SQL: determine that an activity occurs with a given frequency - sql

Question

A common problem in electronic-medical-record (EMR) reporting in determining that an activity occurs with a specific frequency. In this situation, I need to determine that a note was written every 72-hours after admission.
Given:
A D
|-0-|-1-|-2-|-3-|-4-|-5-|-6-|-7-|-8-|-9-|
|---- 1 ----|---- 2 ----|---- 3 ----|-4-|
There would need to be at least one note during periods 1, 2, and 3. Because 4 isn't a full 72-hour period, it doesn't require a note. Failure to find a note in periods 1, 2, and 3 would be a FAIL.
Data:
(ENC):
ENC_ID ADMITTED DISCHARGED PERIODS PASS_FAIL
4114221 06/15/09 18:30 06/24/09 15:40 3 ?
PERIODS: TRUNC(CEIL((DISCHARGED - ADMITTED)/3))
The 'PASS_FAIL' column would indicate if the encounter had an adequate number and timing of notes.
(NOTE):
ENC_ID NOTE_ID NOTE_TIME PERIOD
4114221 1833764 06/17/09 08:42 1
4114221 1843613 06/18/09 08:14 1
4114221 1858159 06/18/09 20:15 2
4114221 1850948 06/18/09 20:15 2
4114221 1850912 06/18/09 20:18 2
4114221 1859315 06/19/09 18:35 2
4114221 1863982 06/20/09 10:29 2
4114221 1868895 06/21/09 22:00 3
4114221 1873539 06/22/09 15:42 3
PERIOD: CEIL((NOTE_TIME - ADMITTED)/3)
Is there an efficient way to solve this problem?

Answer 1

SELECT e.*,
CASE WHEN cnt = TRUNC(CEIL((discharged / admitted) / 3)) THEN 'pass' ELSE 'fail' END AS pass_fail
FROM (
SELECT COUNT(*) AS cnt
FROM enc ei
CROSS JOIN
(
SELECT level AS period
FROM dual
CONNECT BY
level <=
(
SELECT TRUNC(CEIL((discharged / admitted) / 3))
FROM enc
WHERE enc_id = :enc_id
)
) p
WHERE ei.enc_id = :enc_id
AND EXISTS
(
SELECT NULL
FROM note
WHERE enc_id = ei.enc_id
AND note_time >= ei.admitted + (p - 1) * 3
AND note_time < ei.admitted + p * 3
)
) c
JOIN enc e
ON e.enc_id = :enc_id

Answer 2

If I'm reading your question correctly NOTE is a table with the data indicated.
All you really care about is whether the periods 1, 2 & 3 exist in the notes table for each enc_id.
If this is the case it indicates that an analytic function should be used:
select e.enc_id, e.admitted, e.discharged, e.periods
, decode( n.ct
, 'pass'
, 'fail' ) as pass_fail
from enc e
left outer join ( select distinct enc_id
, count(n.period) over ( partition by n.enc_id ) as ct
from note
where period in (1,2,3)
) n
on e.enc_id = n.enc_id
This selects all period's per enc_id from note, which are the ones you want to examine. Then counts them per enc_id. The distinct is there to ensure you only get one row per enc_id in the final result.
If you only want those enc_ids that have a value in note then turn the left outer join into an inner join.
If period is not, as indicated, in the note query, you have to do a distinct on the full query rather than the sub-query and check which period each note_id is in.
I'm sorry about the horrible formatting but I wanted to try to fit it on the page.
select distinct e.enc_id, e.admitted, e.discharged, e.periods
, decode( count( distinct -- number of distinct periods
case when n.note_time between e.admitted
and e.admitted + 3 then 1
when n.note_time between e.admitted
and e.admitted + 6 then 2
when n.note_time between e.admitted
and e.admitted + 9 then 3
end ) -- per enc_id from note
over ( partition by n.enc_id )
-- if it-s 3 then pass
, 3, 'pass'
-- else fail.
, 'fail' ) as pass_fail
from enc e
left outer join note n
on e.enc_id = n.enc_id
Whatever your data-structure the benefits of both ways are that they are simple joins, one index unique scan ( I'm assuming enc.end_id is unique ) and one index range scan ( on note ).