Service A was still active should be flagged 0, Service C started without any service being active will be flagged as 1, same as Service D

Answer 1

I think the flag would be:
SELECT (CASE WHEN NOT EXISTS (SELECT 1
FROM #service r2
WHERE r2.client_ID = r1.client_ID AND
r2.service_ID <> r1.service_ID AND
r1.service_start_date <= r2.service_end_date AND
r2.service_start_date < r1.service_end_date
)
THEN 1
ELSE 0
END) AS First_Overlap;
Notes:
This doesn't actually check for an overlap. I left that out, because you can use the overlaps flag for the check, or include the exists query.
The only difference is < versus <= for the overlap check on the start date.
This might not work as you want when the period of overlap has multiple records beginning at the same time.
Also, I suspect you are trying to solve a gaps-and-islands problem. Using multiple temporary tables and the logic that you are using is unnecessary. You might want to ask another question about the entire problem you want to solve, rather than this one facet.

Answer 2

Its difficult to read your exact goal here, but if you're looking to flag based on the service_start_date, when Overlap = 1. This would suffice.
;WITH CTE (Overlap, client_ID, service_ID, service_start_date) AS (
SELECT * FROM (
VALUES
('1','12345','123','10/27/2009'),
('1','12345','124','12/27/2012'),
('1','12345','125','3/18/2013'),
('1','12345','126','6/29/2014'),
('1','12345','127','6/23/2014'),
('1','12345','128','4/27/2015'),
('1','12345','129','8/1/2015'),
('0','12345','132','7/1/2017'),
('0','12345','133','1/2/2018'),
('0','12345','134','5/3/2018')
) AS A (Overlap, client_ID, service_ID, service_start_date)
)
SELECT CTE.Overlap,
CTE.client_ID,
CTE.service_ID,
CTE.service_start_date,
t2.Result
FROM CTE
LEFT JOIN (
SELECT '1' AS Result,
t2.client_ID,
MIN(t2.service_start_date) AS service_start_date
FROM CTE t2
WHERE t2.Overlap = '1'
GROUP BY client_ID
) t2 ON CTE.client_ID = t2.client_ID
AND CTE.service_start_date = t2.service_start_date
ORDER BY service_ID
This also does not account for anything other than flagging the first Overlap by service_start_date. For instance, if you wanted to flag those that aren't first as 0's, that would need to be added.

Answer 3

UPDATE #overlaps SET IsFirst=1
FROM
(SELECT overlap, client_id client_id, service_start_date service_start_date, service_end_date service_end_date, min(service_id) service_id
FROM #overlaps
WHERE overlap=1
group by overlap, client_id, service_start_date, service_end_date) a
where #overlaps.client_id = a.client_id and #overlaps.service_id = a.service_id
Edit
#marshymell0 - I think I'm understanding what you want. Writing this as a query is pretty tricky, so I'm using a cursor instead. In the section where I have the line PRINT #service_start_date_prev, is where you would update the flag column that determines if the record is the first in the overlapping set.
DECLARE #overlap_prev int, #client_id_prev int, #service_id_prev int
DECLARE #overlap_next int, #client_id_next int, #service_id_next int
DECLARE #service_start_date_prev datetime, #service_end_date_prev datetime
DECLARE #service_start_date_next datetime, #service_end_date_next datetime
DECLARE #part_of_set int = 0
DECLARE o_cursor CURSOR
FOR SELECT overlap, client_id, service_id, service_start_date, service_end_date
FROM #overlaps where overlap=1
ORDER BY service_start_date
OPEN o_cursor
FETCH NEXT FROM o_cursor
INTO #overlap_next, #client_id_next, #service_id_next, #service_start_date_next, #service_end_date_next
WHILE ##FETCH_STATUS = 0
BEGIN
IF (#service_start_date_prev IS NOT NULL)
BEGIN
IF (#part_of_set = 0 AND #service_start_date_prev <= #service_end_date_next AND #service_start_date_next <= #service_end_date_prev)
BEGIN
PRINT #service_start_date_prev
SET #part_of_set = 1
END
ELSE
SET #part_of_set = 0
END
SET #overlap_prev = #overlap_next
SET #client_id_prev = #client_id_next
SET #service_id_prev = #service_id_next
SET #service_start_date_prev = #service_start_date_next
SET #service_end_date_prev = #service_end_date_next
FETCH NEXT FROM o_cursor
INTO #overlap_next, #client_id_next, #service_id_next, #service_start_date_next, #service_end_date_next
END
CLOSE o_cursor;
DEALLOCATE o_cursor;