Extending end date then compare repeatedly - sql

Question

What I'd like to do is to extend RxEndDates until there is no more overlap in the prescriptions. And new extensions do not overlap either.
Context: If Amy takes Humera daily and gets a refill before her current prescription runs out, then add the DaySupply of the 2nd prescription to the first prescription.
sample data:
User Drug RxStartDate DaySupply RxEndDate
Amy Humera 2/12/2017 7 2/18/2017
Amy Humera 2/28/2017 5 3/4/2017 <--Overlap with below
Amy Humera 3/3/2017 5 3/7/2017 <--Overlap with above, need to combine
Amy Humera 3/8/2017 2 3/9/2017
Amy Humera 3/10/2017 7 3/16/2017
Amy Humera 3/17/2017 30 4/15/2017 <--Overlap with all below, combine
Amy Humera 3/22/2017 2 3/23/2017 <--Overlap
Amy Humera 3/24/2017 2 3/25/2017 <--Overlap
Amy Humera 3/31/2017 3 4/2/2017 <--Overlap
Amy Humera 4/7/2017 5 4/11/2017 <--Overlap
Amy Humera 4/13/2017 30 5/12/2017 <--Overlap
So after we combine, we get
User Drug RxStartDate DaySupply RxEndDate
Amy Humera 2/12/2017 7 2/18/2017
Amy Humera 2/28/2017 10 3/9/2017 <-- Combined from above, new overlap
Amy Humera 3/8/2017 2 3/9/2017 <-- Now this overlaps with above
Amy Humera 3/10/2017 7 3/16/2017
Amy Humera 3/17/2017 72 5/27/2017
User Drug RxStartDate DaySupply RxEndDate
Amy Humera 2/12/2017 7 2/18/2017
Amy Humera 2/28/2017 12 3/11/2017 <-- Combined, again, new overlap
Amy Humera 3/10/2017 7 3/16/2017 <-- Now this overlaps with above
Amy Humera 3/17/2017 72 5/27/2017
User Drug RxStartDate DaySupply RxEndDate
Amy Humera 2/12/2017 7 2/18/2017
Amy Humera 2/28/2017 19 3/18/2017 <-- Combined, again, new overlap
Amy Humera 3/17/2017 72 5/27/2017 <-- Now this overlaps with above
User Drug RxStartDate DaySupply RxEndDate
Amy Humera 2/12/2017 7 2/18/2017
Amy Humera 2/28/2017 91 5/29/2017
There is no more overlap…finished!
Is there a way to do this automatically in a loop or something...any ideas?

Answer 1

I think the solution can only be implemented by recursion, as there should be a loop that calculates the accumulated DaySupply and I see no way of doing that with any non-recursive lookups. You can do this with recursive CTE - and according to the official doc, it is available starting with SQL Server 2008.
A possible implementation (I added some test data to challenge it):
DECLARE #test TABLE (
[User] VARCHAR(100),
Drug VARCHAR(100),
RxStartDate DATE,
DaySupply INT,
RxEndDate DATE
)
INSERT #test
VALUES
('Amy', 'Humera', '2/12/2017', '7', '2/18/2017'),
('Amy', 'Humera', '2/28/2017', '5', '3/4/2017'),
('Amy', 'Humera', '3/3/2017', '5', '3/7/2017'),
('Amy', 'Humera', '3/8/2017', '2', '3/9/2017'),
('Amy', 'Humera', '3/10/2017', '7', '3/16/2017'),
('Amy', 'Humera', '3/17/2017', '30', '4/15/2017'),
('Amy', 'Humera', '3/22/2017', '2', '3/23/2017'),
('Amy', 'Humera', '3/24/2017', '2', '3/25/2017'),
('Amy', 'Humera', '3/31/2017', '3', '4/2/2017'),
('Amy', 'Humera', '4/7/2017', '5', '4/11/2017'),
('Amy', 'Humera', '4/13/2017', '30', '5/12/2017'),
('Amy', 'Other', '3/24/2017', '7', '3/30/2017'),
('Amy', 'Other', '3/31/2017', '3', '4/2/2017'),
('Amy', 'Other', '4/7/2017', '5', '4/11/2017'),
('Amy', 'Other', '4/13/2017', '30', '5/12/2017'),
('Joe', 'Humera', '3/24/2017', '8', '3/31/2017'),
('Joe', 'Humera', '3/31/2017', '3', '4/2/2017'),
('Joe', 'Humera', '4/12/2017', '5', '4/16/2017'),
('Joe', 'Humera', '4/23/2017', '30', '5/22/2017'),
('Joe', 'Other', '3/24/2017', '60', '5/23/2017'),
('Joe', 'Other', '3/31/2017', '3', '4/2/2017'),
('Joe', 'Other', '4/7/2017', '5', '4/11/2017'),
('Joe', 'Other', '4/13/2017', '30', '5/12/2017')
-- You can comment this out, it is just to show progress:
SELECT * FROM #test ORDER BY [User], Drug, RxStartDate
DECLARE #test_2 TABLE (
[User] VARCHAR(100),
Drug VARCHAR(100),
RxStartDate_base DATE,
DaySupplyCumulative INT
)
;WITH CTE_RxEndDateExtended as (
SELECT [User], Drug, RxStartDate, DaySupply, DaySupply as DaySupplyCumulative, RxStartDate as RxStartDate_base, RxStartDate as RxStartDateExtended, dateadd (dd, DaySupply, RxStartDate) as RxEndDateExtended
FROM #test
-- WHERE [User] = 'Amy' and Drug = 'Humera' and RxStartDate = '2/28/2017'
UNION ALL
SELECT t.[User], t.Drug, t.RxStartDate, t.DaySupply, c.DaySupplyCumulative + t.DaySupply as DaySupplyCumulative, c.RxStartDate_base, t.RxStartDate as RxStartDateExtended, dateadd (dd, t.DaySupply, c.RxEndDateExtended) as RxEndDateExtended
FROM CTE_RxEndDateExtended as c INNER JOIN #test as t
on c.[User] = t.[User] and c.Drug = t.Drug
and c.RxEndDateExtended >= t.RxStartDate and c.RxStartDateExtended < t.RxStartDate
)
INSERT #test_2
SELECT [User], Drug, RxStartDate_base, MAX (DaySupplyCumulative) as DaySupplyCumulative -- comment this out and use this for debugging: SELECT *
FROM CTE_RxEndDateExtended
GROUP BY [User], Drug, RxStartDate_base -- comment this out for debugging
OPTION (MAXRECURSION 0) -- comment this out and use this for debugging (to avoid infinite loops): OPTION (MAXRECURSION 1000)
-- You can comment this out, it is just to show progress:
SELECT * FROM #test_2
ORDER BY [User], Drug, RxStartDate_base -- comment this out and use this for debugging: ORDER BY [User], Drug, RxStartDate_base, RxStartDate, DaySupplyCumulative
SELECT base.*, dateadd (dd, base.DaySupplyCumulative - 1, base.RxStartDate_base) as RxEndDateCumulative
FROM #test_2 as base LEFT OUTER JOIN #test_2 as filter
on base.[User] = filter.[User] and base.Drug = filter.Drug
and base.RxStartDate_base > filter.RxStartDate_base
and dateadd (dd, base.DaySupplyCumulative, base.RxStartDate_base) <= dateadd (dd, filter.DaySupplyCumulative, filter.RxStartDate_base)
WHERE filter.[User] IS NULL
ORDER BY [User], Drug, RxStartDate_base
Maybe you need to optimize it by simplifying the logic. But be careful not to make an infinite loop. When debugging use OPTION (MAXRECURSION N) with N other than zero.
PS.: this one works also if I add 'Amy', 'Humera', '2/15/2017', '11', '2/25/2017', with which I was criticizing the other solutions... I am curious if it works as you expect - please test!

Answer 2

You can identify where a group starts, using not exists. Then do a cumulative sum to assign a group . . . and aggregate. The following assumes a unique id, which is sort of needed to handle duplicates:
select [user], drug, grp, sum(daysupply), min(RxStartDate), max(RxEndDate)
from (select t.*, sum(flg) over (partition by [user], drug order by RxStartDate) as grp
from (select t.*,
(case when exists (select 1
from #test t2
where t2.[user] = t.[user] and t2.drug = t.drug and
t2.RxStartDate < t.RxStartDate and
t2.RxEndDate >= dateadd(day, -1, t.RxStartDate)
)
then 0 else 1
end) as flg
from #test t
) t
) t
group by [user], drug, grp;

Answer 3

I used a CTE Common Table Expression to perform the grouping. Since some of the days don't technically overlap, I created an alternate end date [RxEndDate_ALT] by adding 1 to the [RxEndDate] in the source_data. Then I was able to group the dates using NOT EXISTS in source_data_grouped. After that, I join back to the source_data_raw to SUM the [DaySupply].
Results
SQL
WITH
source_data_raw
AS
(
SELECT tbl.* FROM (VALUES
( 'Amy', 'Humera', 7, CAST('12-Feb-2017' AS DATE), CAST('18-Feb-2017' AS DATE))
, ( 'Amy', 'Humera', 5, '28-Feb-2017', '04-Mar-2017')
, ( 'Amy', 'Humera', 5, '03-Mar-2017', '07-Mar-2017')
, ( 'Amy', 'Humera', 2, '08-Mar-2017', '09-Mar-2017')
, ( 'Amy', 'Humera', 7, '10-Mar-2017', '16-Mar-2017')
, ( 'Amy', 'Humera', 30, '17-Mar-2017', '15-Apr-2017')
, ( 'Amy', 'Humera', 2, '22-Mar-2017', '23-Mar-2017')
, ( 'Amy', 'Humera', 2, '24-Mar-2017', '25-Mar-2017')
, ( 'Amy', 'Humera', 3, '31-Mar-2017', '15-Apr-2017')
, ( 'Amy', 'Humera', 5, '07-Apr-2017', '16-Apr-2017')
, ( 'Amy', 'Humera', 30, '13-Apr-2017', '27-May-2017')
) tbl ([User], [Drug], [DaySupply], [RxStartDate], [RxEndDate])
)
,
source_data
AS
(
SELECT
sdr.[User]
, sdr.[Drug]
, sdr.[RxStartDate]
, sdr.[RxEndDate]
, [RxEndDate_ALT] = DATEADD(DAY, 1, sdr.[RxEndDate])
FROM
source_data_raw AS sdr
)
,
source_data_grouped
AS
(
SELECT
s1.[User]
, s1.[Drug]
, s1.[RxStartDate]
, [RxEndDate] = MIN(t1.[RxEndDate])
FROM
source_data AS s1
INNER JOIN source_data AS t1 ON s1.[User] = t1.[User] AND s1.[Drug] = t1.[Drug] AND s1.[RxStartDate] <= t1.[RxEndDate_ALT]
AND NOT EXISTS
(
SELECT 1
FROM source_data AS t2
WHERE
1=1
AND t1.[User] = t2.[User]
AND t1.[Drug] = t2.[Drug]
AND t1.[RxEndDate_ALT] >= t2.[RxStartDate]
AND t1.[RxEndDate_ALT] < t2.[RxEndDate_ALT]
)
WHERE
1=1
AND NOT EXISTS
(
SELECT 1
FROM source_data AS s2
WHERE
1=1
AND s1.[User] = s2.[User]
AND s1.[Drug] = s2.[Drug]
AND s1.[RxStartDate] > s2.[RxStartDate]
AND s1.[RxStartDate] <= s2.[RxEndDate_ALT]
)
GROUP BY
s1.[User]
, s1.[Drug]
, s1.[RxStartDate]
)
SELECT
sdg.[User]
, sdg.[Drug]
, [DaySupply] = SUM(sdr.[DaySupply])
, sdg.[RxStartDate]
, sdg.[RxEndDate]
FROM
source_data_grouped AS sdg
INNER JOIN source_data_raw AS sdr ON sdr.[RxStartDate] BETWEEN sdg.[RxStartDate] AND sdg.[RxEndDate]
GROUP BY
sdg.[User]
, sdg.[Drug]
, sdg.[RxStartDate]
, sdg.[RxEndDate]