Counting rows based on consecutive dates

Question

I need some guidance. I have a database of drug fills from our data warehouse. There is a unique row for every drug filled. I a general rule, there is a 30 day prescription filled each month, and my research question needs me to look at consecutive months filled.
So, if I have
Name, Drug_Name, Fill_Date
John Doe, DrugX, 05-2010
John Doe, DrugX, 06-2010
John Doe, DrugX, 07-2010
John Doe, DrugX, 08-2010
Mary Jane, DrugX, 05-2010
Larry Smith, DrugX, 06-2010
Larry Smith, DrugX, 07-2010
John Doe, DrugX, 12-2010
Larry Smith, DrugX, 09-2010
Larry Smith, DrugX, 12-2010
I need to identify patients with at least 3 consecutive months of a drug – so in the case above, that would only be John Doe. I can just do a unique count per patient and look at >= 3, because then I would also pick up Larry Smith who has 4 months of prescriptions, but not 3 consecutive months.
Suggestions?
Thank for the help.
Rich
Edited by: Ricco on Oct 19, 2010 11:19 AM

Answer 1

Hi,
Whenever you have a question, please post the sample data in a form that people can use to re-create the problem and test their ideas.
For example:
CREATE TABLE     table_x
(      name          VARCHAR2 (20)
,      drug_name     VARCHAR2 (10)
,      fill_date     DATE
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('John Doe',    'DrugX', TO_DATE ('05-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('John Doe',    'DrugX', TO_DATE ('06-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('John Doe',    'DrugX', TO_DATE ('07-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('John Doe',    'DrugX', TO_DATE ('08-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('Mary Jane',   'DrugX', TO_DATE ('05-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('Larry Smith', 'DrugX', TO_DATE ('06-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('Larry Smith', 'DrugX', TO_DATE ('07-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('John Doe',    'DrugX', TO_DATE ('12-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('Larry Smith', 'DrugX', TO_DATE ('09-2010', 'MM-YYYY'));
INSERT INTO table_x (name, drug_name, fill_date) VALUES ('Larry Smith', 'DrugX', TO_DATE ('12-2010', 'MM-YYYY'));
COMMIT;Also, post the exact output you want. In this case, it's probably very simple. It's unclear from your description, but I think you want this:
NAME                 DRUG_NAME                                                 
John Doe             DrugX                                                      Here's one way to get those results from that data:
WITH  got_month_num      AS
     SELECT DISTINCT  name
     ,           drug_name
     ,           MONTHS_BETWEEN ( TRUNC (SYSDATE,   'MONTH')
                          , TRUNC (fill_date, 'MONTH')
                         ) AS month_num
     FROM          table_x
--     WHERE          ...     -- Any filtering goes here
,     got_group_num     AS
     SELECT     name
     ,     drug_name
     ,     ROW_NUMBER () OVER ( PARTITION BY  name
                               ,                    drug_name
                         ORDER BY        month_num
                       ) - month_num     AS group_num
     FROM    got_month_num
SELECT DISTINCT  name,     drug_name      
FROM             got_group_num
GROUP BY        name,     drug_name,  group_num
HAVING           COUNT (*)       >= 3
ORDER BY      name,     drug_name
;I'm guessing at some things.
For example, you didn't say what role (if any) drug_name plays in this problem, and it's hard to guess since every row has the same drug_name.
I'm guessing that every distinct combination of name and drug_name is a world of its own, having nothing to do with rows that have the same name but different drug_names, or the same drug_name but different names.
This assumes that name uniquely identifies each patient.
This does not assume that there is only one row per name and drug_name in each month, or that all fill_dates are at the same point in the month.
The query above does require the fill_dates to be in consecutive calendar months. If one fill_date is January 31, 2010, and the next fill_date for the same patient and drug is March 1, 2010, then the rows are not considered to be consecutive, even though there are only 29 days between fill_dates.
Depending on your data and your requirtements, the first sub-query (got_month_num) may not be necessary.
A "group" (as defined by group_num) is a set of rows for the same patient and drug in consecutive months. If we number the months with consecutive integers in oreer by date, and number the rows with consecutive integers also in order by date, then the difference between those integers will be constant for all ros in the same group.
Edited by: Frank Kulash on Oct 19, 2010 2:48 PM