I wish I could correlate an "inline view" - sql

Question

I have a Patient table:
PatientId Admitted
--------- ---------------
1 d/m/yy hh:mm:ss
2 d/m/yy hh:mm:ss
3 d/m/yy hh:mm:ss
I have a PatientMeasurement table (0 to many):
PatientId MeasurementId Recorded Value
--------- ------------- --------------- -----
1 A d/h/yy hh:mm:ss 100
1 A d/h/yy hh:mm:ss 200
1 A d/h/yy hh:mm:ss 300
2 A d/h/yy hh:mm:ss 10
2 A d/h/yy hh:mm:ss 20
1 B d/h/yy hh:mm:ss 1
1 B d/h/yy hh:mm:ss 2
I am trying to create a result set that resembles:
PatientId Numerator Denominator
--------- -------- -----------
1 1 1
2 1 1
3 0 1
Essentially, a patient will have a 1 in the numerator if the have at least one value for measurement A and one value for measurement B. In this example, patient 1 has 3 A measurements and 2 B measures, so the numerator is 1. Patient 2 has 2 A measurements, but no B measurements, so the numerator is 0. Patient has neither an A measurement nor a B measurement, so the numerator is 0.
My query thus far is:
SELECT PatientId, CASE WHEN a.cnt+b.cnt>2 THEN 1 ELSE 0 END Numerator, 1 Denominator
FROM patient p
LEFT OUTER JOIN (
SELECT PatientId, count(*) cnt
FROM PatientMeasurement pm
WHERE MeasurementId='A'
--AND Recorded <= dateadd(hh, 12, Admitted)
GROUP BY PatientId
) a ON p.PatientId=a.PatientId
LEFT OUTER JOIN (
SELECT PatientId, count(*) cnt
FROM PatientMeasurement pm
WHERE MeasurementId='B'
--AND Recorded <= dateadd(hh, 12, Admitted)
GROUP BY PatientId
) b ON p.PatientId=b.PatientId
This works as expected as long as I don't include the correlated, date restriction (Recorded < dateadd(hh, 12, Admitted). Unfortunately, correlating an 'inline view' in this manner is not syntactically valid.
This has forced me to re-write the SQL to:
SELECT PatientId, CASE WHEN v.a+v.b>2 THEN 1 ELSE 0 END Numerator, 1 Denominator
FROM (
SELECT PatientId,
(
SELECT PatientId, count(*) cnt
FROM PatientMeasurement pm
WHERE PatientId=p.PatientId
AND MeasurementId='A'
AND Recorded <= dateadd(hh, 12, Admitted)
GROUP BY PatientId
) a,
(
SELECT PatientId, count(*) cnt
FROM PatientMeasurement pm
WHERE PatientId=p.PatientId
AND MeasurementId='B'
AND Recorded <= dateadd(hh, 12, Admitted)
GROUP BY PatientId
) b
FROM Patient p
) v
My question: Is there a better, more-efficient way to do this?
Thanks for your time.

Answer 1

Try this :
WITH GroupPatients AS
(SELECT MeasurementID, PatientId, Count(*) AS cnt
FROM PatientMeasurement AS pm
INNER JOIN Patient p ON pm.PatientID = p.PatientID
WHERE
MeasurementId IN ('A', 'B')
AND
Recorded <= dateadd(hh, 12, Admitted)
GROUP BY MeasureMentID, PatientId)
SELECT p.PatientID, Case
When IsNull(GPA.cnt, 0) > 0 AND IsNull(GPB.cnt, 0) > 0 Then 1
Else 0
End AS Numerator, 1 AS Denominator
FROM Patient p
LEFT JOIN GroupPatientsA AS GPA ON p.PatientID = GPA.PatientID AND GPA.MeasurementID = 'A'
LEFT JOIN GroupPatientsB AS GPB ON p.PatientID = GPB.PatientID AND GPB.MeasurementID = 'B'
I've made one tweak to the business logic too - your spec says Numerator should be one if a patient has both A and B measurements - however, your clause of a.cnt+b.cnt>2 will erroneously return one if either a.cnt or b.cnt are 3 or more and the other is zero.

Answer 2

Another solution can be close to your original attempt using OUTER APPLY:
SELECT PatientId, CASE WHEN a.cnt+b.cnt>2 THEN 1 ELSE 0 END Numerator, 1 Denominator
FROM patient p
OUTER APPLY (
SELECT count(*) cnt
FROM PatientMeasurement pm
WHERE MeasurementId='A'
AND Recorded <= dateadd(hh, 12, p.Admitted)
AND pm.PatientId = p.PatientId
) AS a(cnt)
OUTER APPLY (
SELECT count(*) cnt
FROM PatientMeasurement pm
WHERE MeasurementId='B'
AND Recorded <= dateadd(hh, 12, p.Admitted)
AND pm.PatientId = p.PatientId
) AS b(cnt)

Answer 3

SELECT p.*,
CASE WHEN
EXISTS
(
SELECT NULL
FROM PatientMeasurement pm
WHERE pm.PatientID = p.ID
AND pm.Type = 'A'
AND pm.Recorded <= DATEADD(hh, 12, p.Admitted)
) AND EXISTS (
SELECT NULL
FROM PatientMeasurement pm
WHERE pm.PatientID = p.ID
AND pm.Type = 'B'
AND pm.Recorded <= DATEADD(hh, 12, p.Admitted)
) THEN 1 ELSE 0 END
FROM Patient p

Answer 4

Assuming you are using Sql 2005 or 2008, the entire query can be simplified using some window functions and a pivot:
with pData as
(
select count(*) over(partition by PatientId, MeasurementId) as cnt,
PatientId, MeasurementId
from PatientMeasurement pm
where MeasurementId in('A','B')
and Recorded <= dateadd(hh, 12, Admitted)
)
select PatientId, coalesce([A],0) as cntA, coalesce([B],0) as cntB,
case when coalesce([A],0) + coalesce([B],0) > 2 then 1 else 0 end as Numerator,
1 as Denominator
from pData
pivot (max(cnt) for MeasurementId in([A],[B])) pvt

Answer 5

DECLARE #TimeSlot int;
SET #TimeSlot = 12;
WITH
pt AS (
SELECT p.PatientID, p.Admitted, m.MeasurementID, m.Recorded,
CASE
WHEN m.Recorded <= dateadd(hh, #TimeSlot, p.Admitted) THEN 1
ELSE 0
END AS "InTimeSlot"
FROM Patient AS p
LEFT JOIN PatientMeasurement AS m ON p.PatientID = m.PatientID
),
cntA AS (
SELECT PatientID, count(*) AS "A_count"
FROM pt WHERE MeasurementID='A' AND InTimeSlot = 1
GROUP BY PatientID
),
cntB AS (
SELECT PatientID, count(*) AS "B_count"
FROM pt WHERE MeasurementID='B' AND InTimeSlot = 1
GROUP BY PatientID
),
cntAB AS (
SELECT p.PatientID
,coalesce(a.A_count, 0) AS "A_cnt"
,coalesce(b.B_count, 0) AS "B_cnt"
FROM Patient as p
LEFT JOIN cntA AS a ON p.PatientID = a.PatientID
LEFT JOIN cntB AS b ON p.PatientID = b.PatientID
),
cntN AS (
SELECT PatientID,
CASE WHEN A_cnt > 0 AND B_cnt > 0 THEN 1 ELSE 0 END AS Numerator
FROM cntAB
)
SELECT PatientID, Numerator, 1 AS Denominator FROM cntN