CDM Workgroup Meeting 3Dec2019

[clairblacketer]

Hi Everyone,

Tomorrow we have a CDM workgroup meeting at 1pm eastern. We will be finishing our discussion of the PERSON table and (hopefully) move on to the OBSERVATION_PERIOD table. The Skype meeting link can be found on our homepage https://ohdsi.github.io/CommonDataModel/index.html and the straw-man description of the OBSERVATION_PERIOD table is below. Please review and come ready to discuss.

Clair

Field Name	User Guide	ETL Conventions
observation_period_id	A Person can have multiple discrete Observation Periods which are identified by the Observation_Period_Id.	Assign a unique observation_period_id to each discrete Observation Period for a Person.
person_id	The Person ID of the PERSON record for which the Observation Period is recorded.
observation_period_start_date	Use this date to determine the start date of the Observation Period	It is often the case that the idea of Observation Periods does not exist in source data. In those cases, the observation_period_start_date can be inferred as the earliest Event date available for the Person. In insurance claim data, the Observation Period can be considered as the time period the Person is enrolled with a payer. If a Person switches plans but stays with the same payer, and therefore capturing of data continues, that change would be captured in PAYER_PLAN_PERIOD.
observation_period_end_date	Use this date to determine the end date of the period for which we can assume that all events for a Person are recorded and any absense of records indicates an absence of events.	It is often the case that the idea of Observation Periods does not exist in source data. In those cases, the observation_period_end_date can be inferred as the last Event date available for the Person. In insurance claim data, the Observation Period can be considered as the time period the Person is enrolled with a payer.
period_type_concept_id	This field can be used to determine the provenance of the Observation Period as in whether the period was determined from an insurance enrollment file, EHR healthcare encounters, or other sources.	Choose the observation_period_type_concept_id that best represents how the period was determined.

[mdewilde (Marcel de Wilde)]

Hi Clair,

unfortunately, I cannot join this meeting today. But I have some things I would like to bring up for the discussion.

As you probably know, our IPCI database has a lot of data outside observationperiod. The convention to have all this valuable information about a person in the CDM was very important for us. After some long discussions we are very happy that its now allowd to have data outside of observation period.

Data outside observation period

Convention 6:
Events CAN fall outside of an observation period
though they should fall in a valid payer plan period,
such as Medicare Part D, which can overlap an
observation period. However, time outside of an
observation period cannot be used to identify people.
To ensure quality, events outside of an observation
period should not be used for analysis. THEMIS
issue #23

The phrases in the convention “cannot be used to identify people” and “should not be used for analysis” are pretty hard statements and are basically disabling the rule to have data outside observationperiod.

We actually use this data before observation_period a lot for our analysis. Also to identify and excude patients. For example: we would miss about 30% of the DM2 patients if we ignore the DM2 codes outside observation period. Simply because in most dossiers the GP is registering this DM2 code only once in the system and has no reason to repeat this. And in or database this is the case for most chronical diseases. But also the other tables out database contain a lot of valuable information about patients before observation period.

Its clear, that an event outside observationperiod cannot be used for incidence rates by itself. But it can be needed to detect that the first event within observation period is NOT the first event for the patient, and cannnot always be interpreted as an incident code.

For claims databases I can imagine that data outside observation period is probalby not reliable. How can there be a claim before the patient was even insured? But for EHR databases a lot of valuable data can be in the dossiers before observation period. It can contain a lot of history data. In the Netherlands, if a patient moves to a new GP the data of the previous GP is also transfered to the new GP. The observation period is starting when registering with the new GP, but now we can have many years of history data as well. Often starting all they way back to the year 2000 or even before.

Also, if we ignore the data after observation period, we can miss the cause of death and other death related information. An authopsy can contain the cause of death and is hopefuly after death.

Multiple observation_periods

In IPCI a patient only had 1 observation period.

A patient can have multiple observation periods and its not required that these are connected. Also we may aussume that all data within observation period can be used to calculate duration and time between. But if we have patients with two observationperiods and there are many years in between, we actually have a gap in the data. If this patient has DM2 in the first obsevbationperiod, and gap of 10 years and a first insuline in the next observation its very likely this will be interpreted as a patient that started with insuline 10 years after the DM2 diagnosis. So, basically we then have a patient with gaps in observationperiod. How should analysis deal with this situation? Are there easy ways to deal with this? How important is it to have multiple (disconnectedO observation periods?

End date

A simpe one: its not clear to me if the observation_end_date is included or excluded.

Greetings,
Marcel