Rolling total in SQL that Resets to 0 when going above 90 - sql

Question

First time post. Learning SQL over the past 6 months so help is appreciated. I have data structured as below:
DECLARE #tmp4 as TABLE (
AccountNumber int,
Date date,
DateRank int
)
INSERT INTO #tmp4
VALUES (001, '11/13/2018' , 1)
, (002, '12/19/2018', 2)
, (003, '1/23/2019' , 3)
, (004, '2/5/2019' , 4)
, (005, '3/10/2019' , 5)
, (006, '3/20/2019' , 6)
, (007, '4/8/2019' , 7)
, (008, '5/20/2019' , 8)
What I need to do with this data is calculate a rolling total that resets to 0 once a threshold of 90 days is reached. I have used the DateDiff function to calculate the DateDiffs between consecutive dates and have tried multiple things using LAG and other window functions but can't make it reset. The goal is to find "index visits" which can only occur once every 90 days. So my plan is to have a field that reads 0 on the first visit and resets to 0 for the next stay after 90 days is up from the first visit then only pull visits with a value of 0.
One solution I tried was correct for most sets but did not return the right values for the above set (rows 4 and 8 should start over as "index visits").
The results I would expect for this query would be:
Account Date DateRank RollingTotal
001 |'11/13/2018' | 1 | 0
002 |'12/19/2018' | 2 | 35
003 |'1/23/2019' | 3 | 71
004 |'2/5/2019' | 4 | 84
005 |'3/10/2019' | 5 | 0 (not 117)
006 |'3/20/2019' | 6 | 10
007 |'4/8/2019' | 7 | 29
008 |'5/20/2019' | 8 | 71
Thanks for any help.
Here's the code I tried:
DECLARE #tmp2 as TABLE
(EmrNumber varchar(255)
, AdmitDateTime datetime
, DateRank int
, LagDateDiff int
, RunningTotal int
)
INSERT INTO #tmp2
SELECT tmp1.EmrNumber
, tmp1.AdmitDateTime
, tmp1.DateRank
--, LAG(tmp1.AdmitDateTime) OVER(PARTITION BY tmp1.EmrNumber ORDER BY tmp1.DateRank) as NextAdmitDate
, -DATEDIFF(DAY, tmp1.AdmitDateTime, LAG(tmp1.AdmitDateTime) OVER(PARTITION BY tmp1.EmrNumber ORDER BY tmp1.DateRank)) LagDateDiff
, IIF((SELECT SUM(sumt.total)
FROM (
SELECT -DATEDIFF(DAY, tmpsum.AdmitDateTime, LAG(tmpsum.AdmitDateTime) OVER(PARTITION BY tmpsum.EmrNumber ORDER BY tmpsum.DateRank)) total
FROM #tmp tmpsum
WHERE tmp1.EmrNumber = tmpsum.EmrNumber
AND tmpsum.AdmitDateTime <= tmp1.AdmitDateTime
) sumt) IS NULL, 0, (SELECT SUM(sumt.total)
FROM (
SELECT -DATEDIFF(DAY, tmpsum.AdmitDateTime, LAG(tmpsum.AdmitDateTime) OVER(PARTITION BY tmpsum.EmrNumber ORDER BY tmpsum.DateRank)) total
FROM #tmp tmpsum
WHERE tmp1.EmrNumber = tmpsum.EmrNumber
AND tmpsum.AdmitDateTime <= tmp1.AdmitDateTime
) sumt) ) as RunningTotal
FROM #tmp tmp1
SELECT *
, CASE WHEN LagDateDiff >90 THEN 0
WHEN RunningTotal = 0 THEN 0
ELSE LAG(LagDateDiff) OVER(PARTITION BY EmrNumber ORDER BY DateRank) + RunningTotal END AS RollingTotal
FROM #tmp2

Answer 1

You need a recursive query for this, because the running total has to be checked iteratively, row after row:
with cte as (
select
Account,
Date,
DateRank,
0 RollingTotal
from #tmp4
where DateRank = 1
union all
select
t.Account,
t.Date,
t.DateRank,
case when RollingTotal + datediff(day, c.Date, t.Date) > 90
then 0
else RollingTotal + datediff(day, c.Date, t.Date)
end
from cte c
inner join #tmp4 t on t.DateRank = c.DateRank + 1
)
select * from cte
The anchor of the cte selects the first record (as indicated by DateRank. Then, the recursive part processes rows one by one, and resets the running count when it crosses 90.