How to populate death_date when source does not provide

[clairblacketer]

When the person's death_date is not provided but the patient is deceased

CDM or THEMIS convention?

Themis convention as it is not applicable to all and the CDM covers incomplete death_date and death_datetime data. https://ohdsi.github.io/CommonDataModel/cdm54.html#death

Is this a general convention?

No

Summary of issues

• With the death_date required, some have argued for the allowance of a death indicator flag when the date is not available.
• The death_date and death_datetime missing accuracy has been addressed in the CDM, but not the absence of the date entirely to indicate the deceased status of the person.
• The issue may be viewed as a data quality issue prior to ETL that should be resolved.

Summary of answer

• Recording the last known Provider-Person interaction based on other clinical entries. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/19

• The most obvious recommendation is to not use the death_date if your ETL source data does not have a method to properly insert death_date with the accuracy expected. Not a fix, but the best recommendation I can find. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/41

Related links

https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241 https://github.com/OHDSI/CommonDataModel/issues/210 https://ohdsi.github.io/CommonDataModel/cdm54.html#death

Other comments/notes

[clairblacketer]

Hi @assimilateDave I opened this as a Themis issue so I could comment. The note about the removal of the death table in v6.0 relates to multiple causes of death and not necessarily what to do when death date is not provided in the data. Anything in v6.0 by definition will not be retained in the v5 series as many of them were breaking changes.

Have you found a suggestion for how to populate death_date if it is not given in the source?

[assimilateDave]

Suggestions for "writing" an estimated death date include:

• Recording the last known Provider-Person interaction based on other clinical entires. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/19
• Post a fictitious future date e.g.2099-01-01
• Bring in a death "flag" indicator in addition to allowing the death_date to be filled
• The most obvious recommendation is to not use the death_date if your ETL data does not have a method to properly complete data with the accuracy you expect! Not a fix, but the best recommendation I can find. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/41

[clairblacketer]

Thanks @assimilateDave for working on this one! I would suggest we not include the fictitious future date as this has implications for the observation period and is almost the same as having a null. There would be no way to use such a date. Similarly, the flag option was brought up in the past but the OMOP CDM specifically typically doesn't include flags but instead relies on the records themselves to indicate that something happened to the person. In this way the dates are critical because wthout a date there is no way to temporally associate the death with any other events. In this case I would use the first and fourth bullets you have as the methods for answering this question

[assimilateDave]

I am unable to edit the original document content and would like to replace the "Summary of answer" with the following:

• Recording the last known Provider-Person interaction based on other clinical entries. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/19

• The most obvious recommendation is to not use the death_date if your ETL source data does not have a method to properly insert death_date with the accuracy expected. Not a fix, but the best recommendation I can find. https://forums.ohdsi.org/t/what-to-do-with-null-death-dates-in-omop/4241/41

[clairblacketer]

Thanks @assimilateDave ! I updated it, I opened the issue so I guess it didn't give you edit rights

[MelaniePhilofsky]

Issue # and location

Forum discussion

• Forum
• Forum

Issue summary

In order to insert a record in the death table, the death_date must be populated. However, some data sources only contain a death flag. In this case, what should the ETL do when there is not a date at the source, but there is a flag indicating the person is deceased?

Convention type

Table

CDM table

Death

CDM field

NA

Links to issue discussion

• Forum
• Forum

Provenance of data

All

The ratified convention

If your source data do not have a method to properly identify a realistic death_date, then do not create a record in the Death table. If you think you have a robust method to do the imputation, and you checked that against the data where you do have the death date, you use it. One method of imputation uses the date of the last Patient-Provider interaction.

Date of ratification/published

January 29, 2022

Downstream implications

No

Link to DQD check

isRequired.

Related conventions/further information

NA

#Tags Dates, Death