Detecting duplicates which fall outside of a date interval - sql

Question

I searched in SO but couldnt find a direct answer.
There are patients, hospitals, medical branches(ER,urology,orthopedics,internal disease etc), medical operation codes (examination,surgical operation, MRI, ultrasound or sth. else) and patient visiting dates.
Patient visits doctor, doctor prescribes medicine and asks to come again for control check.
If patient returns after 10 days, (s)he has to pay another examination fee to the same hospital. Hospitals may appoint a date after 10 days telling there are no available slots in following 10 days, in order to get the examination fee.
Table structure is like:
Patient id.no Hospital Medical Branch Medical Op. Code Date
1 H1 M0 P1 01/05/2011
5 H1 M1 P9 03/05/2011
3 H2 M0 P2 09/05/2011
1 H1 M0 P1 14/05/2011
3 H1 M0 P2 20/05/2011
5 H1 M2 P9 25/05/2011
1 H1 M0 P3 26/05/2011
Here, visiting patients no. 3 and 5 does not constitute a problem as patient no. 3 visits different hospitals and patient no.5 visits different medical branches. They would pay the examination fee even if they visited within 10 days.
Patient no.1, however, visits same hospital, same branch and is subject to same process (P1: examination) on 01/05 and 14/05.
26/05 doesnt count because it is not medical examination.
What I want to flag is same patient, same hospital, same branch and same medical operation code (that is specifically medical examination : P1 ), with date range more than 10 days.
The format of resulting table:
HOSPITAL TOTAL NUM. of PATIENTS NUM. of PATIENTS OUT OF DATE RANGE
H1 x a
H2 y b
H3 z c
Thanks.

Answer 1

Once again, it's analytic functions to the rescue.
This query uses the LAG() function to link a record in YOUR_TABLE with the previous (defined by DATE) matching record (defined by PATIENT_ID) in the table.
select hospital_id
, count(*) as total_num_of_patients
, sum (out_of_range) as num_of_patients_out_of_range
from (
select patient_id
, hospital_id
, case
when hospital_id_1 = hospital_id_0
and visit_1 > visit_0 + 10
and med_op_code_1 = med_op_code_0
then 1
else 0
end as out_of_range
from (
select patient_id
, hospital_id as hospital_id_1
, date as visit_1
, med_op_code as med_op_code_1
, lag (date) over (partition by patient_id order by date) as visit_0
, lag (hopital_id) over (partition by patient_id order by date) as hopital_id_0
, lag (med_op_code) over (partition by patient_id order by date) as med_op_code_0
from your_table
where med_op_code = 'P1'
)
)
group by hospital_id
/
Caveat: I haven't tested this code, so it may contain syntax errors. I will check it the next time I can access an Oracle database.

Answer 2

This is a little rough, as I haven't got an Oracle DB to hand, but the key feature is the same: the analytical function LAG(). Along with its companion function, LEAD(), they're great for helping to deal with things like periods of activity.
Here's my attempt at the code:
select n.hospital, COUNT(n.patient_id) as patients_out_of_date_range
from (
select *
from (
select d.*, lag(date, 1) over (partition by d.patient_id, d.hospital, d.medical_branch, d.medical_op_code order by d.date) as prev_date
from datatable d inner join
(
select d.patient_id, d.hospital, d.medical_branch, d.medical_op_code
from datatable d
where d.medical_op_code = 'P1'
group by d.patient_id, d.hospital, d.medical_branch, d.medical_op_code
having COUNT(d.date) > 1
) t on d.patient_id = t.patient_id and d.hospital = t.hospital and d.medical_branch = t.medical_branch and d.medical_op_code = t.medical_op_code
) m
where date - prev_date > 10
) n
group by n.hospital
Like I say, this isn't tested, but it should at least get you started in the right direction.
Some references:
http://www.adp-gmbh.ch/ora/sql/analytical/lag.html
http://www.oracle-base.com/articles/misc/LagLeadAnalyticFunctions.php

Answer 3

I think this is what you're trying for:
WITH Patient_Visits (Patient_Id, Hospital_Id, Branch_Id, Visit_Date, Visit_Order) as (
SELECT Patient_Id, Hospital_Id, BranchId, Visit_Date,
ROW_NUMBER() OVER(PARTITION BY Patient_ID, Hospital_Id, Branch_Id,
ORDER_BY Patient_Id, Hospital_Id, Branch_Id, Visit_Date)
FROM Hospital_Visits
WHERE Procedure_Id = 'P1'),
Hospital_Recent_Visits (Hospital_Id, Recent_Visitor_Count) as (
SELECT a.Hospital_Id, COUNT(DISTINCT a.Patient_Id)
FROM Patient_Visits as a
JOIN Patient_Visits as b
ON b.Hospital_Id = a.Hospital_Id
AND b.Branch_Id = a.Branch_Id
AND b.Patient_Id = a.Patient_Id
AND b.Visit_Order = a.Visit_Order - 1
AND b.Visit_Date + 10 > a.Visit_Date
GROUP BY a.Hospital_Id, a.Patient_Id),
Hospital_Patient_Count (Hospital_Id, Patient_Count) as (
SELECT Hospital_Id, COUNT(DISTINCT Patient_Id)
FROM Hospital_Visits
GROUP BY Hospital_Id, Patient_Id)
SELECT a.Hospital_Id, b.Patient_Count, c.Recent_Visitor_Count
FROM Hospitals as a
LEFT JOIN Hospital_Patient_Count as b
ON b.Hospital_Id = a.Hospital_Id
LEFT JOIN Hospital_Recent_Visits as c
ON c.Hospital_id = a.Hospital_Id
Please note that this was written and tested against a DB2 system. I think Oracle databases have the relevant functionality, so the query should still work as written. However, DB2 appears to lack some of the OLAP functions Oracle has (my version, at least), which could be useful in knocking out some of the CTEs.