OHDSI Home | Forums | Wiki | Github

Resolving observation_period_end_date without exact information on a last day


(Anton Agranov) #1

I am sorry, if the question has been already resolved, still I didn’t find exact answer.
The basic question is “What is the best option to choose observation_period_end_date?”

The source set of data provides information on starting and ending month of observation for every unique person in format “01-mm-yyyy”. It is fine with the startind date because of the 1st day of every month, and totally a problem when it is about the end date. Worth to mention that some observation_period_end_dates tend to be the end_dates of data when is was added to the dataset. So in some cases there are event after observation_period_end_date, that is fine, taking into accountant format “01-mm-yyyy”.

Also the issue is that data was gathered from insurances sources, but there are no specific information on insurance policy, neither the period of it, only the sums to be recovered, meaning that particular data is more about hospitalization than insurance. So it is a tricky question how to populate fields based on the provenance of the visit record as “claims”, “EHR” or any others.

According to:

it is fine having events outside of observation period

And concenrning this:

it is a possible to use event with maximum date with additional 60 days for person with records where contains evidence about the death.

So, what is the best solution for observation_period_end_date for particular dataset?
Should it be MAX from event_dates meaning that observation_period_end_date will not be used?
Or it is possible to fill observation_period_end_date with the last day of corresponding month in order to cover all events?

(Christian Reich) #2

Why is this such a big problem? In the worst case, you are one month off. Remember: We need the observation_period for denominator and wash-out/follow-up purposes. If you cut it too short you lose data. If you are leaving it too wide you will include denominator time even though data is no longer actively captured. Can’t see how a couple of days one way or another will have a big impact.

You may want to use the CDM documentation for conventions.

Ditto. There is a section ETL conventions. It really is up to you to figure out what’s the best shot - data driven or by some enrolment information.