SSRS 2008 R2 - evaluating running total only on change of group - sql-server

Question

I have a report where I capture patient information, some of which is stored in the patient table and some of which is stored in the observations table. Taking date of birth as my example, if I count all the records for which the DOB has been supplied, I get significantly more than the total number of patients, because of the join to the observations table. How do I evaluate the running total only once for each group?
Edit: some sample data over at http://sqlfiddle.com/#!3/27b91/1/0. If I count birthdates from that query, I want 2 as the answer; same for race and ethnicity.

Answer 1

The following may or may not be the right approach for your specific situation, but it can be a useful technique to have at your disposal.
You can add some code to your select statement to help yourself answer questions like these 'downstream' (either via added criteria or via SSRS). See this modification of your SQL Fiddle:
select pid, firstName, lastName, dateOfBirth, obsName, obsValue, obsDate,
rowRank, CASE rowRank WHEN 1 THEN 1 ELSE 0 END AS countableRow
from
(
select Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth
, Obs.obsName, Obs.obsValue, Obs.obsDate,
ROW_NUMBER() OVER (PARTITION BY Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth ORDER BY Obs.obsDate) AS rowRank
from Person
join Obs on Person.pId = Obs.pId
) rankedData
The rowRank field will create a group-relative ranking number, which may or may not be useful to you downstream. The countableRow field will be either 1 or 0 such that each group will have one and only one row with a 1 in it. Doing SUM(countableRow) will give you the proper number of groups in your data.
Now, you can extend this functionality (if you wish) by dumping out actual field values instead of a constant scalar like 1 in the first row of each group. So, if you had CASE rowRank WHEN 1 THEN dateOfBirth ELSE NULL END AS countableDOB, you could then, for example, get the total number of people with each distinct birthday using just this dataset.
Of course, you can do all those things using methods like #Russell's with SQL anyway, so this would be most relevant with specific downstream requirements that may not match your situation.
EDIT
Obviously the countableRow field there isn't a one-size-fits-all solution to the types of queries you want. I have added a few more examples of the PARTITION BY strategy to another SQL Fiddle:
select pid, firstName, lastName, dateOfBirth, obsName, obsValue, obsDate,
rowRank, CASE rowRank WHEN 1 THEN 1 ELSE 0 END AS countableRow,
valueRank, CASE valueRank WHEN 1 THEN 1 ELSE 0 END AS valueCount,
dobRank, CASE WHEN dobRank = 1 AND dateOfBirth IS NOT NULL THEN 1 ELSE 0 END AS dobCount
from
(
select Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth
, Obs.obsName, Obs.obsValue, Obs.obsDate,
ROW_NUMBER() OVER (PARTITION BY Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth ORDER BY Obs.obsDate) AS rowRank,
ROW_NUMBER() OVER (PARTITION BY Obs.obsName, Obs.obsValue ORDER BY Obs.obsDate) AS valueRank,
ROW_Number() OVER (PARTITION BY Person.dateOfBirth ORDER BY Person.pid) AS dobRank
from Person
join Obs on Person.pId = Obs.pId
) rankedData
Lest anyone misunderstand me as suggesting this is always appropriate, it obviously isn't. This isn't a better solution to getting specific answers using additional SQL queries. What it allows you to do is encode enough information to simply answer such questions in the consuming code all in a single result set. That's where it can come in handy.
SECOND EDIT
Since you were wondering whether you can do this if race data is stored in more than one place, the answer is, absolutely. I have revised the code from my previous SQL Fiddle, which is now available in a new one:
select pid, firstName, lastName, dateOfBirth, obsName, obsValue, obsDate,
rowRank, CASE rowRank WHEN 1 THEN 1 ELSE 0 END AS countableRow,
valueRank, CASE valueRank WHEN 1 THEN 1 ELSE 0 END AS valueCount,
dobRank, CASE WHEN dobRank = 1 AND dateOfBirth IS NOT NULL THEN 1 ELSE 0 END AS dobCount,
raceRank, CASE WHEN raceRank = 1 AND (race IS NOT NULL OR obsName = 'RACE') THEN 1 ELSE 0 END AS raceCount
from
(
select Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth, Person.[race]
, Obs.obsName, Obs.obsValue, Obs.obsDate,
ROW_NUMBER() OVER (PARTITION BY Person.pid, Person.firstName, Person.lastName, Person.dateOfBirth ORDER BY Obs.obsDate) AS rowRank,
ROW_NUMBER() OVER (PARTITION BY Obs.obsName, Obs.obsValue ORDER BY Obs.obsDate) AS valueRank,
ROW_NUMBER() OVER (PARTITION BY Person.dateOfBirth ORDER BY Person.pid) AS dobRank,
ROW_NUMBER() OVER (PARTITION BY ISNULL(Person.race, CASE Obs.obsName WHEN 'RACE' THEN Obs.obsValue ELSE NULL END) ORDER BY Person.pid) AS raceRank
from Person
left join Obs on Person.pId = Obs.pId
) rankedData
As you can see, in the new Fiddle, this properly counts the number of Races as 3, with 2 being in the Obs table and the third being in the Person table. The trick is that PARTITION BY can contain expressions, not just raw column output. Note that I changed the join to a left join here, and that we need to use a CASE to only include obsValue WHERE obsName is 'RACE'. It is a little complicated, but not overwhelmingly so, and it handles even fairly complex cases gracefully.

Answer 2

It turned out that Jeroen's pointer to RunningValue was more on-target than I thought. I was able to get the results I wanted with the following code:
=RunningValue(Iif(Not IsNothing(Fields!DATEOFBIRTH.Value)
, Fields!PATIENTID.Value
, Nothing)
, CountDistinct
, Nothing
)
Thanks particularly to Dominic P, whose technique I'll keep in mind for next time.

Answer 3

This will only pull one record per patient, unless they reported different DOBs:
SELECT P.FOO,
P.BAR,
(etc.),
O.DOB
FROM Patients P
INNER JOIN Observations O
ON P.PatientID = O.PatientID
GROUP BY P.FOO, P.BAR, (P.etc), O.DOB