SQL-based, FIFO logic - sql

Question

I need to build a query that will aggregate the # of units of blood that has been transfused so it can be compared to the # of units of blood that has been cross-matched. Blood (a precious resource) that is cross-matched, but not transfused is wasted.
Providers are supposed to check the system-of-record (Epic) for 'active' cross-match orders before creating new ones. Providers that don't do this are 'penalized' (provider 20). No penalty applies (it seems) for providers that don't transfuse all of the blood that they've cross-matched (provider 10).
Cross-match orders:
|ENC_ID|PROV_ID|ORDER_ID|ORDER_TIME |UNITS|
| 1| 10| 100|26-JUL-12 13:00| 4|
| 1| 20| 231|26-JUL-12 15:00| 2|
Transfusion orders:
|ENC_ID|PROV_ID|ORDER_ID|ORDER_TIME |UNITS|
| 1| 10| 500|26-JUL-12 13:05| 1|
| 1| 10| 501|26-JUL-12 13:25| 1|
| 1| 20| 501|26-JUL-12 15:00| 1|
| 1| 20| 501|26-JUL-12 15:21| 2|
Rules:
compare transfusions to cross-matches for same encounter (transfusion.END_ID=cross-match.ENC_ID)
transfusion orders are applied to cross-match orders in a FIFO manner
transfusion.ORDER_TIME >= cross-match.ORDER_TIME
a provider may transfuse more than their cross-match order, as long as all of the 'active' cross-match order still have available units (provider 20's second transfusion order)
Desired result:
|ENC_ID|PROV_ID|ORDER_ID|ORDER_TIME |CROSS-MATCHED|TRANSFUSED|
| 1| 10| 100|26-JUL-12 13:00| 4| 4|
| 1| 20| 231|26-JUL-12 15:00| 2| 1|
Provider 10 'credited' with Provider 20's transfusions.
Can this logic be implemented without resorting to a procedure?

Answer 1

You could do it in a single SQL query. Here's an example (tested on 11gR2, should work on 10g):
SETUP:
CREATE TABLE cross_match as (
SELECT 1 ENC_ID, 10 PROV_ID, 100 ORDER_ID,
to_date('2012-07-26 13', 'yyyy-mm-dd hh24') ORDER_TIME, 4 UNITS
FROM DUAL
UNION ALL SELECT 1, 20, 231, to_date('2012-07-26 15', 'yyyy-mm-dd hh24'), 2 FROM DUAL
);
CREATE TABLE transfusion as (
SELECT 1 ENC_ID, 10 PROV_ID, 500 ORDER_ID,
to_date('2012-07-26 13:05', 'yyyy-mm-dd hh24:mi') ORDER_TIME, 1 UNITS
FROM DUAL
UNION ALL SELECT 1, 10, 501, to_date('2012-07-26 13:25', 'yyyy-mm-dd hh24:mi'), 1 FROM DUAL
UNION ALL SELECT 1, 20, 501, to_date('2012-07-26 15:00', 'yyyy-mm-dd hh24:mi'), 1 FROM DUAL
UNION ALL SELECT 1, 20, 501, to_date('2012-07-26 15:21', 'yyyy-mm-dd hh24:mi'), 2 FROM DUAL
);
The following query will build a list of blood units numerically and join each unit from the cross_match table to the corresponding one (if it exists) in the transfusion table:
WITH cross_order as (
SELECT rownum rn FROM DUAL
CONNECT BY level <= (SELECT MAX(units) FROM cross_match)
),
transfusion_order as (
SELECT rownum rn FROM DUAL
CONNECT BY level <= (SELECT MAX(units) FROM transfusion)
)
SELECT c.enc_id, c.prov_id, c.order_id, c.order_time,
count(*) cross_matched,
count(t.enc_id) transfused
FROM (SELECT cm.*,
row_number() over (partition by cm.enc_id
order by cm.order_time) cross_no
FROM cross_match cm
JOIN cross_order co ON cm.units >= co.rn) c
LEFT JOIN (SELECT t.*,
row_number() over (partition by t.enc_id
order by t.order_time) trans_no
FROM transfusion t
JOIN transfusion_order tor ON t.units >= tor.rn) t
ON c.enc_id = t.enc_id
AND c.cross_no = t.trans_no
GROUP BY c.enc_id, c.prov_id, c.order_id, c.order_time;
ENC_ID PROV_ID ORDER_ID ORDER_TIME CROSS_MATCHED TRANSFUSED
-----------------------------------------------------------
1 20 231 07/26/2012 2 1
1 10 100 07/26/2012 4 4
This may be efficient if the maximum number of units remains small, otherwise this 1-to-1 relationship may become cumbersome.
This can be improved by using a running total of units on both sides instead of a basic 1-1. The join condition would be like an interval intersection between begin unit and end unit:
SELECT c.enc_id, c.prov_id, c.order_id, c.order_time,
sum(c.unit_end - nvl(c.unit_start,0))/count(*) cross_matched,
sum(least(c.unit_end, t.unit_end)
-greatest(nvl(c.unit_start, 0), nvl(t.unit_start, 0))) transfused
FROM (SELECT cm.*,
sum(cm.units) over (partition by cm.enc_id
order by cm.order_time
rows between unbounded preceding
and 1 preceding) unit_start,
sum(cm.units) over (partition by cm.enc_id
order by cm.order_time) unit_end
FROM cross_match cm) c
LEFT JOIN (SELECT t.*,
sum(t.units) over (partition by t.enc_id
order by t.order_time
rows between unbounded preceding
and 1 preceding) unit_start,
sum(t.units) over (partition by t.enc_id
order by t.order_time) unit_end
FROM transfusion t) t
ON c.enc_id = t.enc_id
AND c.unit_end > nvl(t.unit_start, 0)
AND t.unit_end > nvl(c.unit_start, 0)
GROUP BY c.enc_id, c.prov_id, c.order_id, c.order_time;
ENC_ID PROV_ID ORDER_ID ORDER_TIME CROSS_MATCHED TRANSFUSED
-----------------------------------------------------------
1 20 231 07/26/2012 2 1
1 10 100 07/26/2012 4 4