OHDSI Home | Forums | Wiki | Github

Death CDM v 6.0

Hello!
I have read a discussion about death and have several questions:
Case #1
There are several causes of death with different dates in condition_occurrence table.

Should I keep all of them in condition_occurrence table with type_concept_id that has class ‘Death Type’, but choose only one date as a date of death and populate cdm.person.death_datetime field with this date? Or I should remove record with the wrong date of death from condition_occurrence table and add a record to observation table as an alternative date of death?
For example:
condition_occurrence:
record#1: 2014-02-05 ICD9CM: 799 Other ill-defined and unknown causes of morbidity and mortality
record#2: 2014-02-10 ICD9CM: 799.9 Other unknown and unspecified cause of morbidity and mortality

Let’s say we have the rule to choose the earliest date of death as a real date of death.
Therefore cdm.person.death_datetime = 2014-02-05

Option 1:
Should I remove record#2 from condition_occurrence and create a record in observation table?:
condition_occurrence:
record#1: 2014-02-05 ICD9CM: 799 Other ill-defined and unknown causes of morbidity and mortality
observation:
2014-02-10 concept_id = 4306655 ‘Death’
person:
death_datetime = 2014-02-05

Option 2:
Or I can keep 2 records and assign both records type_concept_ids with class ‘Death type’?:
condition_occurrence:
record#1: 2014-02-05 ICD9CM: 799 Other ill-defined and unknown causes of morbidity and mortality
record#2: 2014-02-10 ICD9CM: 799.9 Other unknown and unspecified cause of morbidity and mortality
person:
death_datetime = 2014-02-05

If I understood correctly, observation table might contain several dates of death, but what about condition_occurrence table?

Case 2
There is only death indcator in source data.
Should I create a record in observation table using concept 4306655 ‘Death’? Or storing death date in person table will be enough?

There is a cause of death in source data. It goes to condition_occurrence table. Should I create a record in observation table using concept 4306655 ‘Death’? Or storing death date in person table and cause of death in condition_occurrence table will be enough?

Thank you in advance!

@Alexandra_Orlova:

Don’t try to stuff data into the OBSERVATION table so you don’t need to throw them away. It’s also called “carbage can”, so you are throwing things away one way or another. :slight_smile:

In your example, the cause of death is a non-mappable record (unknown cause of death). For such records, I would not worry about what your choice would be.

But if you really have different causes of death I would:

  • Pick the best (latest) death day and record it in PERSON
  • Record the cause at that date as a CONDITION with the appropriate Death Type
  • Take the other record and put it into CONDITION with the appropriate Death Type with the date as in the source (and thus differing from death_datetime in PERSON)

@Christian_Reich
Thank you for very quick response!

What about Case 2?:
If there is only death indcator in source data, should I create a record in observation table using concept 4306655 ‘Death’? Or storing death date in person table will be enough?

I am asking about observation table because there is the phrase in wiki: “The DEATH_DATETIME in the PERSON table should not be used as the way to find all deaths …deaths should be found through the OBSERVATION table and the PERSON table is only used to determine which death date should be used in analysis”
It stumped me because I understood that I must create records in observation table for each death, probably I misunderstood :frowning:

Why not? Why store the same thing twice? It’s bad database modeling.

That’s an awful phrase. Will fix it. The OBSERVATION table has nothing to do with it.

1 Like

@Christian_Reich,

What about the case where you only have an indication of death, no date, what do you do then? You can’t put anything in the person table’s DEATH_DATETIME field, but you need to know the person has died. If you add a record to another table then people have to know that they have to look in multiple tables to find out who is deceased.

@jscherdin:

Hm. I am not sure. We all die. So, adding death to the patient records really doesn’t add any information (except that the death was at some point before “now”). In order to do analytics you need to know the context and timing. Otherwise it’s a entity to be discussed by philosophers.

That’s not entirely true. There are times that just knowing deceased status is needed. For instance, if you are doing patient recruitment, you would only want those patients that are alive so being able to filter out those that are known to have died is necessary.

We could create a convention like 31-Dec-2099 as an unknown death date. But the consequence is that people have to filter that thing out each time they want to use real death dates. This is a proposal you’d have to get through the OMOP CDM WG. Want to do that? Let @clairblacketer know.

t