Select First, Max, and Last non-null value per group - sql-server

Question

Trying to select, per group, the first and last values (chronologically) as well as the max value. I had written a query that works fine except it does not handle the NULL values. I need it to ignore NULL values.
Here's an example:
DECLARE #T table (
LabName VARCHAR(20)
, CreatedOn date
, LabValue int
)
INSERT INTO #T
( LabName,CreatedOn,LabValue )
VALUES
('Creatinine', '2016-01-01', NULL)
, ('Creatinine', '2016-02-01', 15)
, ('Creatinine', '2016-03-01', 20)
, ('Creatinine', '2016-04-01', 19)
, ('SGOT (ST)', '2016-01-01', 25)
, ('SGOT (ST)', '2016-02-01', 31)
, ('SGOT (ST)', '2016-03-01', 25)
, ('SGOT (ST)', '2016-04-01', NULL)
SELECT DISTINCT
*
FROM (
SELECT
LabName
, FIRST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC) AS FirstValue
, MAX(LabValue) OVER(PARTITION BY LabName) AS MaxValue
, LAST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) LastValue
FROM #T
) AS T
It was working fine until I realized some labs aren't run on some dates. Once I put some NULLs into the test data, the results for First and Last will include them.
Here is the result I get:
+------------+------------+----------+-----------+
| LabName | FirstValue | MaxValue | LastValue |
+------------+------------+----------+-----------+
| Creatinine | NULL | 20 | 19 |
| SGOT (ST) | 25 | 31 | NULL |
+------------+------------+----------+-----------+
Here is the result I want:
+------------+------------+----------+-----------+
| LabName | FirstValue | MaxValue | LastValue |
+------------+------------+----------+-----------+
| Creatinine | 15 | 20 | 19 |
| SGOT (ST) | 25 | 31 | 25 |
+------------+------------+----------+-----------+

Answer 1

Use conditional aggregation with ROW_NUMBER():
SELECT LabName,
MAX(CASE WHEN seqnum_asc = 1 THEN LabValue END) as FirstValue,
MAX(LabValue) as MaxValue,
MAX(CASE WHEN seqnum_desc = 1 THEN LabValue END) as LastValue
FROM (SELECT t.*,
ROW_NUMBER() OVER (PARTITION BY LabName
ORDER BY (CASE WHEN LabValue IS NOT NULL THEN 1 ELSE 2 END),
CreatedOn
) as seqnum_asc,
ROW_NUMBER() OVER (PARTITION BY LabName
ORDER BY (CASE WHEN LabValue IS NOT NULL THEN 1 ELSE 2 END),
CreatedOn DESC
) as seqnum_desc
FROM #T t
) T
GROUP BY LabName;

Answer 2

As you said there are 13 such columns where you need to check not null values.
I think you should first filter all not null values using CTE,then using CTE you can write your actual query.CTE will reduce your result set and applying window function on reduce resultset will give better performance.
BTW,13 such columns appear t be bad DB design.you may have to 100 query in future.
IMHO, DISTINCT often indicate bad DB design than query.
;With CTE as
(-- try to reduce resultset if possible
SELECT * FROM #T
WHERE LabValue IS NOT NULL
)
SELECT DISTINCT
*
FROM (
SELECT
LabName
, FIRST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC) AS FirstValue
, MAX(LabValue) OVER(PARTITION BY LabName) AS MaxValue
, LAST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) LastValue
FROM CTE
) AS T

Answer 3

Your database is handling NULL values properly.
First value for Creatinine is actually null and last value for SGOT (ST) is null as well.
If you wish to discard rows with null values just add it in the WHERE clause:
SELECT DISTINCT
*
FROM (
SELECT
LabName
, FIRST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC) AS FirstValue
, MAX(LabValue) OVER(PARTITION BY LabName) AS MaxValue
, LAST_VALUE(LabValue) OVER(PARTITION BY LabName ORDER BY CreatedOn ASC ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING) LastValue
FROM #T
WHERE LabValue IS NOT NULL
) AS T;