Counting rows of a subgroup while ignoring duplicates - sql

Question

I can't find a way to describe my problem in an abstract and general manner, so I'll just provide a minimal example:
Let's say I have these 3 simple tables:
CREATE TABLE Document(
[Id] int IDENTITY(1, 1) NOT NULL PRIMARY KEY,
[Title] nvarchar(MAX),
[Patient] nvarchar(MAX)
);
CREATE TABLE Link(
DocumentId INT FOREIGN KEY REFERENCES Document(Id),
Text nvarchar(max)
);
CREATE TABLE ReadStatus(
DocumentId INT FOREIGN KEY REFERENCES Document(Id),
IsRead Bit NOT NULL,
UserId Int NOT NULL
);
We have a set of documents
A document can have 0 or more links
Documents can be read by users - this is tracked by the ReadStatus table, which associates a user with a document, and where IsRead=1 means the document has been read by that user and IsRead=0 means it hasn't been read by that user yet.
If, for document X and user A, a row does not exist in the ReadStatus table, we assume User A hasn't read document X yet.
Now, I need to run a query to select all patients. For each patient, I need the total number of documents available AND the number of documents that have already been read (i.e. IsRead=1). This is what I have so far:
SELECT d.Patient,
COUNT(DISTINCT d.Id) AS DocumentCount,
COUNT(NULLIF(rs.IsRead,0)) AS ReadDocumentCount,
COUNT(*) OVER () AS TotalPatientCount
FROM Document d
LEFT OUTER JOIN ReadStatus AS rs ON d.Id = rs.DocumentId AND rs.UserId = 123
INNER JOIN Link AS l ON d.Id = l.DocumentId AND l.Text IN ('Link W', 'Link X', 'Link T', 'Link Z')
GROUP BY d.Patient
The problem happens when a document (that has already been read) has more than one link. If that document has 3 links, the cartesian product produced by the INNER JOIN with the Link table will cause the ReadDocumentCount selection to be 3 instead of 1.
In other words, given this data:
INSERT INTO Document(Title, Patient) VALUES('Doc A', 'Mike')
INSERT INTO Document(Title, Patient) VALUES('Doc B', 'Mike')
INSERT INTO Link(DocumentId, Text) VALUES(1, N'Link W')
INSERT INTO Link(DocumentId, Text) VALUES(1, N'Link X')
INSERT INTO Link(DocumentId, Text) VALUES(1, N'Link Y')
INSERT INTO Link(DocumentId, Text) VALUES(2, N'Link Z')
INSERT INTO ReadStatus(DocumentID, IsRead, UserId) VALUES(1, 1, 123)
INSERT INTO ReadStatus(DocumentID, IsRead, UserId) VALUES(2, 0, 123)
I'm getting this as a result:
Patient DocumentCount ReadDocumentCount TotalPatientCount
Mike 2 3 1
Whereas this is what I want:
Patient DocumentCount ReadDocumentCount TotalPatientCount
Mike 2 1 1
SQL fiddle: http://sqlfiddle.com/#!6/e06bf/3

Answer 1

You can use COUNT(DISTINCT) conditionally as well:
SELECT d.Patient,
COUNT(DISTINCT d.Id) AS DocumentCount,
COUNT(DISTINCT (CASE WHEN rs.IsRead <> 0 THEN d.id END)) AS ReadDocumentCount,
COUNT(*) OVER () AS TotalPatientCount
FROM Document d LEFT OUTER JOIN
ReadStatus rs
ON d.Id = rs.DocumentId AND rs.UserId = 123 INNER JOIN
Link l
ON d.Id = l.DocumentId AND l.Text IN ('Link W', 'Link X', 'Link T', 'Link Z')
GROUP BY d.Patient;