Distinct values based on correlated MIN() and MAX() values - sql

Question

I have a table called EVENTS and I need to write 4 scripts to generate 4 separate tables of distinct EVENT_ID values. Below is the first script and the other 3 will be of a similar logic. Can anyone help with this first script please so I can then hopefully use it as a template for the other 3 scripts. I am writing these scripts in SQL2005 which must be backwards compatible with SQL2000. I have removed any duplicates so there shouldn't be a need to involve the rank of the EVENT_ID in the logic.
For each CARE_ID select the value of EVENT_ID which has an EVENT_TYPE
of CP and has a MAX(EVENT_DATE) which is <= the MIN(EVENT_DATE) where
the EVENT_TYPE is in ('B','CH','S', 'T')
CARE_ID EVENT_ID EVENT_DATE EVENT_TYPE
3 194 01/10/2012 S
3 228 07/07/2010 S
3 104 12/05/2010 CH
3 16 12/07/2010 B
3 17 13/07/2010 B
3 43 15/01/2010 P
3 189 15/04/2010 S
39 45 09/10/2009 T
39 4 21/07/2009 P
39 6 21/07/2009 CH
78 28 08/07/2009 S
78 706 08/12/2010 CP
78 707 09/12/2010 CP
78 9 28/07/2009 T
78 11 28/07/2009 CH
95 21 31/07/2009 CH
95 21 31/07/2009 T
107 1474 21/09/2012 S
107 93 23/02/2010 CP
107 59 29/10/2012 P
107 58 29/12/2009 P
151 186 19/03/2010 S
151 49 21/03/2010 T
152 69 26/08/2009 CH
206 85 21/08/2009 CP
206 84 28/07/2009 CP
217 158 18/02/2010 S
217 102 30/03/2010 CH
218 159 12/03/2010 S
227 1378 01/04/2011 CP
355 19 13/07/2010 B
355 20 13/07/2010 B
355 239 13/07/2010 S
355 56 16/07/2010 T
355 111 16/07/2010 CH
364 1136 18/02/2011 CP
364 569 19/02/2011 S
364 774 23/08/2012 CH
364 1122 26/01/2011 CP
367 247 01/07/2010 S
367 151 21/06/2010 CP
369 108 26/07/2010 P
369 152 27/07/2010 CP
369 109 28/07/2010 P
369 117 28/07/2010 CH
369 248 28/07/2010 S
380 277 08/07/2011 T
396 1573 06/06/2011 CP
481 63 07/09/2010 T
481 116 07/09/2010 P
481 194 07/09/2010 CP
481 289 07/09/2010 S
502 200 13/08/2010 CP
530 220 14/06/2010 CP
535 222 05/07/2010 CP
535 303 13/07/2010 S
535 223 19/07/2010 CP
535 224 26/07/2010 CP
536 135 10/09/2010 CH
536 225 23/08/2010 CP
568 155 06/10/2010 P
568 315 15/10/2010 S
631 148 02/02/2010 CH
631 74 15/01/2010 T
631 256 15/12/2009 CP
631 345 15/12/2009 S
631 147 25/12/2009 CH
632 259 18/09/2010 CP
653 189 29/10/2010 P
653 360 30/09/2010 S
655 1570 06/06/2011 CP
680 569 08/12/2010 CP
680 1191 24/11/2011 S
680 530 25/01/2011 S
680 151 30/09/2010 P
680 281 30/09/2010 CP
680 480 30/11/2010 CP
689 306 02/11/2010 CP
689 158 06/10/2010 P
689 372 06/10/2010 S
689 2720 06/11/2012 CP
689 2736 11/11/2012 CP
689 2752 13/11/2012 CP
689 2765 15/11/2012 CP
689 2125 22/09/2011 CP
689 2654 24/09/2012 CP
689 1944 26/08/2011 CP
689 307 26/10/2010 CP
689 1947 27/08/2011 CP
729 299 15/09/2010 CP
811 413 27/10/2010 S
834 622 01/01/2012 CH
834 1233 06/01/2012 S
834 624 15/01/2012 CH
834 625 23/01/2012 CH
834 627 23/01/2012 CH
838 629 02/01/2012 CH
838 630 20/01/2012 CH
838 632 27/01/2012 CH
846 416 05/10/2010 S
849 195 03/11/2010 P
849 336 21/02/2011 CP
923 441 26/07/2010 S
963 371 29/10/2010 CP
981 624 23/03/2011 S
984 384 13/11/2010 CP
984 392 18/11/2010 CP

Answer 1

Tried using a HAVING clause? Maybe I've messed something up, but your test data only seems to have one such entry, namely for the S case, as follows:
SELECT e.[CARE_ID], e.[EVENT_ID]
FROM dbo.EVENTS e
WHERE e.[EVENT_TYPE] = 'CP'
GROUP BY e.[CARE_ID], e.[EVENT_ID]
HAVING MAX( e.[EVENT_DATE] ) <= ( SELECT MIN( [EVENT_DATE] )
FROM [EVENTS]
WHERE [EVENT_ID] = e.[EVENT_ID]
AND [EVENT_TYPE] = 'S' );
Here's an SQL Fiddle to help sort you out!
Edit: The original fiddle is looking for the maximum date of the EVENT_ID, when we're supposed to be looking for the maximum date of the CARE_ID. I think this will get you on the right track!
SELECT e.[CARE_ID], e.[EVENT_ID]
FROM dbo.EVENTS e
WHERE e.[EVENT_TYPE] = 'CP'
GROUP BY e.[CARE_ID], e.[EVENT_ID]
HAVING MAX( e.[EVENT_DATE] ) <= ( SELECT MIN( [EVENT_DATE] )
FROM [EVENTS]
WHERE [CARE_ID] = e.[CARE_ID]
AND [EVENT_TYPE] = 'S' );
Edit 3: Now with proper DATETIME!
Edit 4: Distinct EVENT_ID!
SELECT DISTINCT e.[EVENT_ID]
FROM dbo.EVENTS e
WHERE e.[EVENT_TYPE] = 'CP'
GROUP BY e.[CARE_ID], e.[EVENT_ID]
HAVING MAX( e.[EVENT_DATE] ) <= ( SELECT MIN( [EVENT_DATE] )
FROM dbo.[EVENTS]
WHERE [CARE_ID] = e.[CARE_ID]
AND [EVENT_TYPE] = 'S' );

Answer 2

This translate quite directly into SQL using window functions:
select care_id, event_id
from (select e.*,
max(case when event_type = 'CP' then event_date end) over (partition by care_id) as MaxED_CP,
min(case when event_type = ('B','CH','S', 'T') then event_date end) over (partition by care_id) as MinED_others,
from events e
) e
where event_type = 'CP' and
MaxED_CP <= MinED_others and
event_date = MaxED_CP;
These functions calculate the max() and min() values on every row. The outer query just selects the appropriate rows.
Note that if no such event exists for a given care_id then the care_id is not in the output.
This is not backwards compatible with SQL Server 2000. If it works, then you can replace the window functions with subqueries.