Convention for Missing Clinical Event Dates

[katy-sadowski]

Issue Name:
We're missing an official convention for how to handle clinical events that are missing a date in the source data. Since the x_start_date field is required in all clinical event tables, we should provide guidance for users to ensure missing dates are handled consistently across CDMs. In this issue the focus will be on the following tables:

• condition_occurrence
• drug_exposure
• procedure_occurrence
• device_exposure
• observation

These types of events may be found in medical history / patient reported contexts in which they may not have dates recorded. For other types of events (visits, measurements) the use case & implementation of such a convention is less clear so we'll focus on these 5 in this issue.

Issue Type: Lack of an ETL convention for a CDM table and/or field

Description:
Current documentation:

• https://ohdsi.github.io/CommonDataModel/cdm54.html#CONDITION_OCCURRENCE; https://ohdsi.github.io/CommonDataModel/cdm54.html#DRUG_EXPOSURE; https://ohdsi.github.io/CommonDataModel/cdm54.html#PROCEDURE_OCCURRENCE; https://ohdsi.github.io/CommonDataModel/cdm54.html#DEVICE_EXPOSURE - these docs do not describe how to handle source records w/ missing (start) date
• https://ohdsi.github.io/CommonDataModel/cdm54.html#OBSERVATION - for observation_start_date, states "For some observations the ETL may need to make a choice as to which date to choose."

We should update docs with explicit instructions on how to handle missing dates under certain circumstances. Given the "junk drawer" nature of OBSERVATION we won't be able to cover all circumstances, but there are some well-understood use cases like "history of" events we could make a convention for.

Related forum post: https://forums.ohdsi.org/t/how-to-handle-missing-dates-in-all-cdm-tables/9659

The reason I re-kindled that thread is that in DQD we're trying to provide guidance to users on how to resolve failing data quality checks. For isRequired I'd like to be able to point to CDM docs in order to recommend how users should handle rows w/ a NULL x_start_date (at least in certain cases).

Suggested solution:
For conditions, drug, procedures, devices missing a date in EHR we could recommend putting them in the observation table as "history of" observations (a common scenario is for historical diagnoses to be recorded on a problem list without a date). We'll need a convention for how to populate the observation start date for these events (perhaps: if they are associated with a visit, use the visit start date, otherwise ???).

I'm curious if there are other scenarios we can/should address in addition to the one above; this may be worth discussing.

Reported By: Katy

[MelaniePhilofsky]

@katy-sadowski Do we want to discuss on a Data Quality WG call or a Themis WG call?

[katy-sadowski]

Yes! But personally my plate is too full to push this forward right now...things are about to get crazy with the Symposium, and then the holidays 😅 maybe we can put this on a Themis call in the new year?

[MelaniePhilofsky]

Yes, no rush! 2025 it is :)