如何将此T-SQL游标操作转换为基于集合的操作

Question

我有一个医疗报告项目,需要从诊断日期列表中计算"开始"日期.

患者第一次出现某种诊断,即发病日期为#1.在接下来的30天内,任何后续诊断日期都计入该发病次数的一部分,因此可以删除后续日期.

30天过后,下一个诊断日期计为起始#2.

等等..

示例输入

2015-09-10 (this is onset #1)
2015-09-19 (within 30 days from onset #1, so drop)
2015-09-29 (within 30 days from onset #1, so drop)
2015-10-17 (>= 30 days from onset #1, so this is onset #2)
2015-10-19 (within 30 days from onset #2, so drop)
2015-11-13 (within 30 days from onset #2, so drop)
2015-11-29 (>= 30 days from onset #2, so this is onset #3)

示例输出

2015-09-10 (onset #1)
2015-10-17 (onset #2)
2015-11-29 (onset #3)

这可以用光标完成,下面包含一个最小的例子.

我听说它说任何游标操作都可以表示为基于集合的操作.但我无法弄清楚如何以基于集合的方式处理这种特定算法,因为计算依赖于以前的算法.我看不出如何在一个基于集合的"通行证"中完成它.

可以吗？如果是这样,怎么样？

任何解决方案都应该在SQL Server 2012中运行.

DECLARE @dx_list TABLE(dx_dt date);
INSERT INTO @dx_list(dx_dt)
    VALUES ('2015-09-10') --this is onset #1
    , ('2015-09-19') 
    , ('2015-09-29')
    , ('2015-10-17') --date is >= 30 days from last onset, so this is onset #2
    , ('2015-10-19')
    , ('2015-11-13')
    , ('2015-11-29'); --date is >= 30 days from last onset, so this is onset #3

DECLARE @mycursor AS cursor;
SET @mycursor = CURSOR FOR
SELECT dx_dt
FROM @dx_list
ORDER BY dx_dt; --make sure dates are in order

DECLARE @possible_dt AS date;
DECLARE @onset_list TABLE(onset_dt date);

OPEN @mycursor;
FETCH NEXT FROM @mycursor INTO @possible_dt;
--First date is always an onset date
INSERT INTO @onset_list(onset_dt) VALUES (@possible_dt);

WHILE @@FETCH_STATUS = 0
BEGIN
    FETCH NEXT FROM @mycursor INTO @possible_dt;
    --If this date is 30 days or more from last onset date, add it
    IF @possible_dt >= DATEADD(dd, 30, (SELECT MAX(onset_dt) FROM @onset_list))
    BEGIN
        INSERT INTO @onset_list(onset_dt) VALUES (@possible_dt);
    END
END

CLOSE @mycursor;
DEALLOCATE @mycursor;

--Show results
SELECT * FROM @onset_list;

Answer 1

你可以使用递归CTE来做到这一点:

;WITH OnsetDates AS (
   SELECT TOP 1 dx_dt
   FROM dx_list
   ORDER BY dx_dt

   UNION ALL

   SELECT dx_dt
   FROM (  
     SELECT d1.dx_dt,
            ROW_NUMBER() OVER (ORDER BY d1.dx_dt) AS rn
     FROM dx_list AS d1
     INNER JOIN OnsetDates AS d2 ON d1.dx_dt > d2.dx_dt
     WHERE DATEDIFF(d, d2.dx_dt, d1.dx_dt) >= 30 ) AS t
   WHERE t.rn = 1
)
SELECT *
FROM OnsetDates

所谓固定件的CTE只是顶层的日期.该递归成员取出下一发病日期:这是第一次约会是过去30天+由递归CTE的前一个调用返回的日期.

请注意,为了获得第一个日期,我们必须使用ROW_NUMBER和子查询,因为TOP 1并且ORDER BY不允许在CTE的递归成员中.

在这里演示