SQL Conditional sum and grouping - sql

Question

I have a query that's trying to sum up a patient's length of stay at a hospital. Here is an example of the data
| Patient | Admission_ID | Admission_Event_ID | Admission_Event_Type | Start Date | End Date | Duration | Linked_Admission |
| P0001 | ADM0001 | AE1 | (formal) Separation | 2012-12-18 | 2012-12-18 | 0 | ADM0002 |
| P0001 | ADM0001 | AE2 | Statistical Admission | 2012-12-17 | 2012-12-18 | 1 | ADM0002 |
| P0001 | ADM0002 | AE3 | Statistical Separation| 2012-12-17 | 2012-12-17 | 0 | NULL |
| P0001 | ADM0002 | AE4 | (formal) Admission | 2012-11-30 | 2012-12-17 | 17 | NULL |
| P0002 | ADM0003 | AE5 | (formal) Admission | 2012-11-30 | 2012-12-25 | 25 | NULL |
. . .
EDIT: Forgot to mention, there is a column that links the admission ID (only used when the patient is statistically separated and admitted)
By definition, the length of stay is calculated for each patient from the start of their admission until they are separation (statistical separations and admission carry on with the admission, but they're given a new Admission ID
A report is run to find out the average length of stay (ALOS) for the hospital and it's unit, the user selects two dates to report between. I've used a CTE (lets call it CTESep) to get all the patient's that have been formally separated between the reporting period. I then use another CTE (called CTEAdmissions) to get all the admissions of the patients within CTESep. This is where I get stuck.
I need to sum up the Durations of the patient to get their total length of stay for that admission (which is a combination of ADM0001 and ADM0002) so the total LOS will be 18, rather than 17 and 1.
My idea was to
ORDER BY Patient
, End_Date DESC
, adm_id
, CASE WHEN
Admission_Event_Type = '(formal) Separation ' THEN 1
WHEN Admission_Event_Type = 'Statistical Admission ' THEN 2
WHEN Admission_Event_Type = 'Statistical Separation' THEN 3
WHEN Admission_Event_Type = '(formal) Admission ' THEN 4
END ASC
Then sum up the duration on based on a condition. The condition rule is 'Start summing up the duration of each patient's admission from a formal separation to a formal admission'. Which I'm not sure how to do.
I've tried:
SELECT SUM(Duration) OVER(PARTITION BY Patient) AS 'Sum'
But that will give me the total LOS for the patient across ALL their admissions (if they have more than one separation within that reporting period)
I've also tried
SELECT SUM(Duration) OVER(PARTITION BY Patient, Admission_ID) AS 'Sum'
But of course that gives me the LOS of a patient between a formal admission and a statistical separation (and not the LOS by its actual definition).
Anyone got a different way of tackling this problem? By the way, using Sybase

Answer 1

How about this:
select patientid,
admissionid,
datediff(day,
max(case when Admission_Event_Type = '(formal) Separation ' then startdate end),
max(case when Admission_Event_Type = '(formal) Admission ' then enddate end)
) as total_length
from data
group by patientid, admissionid