为特定子群体创建连续注册孤岛

Question

我的最终目标是为单个子群体的每个 CLIENTID 创建连续注册天数的孤岛：'Adult Expansion' 日历年 2019 和 2020。一个 CLIENTID 可以是在一个日历年内与多个子人群相关联，但永远不能同时与多个子人群相关联（入学时没有重叠）。我的数据可以追溯到 2016 年，但我只对 2019 年和 2020 年感兴趣。数据的结构是每一行都是一个注册期，有注册的开始和结束日期，与一个子人群相关联。

我在下面包含了一些虚拟数据和所需的输出以更好地说明我的目标：

CREATE TABLE #t (
CLIENTID NVARCHAR(9),
DEMONSTRATION_POPULATION NVARCHAR(30),
ELIGBEGIN DATE,
ELIGEND DATE, 
AGE INT
)
INSERT INTO #t
VALUES 
('123456789', 'Adult Expansion', '2019-12-16', '2019-12-31', 52)
, ('123456789', 'Adult Expansion', '2020-01-01', '2020-01-15', 52)
, ('123456789', 'Adult Expansion', '2020-03-01', '2020-03-31', 52)
, ('123456789', 'Adult Expansion', '2020-04-01', '2020-04-30', 52)
, ('123456789', 'Adult Expansion', '2020-05-01', '2020-05-31', 52)
, ('123456789', 'Adult Expansion', '2020-06-01', '2020-06-30', 52)
, ('123456789', 'Adult Expansion', '2020-07-01', '2020-07-31', 52)
, ('123456789', 'Adult Expansion', '2020-08-01', '2020-08-31', 52)
, ('123456789', 'Adult Expansion', '2020-09-01', '2020-09-30', 52)
, ('123456789', 'Adult Expansion', '2020-10-01', '2020-10-31', 52)
, ('123456789', 'Adult Expansion', '2020-11-01', '2020-11-30', 52)
, ('123456789', 'Adult Expansion', '2020-12-01', '2020-12-31', 52)
------------------------NEW CLIENTID-----------------------------
,('012345678', 'Demonstration Population #3', '2019-10-01', '2019-10-31', 52)
,('012345678', 'Demonstration Population #3', '2019-11-01', '2019-11-30', 52)
,('012345678', 'Demonstration Population #3', '2019-12-01', '2019-12-31', 52)
,('012345678', 'Demonstration Population #3', '2020-01-01', '2020-01-31', 52)
,('012345678', 'Adult Expansion', '2020-02-01', '2020-02-28', 52)
,('012345678', 'Demonstration Population #3', '2020-02-29', '2020-02-29', 52)
,('012345678', 'Adult Expansion', '2020-03-01', '2020-03-31', 52)
,('012345678', 'Adult Expansion', '2020-04-01', '2020-04-30', 52)
,('012345678', 'Adult Expansion', '2020-05-01', '2020-05-31', 52)
,('012345678', 'Adult Expansion', '2020-06-01', '2020-06-30', 52)
,('012345678', 'Adult Expansion', '2020-07-01', '2020-07-31', 52)
,('012345678', 'Adult Expansion', '2020-08-01', '2020-08-31', 52)
,('012345678', 'Adult Expansion', '2020-09-01', '2020-09-30', 52)
,('012345678', 'Adult Expansion', '2020-10-01', '2020-10-31', 52)
,('012345678', 'Adult Expansion', '2020-11-01', '2020-11-30', 52)
,('012345678', 'Adult Expansion', '2020-12-01', '2020-12-31', 52)
---------------------------NEW CLIENTID---------------------------
,('020234587', 'Adult Expansion',   '2019-06-01', '2019-06-30', 36)
,('020234587', 'Adult Expansion',   '2019-08-01', '2019-08-31', 36)
,('020234587', 'Adult Expansion',   '2019-09-01', '2019-09-30', 36)
,('020234587', 'Adult Expansion',   '2019-10-01', '2019-10-31', 36)
,('020234587', 'Adult Expansion',   '2019-11-01', '2019-11-30', 36)
,('020234587', 'Non-1115-Waiver',   '2019-12-01', '2019-12-31', 36)
,('020234587', 'Non-1115-Waiver',   '2020-01-01', '2020-01-31', 36)
,('020234587', 'Non-1115-Waiver',   '2020-02-01', '2020-02-29', 36)

期望输出：

CLIENTID	AGE	ELIGBEGIN	ELIGEND	Sequence_ID
123456789	52	2019-12-19	2019-12-31	1
123456789	52	2020-01-01	2020-01-15	2
123456789	52	2020-03-01	2020-12-31	3
012345678	52	2020-02-01	2020-28-20	1
012345678	52	2020-03-01	2020-03-31	2
020234587	36	2019-06-01	2019-06-30	1
020234587	36	2019-08-01	2019-11-30	2

然后这是我用来尝试解决这个问题的当前代码。

SELECT * INTO #y1 FROM #t
WHERE YEAR(ELIGBEGIN) = '2019'

SELECT s1.CLIENTID
       , s1.AGE
       , CAST(s1.ELIGBEGIN AS DATETIME)    AS ELIGBEGIN
       , MIN(CAST(t1.ELIGEND AS DATETIME)) AS ELIGEND
       , ROW_NUMBER() OVER(PARTITION BY s1.CLIENTID ORDER BY CAST(s1.ELIGBEGIN AS DATETIME)) AS Sequence_ID
INTO #CY19
FROM --[dbo].[Eligs] 
 --#t s1
 #y1 s1
INNER JOIN #y1--#t
t1 ON t1.CLIENTID = s1.CLIENTID
                   AND CAST(s1.ELIGBEGIN AS DATETIME) <= CAST(t1.ELIGEND AS DATETIME)
                   AND t1.DEMONSTRATION_POPULATION = 'Adult Expansion'
    AND NOT EXISTS
    (
    SELECT * FROM #y1 t2--#t t2
                 WHERE t2.CLIENTID = t1.CLIENTID
                   AND (CAST(t1.ELIGEND AS DATETIME) + 1) >= CAST(t2.ELIGBEGIN AS DATETIME)
                         AND CAST(t1.ELIGEND AS DATETIME) < CAST(t2.ELIGEND AS DATETIME)
                   )
WHERE --s1.DEMONSTRATION_POPULATION = 'Adult Expansion' AND 
    NOT EXISTS
    (
    SELECT * FROM #y1 s2--#t s2
                 WHERE s2.CLIENTID = s1.CLIENTID
                   AND CAST(s1.ELIGBEGIN AS DATETIME)       > CAST(s2.ELIGBEGIN AS DATETIME)
                   AND (CAST(s1.ELIGBEGIN AS DATETIME) - 1) <= CAST(s2.ELIGEND AS DATETIME)
                   )
        --AND s1.DEMONSTRATION_POPLUATION = 'Adult Expansion'
        --AND s1.UMIC = '1'
GROUP BY s1.CLIENTID, CAST(s1.ELIGBEGIN AS DATETIME), s1.AGE, s1.ELIGBEGIN, s1.DEMONSTRATION_POPULATION
ORDER BY s1.CLIENTID, CAST(s1.ELIGBEGIN AS DATETIME)

--do the same for the 2020 calendar year:
SELECT * INTO #y2
FROM #t
WHERE YEAR(ELIGBEGIN) = '2020'

SELECT s1.CLIENTID
       , s1.AGE
       , CAST(s1.ELIGBEGIN AS DATETIME)    AS ELIGBEGIN
       , MIN(CAST(t1.ELIGEND AS DATETIME)) AS ELIGEND
       , ROW_NUMBER() OVER(PARTITION BY s1.CLIENTID ORDER BY CAST(s1.ELIGBEGIN AS DATETIME)) AS Sequence_ID
INTO #CY20
FROM --[dbo].[Eligs] 
 --#t s1
 #y2 s1
INNER JOIN #y2--#t
t1 ON t1.CLIENTID = s1.CLIENTID
                   AND CAST(s1.ELIGBEGIN AS DATETIME) <= CAST(t1.ELIGEND AS DATETIME)
                   AND s1.DEMONSTRATION_POPULATION = 'Adult Expansion'
    AND NOT EXISTS
    (
    SELECT * FROM #y2 t2--#t t2
                 WHERE t2.CLIENTID = t1.CLIENTID
                   AND (CAST(t1.ELIGEND AS DATETIME) + 1) >= CAST(t2.ELIGBEGIN AS DATETIME)
                         AND CAST(t1.ELIGEND AS DATETIME) < CAST(t2.ELIGEND AS DATETIME)
                   )
WHERE --s1.DEMONSTRATION_POPLUATION = 'Adult Expansion' 
    --AND 
    NOT EXISTS
    (
    SELECT * FROM #y2 s2--#t s2
                 WHERE s2.CLIENTID = s1.CLIENTID
                   AND CAST(s1.ELIGBEGIN AS DATETIME)       > CAST(s2.ELIGBEGIN AS DATETIME)
                   AND (CAST(s1.ELIGBEGIN AS DATETIME) - 1) <= CAST(s2.ELIGEND AS DATETIME)
                   )
        --AND s1.DEMONSTRATION_POPLUATION = 'Adult Expansion'
        --AND s1.UMIC = '1'
GROUP BY s1.CLIENTID, CAST(s1.ELIGBEGIN AS DATETIME), s1.AGE
--ORDER BY s1.CLIENTID, CAST(s1.ELIGBEGIN AS DATETIME)
;

SELECT CLIENTID
, AGE
, ELIGBEGIN
, ELIGEND
, Sequence_ID = ROW_NUMBER() OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN)
INTO #testcase
FROM (
SELECT * FROM #CY19
UNION
SELECT * FROM #CY20) a

SELECT * FROM #testcase
ORDER BY CLIENTID

然而，最终SELECT * FROM #testcase的实际输出如下：

实际结果：

CLIENTID	AGE	ELIGBEGIN	ELIGEND	Sequence_ID
123456789	52	2019-12-19	2019-12-31	1
123456789	52	2020-01-01	2020-01-15	2
123456789	52	2020-03-01	2020-12-31	3
020234587	36	2019-06-01	2019-06-30	1
020234587	36	2019-08-01	2019-11-30	2

如您所见，我面临三个主要问题，即实际输出与所需输出不匹配。

我必须运行在不同的年份两次进行相同的查询，因为我不知道如何对从 2019-12-31 到 2020-01 连续注册的会员进行细分注册- 31岁及以上。如果我运行代码 table #t 而不是 table #y1，第一个 CLIENTID 的输出将是 12/16/19 到 1/15/20，我不想要。
我完全失去了一个 CLIENTID，因为他们在 'Adult Expansion' 子群体中，然后在 'Demonstration Population #3' 中，然后又回到 'Adult Expansion'
出于某种原因，代码将正确忽略前面的行，其中客户的 DEMONSTRATION_POPULATION != 'Adult Expansion' 但如果 CLIENTID 从 'Adult Expansion' 更改为不同的子群体，则实际输出仍包括注册日期来自这个不同的子群体，这直接反对我想要的输出。当一个人在 'Adult Expansion'.

如有哪位大侠能协助解决以上三个问题，将不胜感激！

提前感谢任何提示！

Answer 1

这是一个使用间隙和孤岛方法的解决方案：

;WITH prevNextCTE
AS
(
    SELECT CLIENTID,
           DEMONSTRATION_POPULATION,
           ELIGBEGIN,
           ELIGEND,
           AGE,
           -- compare the year of the previous record to the current to create split at year end
           CASE WHEN YEAR(LAG(ELIGEND,1) OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN)) = YEAR(ELIGBEGIN)
                THEN LAG(ELIGEND,1) OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN)
           END AS prevELIGEND,
           -- compare the year of the next record to the current to create split at year end
           CASE WHEN YEAR(LEAD(ELIGBEGIN,1) OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN)) = YEAR(ELIGEND)
                THEN LEAD(ELIGBEGIN,1) OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN)
           END AS nextELIGBEGIN
    FROM #t
    WHERE DEMONSTRATION_POPULATION = 'Adult Expansion'
)
,islandStartCTE
AS
(
    SELECT *,
           ROW_NUMBER() OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN) AS sequence_no
    FROM prevNextCTE
    WHERE prevELIGEND IS NULL
    OR DATEADD(DAY,1,prevELIGEND) < ELIGBEGIN
)
,islandEndCTE
AS
(
    SELECT *,
           ROW_NUMBER() OVER (PARTITION BY CLIENTID ORDER BY ELIGBEGIN) AS sequence_no
    FROM prevNextCTE
    WHERE nextELIGBEGIN IS NULL
    OR DATEADD(DAY,1,ELIGEND) < nextELIGBEGIN
)
SELECT iss.CLIENTID,
       iss.AGE,
       iss.ELIGBEGIN,
       ise.ELIGEND,
       iss.sequence_no
FROM islandStartCTE AS iss
-- left join here is not necessary but makes it easier to spot errors 
-- when a start position has no matching end (which should not occur)
LEFT
JOIN islandEndCTE AS ise 
ON ise.CLIENTID = iss.CLIENTID
AND ise.sequence_no = iss.sequence_no

为特定子群体创建连续注册孤岛

Create Continuous Enrollment islands for a certain sub-population

sql

sql-server

tsql

gaps-and-islands