| OHDSI Home | Forums | Wiki | Github |
In my Death table, I’ve got persons who have multiple death causes because a single ICD code maps to two SNOMED codes. For instance:
ICD10
J96.21 Acute on chronic respiratory failure with hypoxia (HCC)
maps to SNOMED
389086002 Hypoxia
67905004 Acute-on-chronic respiratory failure
The specifications still say a person can have up to one death record. Seems to me this is a similar issue to the Conditions table. How do we resolve it?
I’m adding the Themis tag to this conversation since there isn’t a convention on how to handle this issue.
In Colorado we use a min function to take one concept_id for the record.
What would be the down-side of having multiple death records as long as they have the same date? It’s how the same issue is handled in the Conditions table.
Since a death certificate can have multiple reasons for death on it, would it not make sense, for those patients that one has said certificate for, the death table have the ability to match said reasons?
Luckily, we are outpatient only, so this is not something I have to worry about. I say this as a caveat that any extra work that this might cause, would not cause any issue for me. 
Multiple death records of the same death date is common. It has been discussed here and has been approved in previous themis discussion.
So the approved method from the link is:
“If multiple death records occur, the date and the person have to be the same, but the cause can be different. Can be reported by different sources as well.”
However, at the THEMIS discussion today (which I unfortunately missed), they proposed a different solution based on this discussion, which is also somehow marked as Accepted.
This solution is to put the causes of death in the condition table. I think this was precipitated by the envisioned move to v6.0 where the death date would be moved to the Person table and the causes to the Conditions table. However v6.0 never happened and never will. Is the next major release still planning on deprecating the Death table?
However, in current version 5.4, there is still a Death table which has cause-of-death data. It was suggested to put one cause-of-death in the Death table and the others in the Condition table, but that doesn’t make any sense to me from either a database perspective or an analysis perspective.
Remember, this is not 2 causes of death, it’s one ICD10CM cause of death (J96.21) that maps to two SNOMED codes (389086002 & 67905004).
So if I use a MIN() function to select just one of these codes, it will always be “Hypoxia”. Researchers looking for “Acute-on-chronic respiratory failure” will have to know to look in the Conditions table to find any other codes. There might as well not be a cause-of-death field in the Death table. Which was the point in version 6.0. But we’re not there.
So for the current version (v5.4) I think the convention should be multiple records (concept-ids) should be allowed for a Person as long as they have the same death date, which appears to be the established convention as quoted above.
If and when a new major version arrives which eliminates the Death table, this convention can be changed to put the cause of death in the Condition table.
Thoughts?
There is a recording of the THEMIS meeting here, but I don’t know if everyone can see it. The discussion starts at 11:24
From a point of logic, what you are purposing to do, makes sense. My question is to the tool implementers, will this break any existing said tools?
Also, will the cause you to have DQ errors on your next AOU upload? That is separate from this discussion.
There are two reasons we may want multiple death records (all at the same day, though):
In the discussion, we emphasized the second, while all those causes of death may go to CONDITION_OCCURRENCE.
But I hear you: if we have a cause_concept_id field people expect it to be filled with meaningful information, and a convention to not use it is suboptimal while we are still in 5.4 (in 6.1 we can go back and eliminate all of this as a problem).
Should we? The only convention remaining would be that only one death_date must exist?
Correct, we need to pass our conventions over to the DQD folks. This is true for all of them, not just this.
On Thursday, August 3rd, the Themis group discussed the topic of placing multiple death records in the Death table.
Findings: It was discovered the most recent convention for multiple causes of Death is contained in the CDM v6.0 documentation, “Similarly, the cause of death is stored in the CONDITION_OCCURRENCE table using the CONDITION_STATUS_CONCEPT_ID 32891 for ‘Cause of death’.”
The CDM v5.4 doesn’t mention what to do with multiple death records. It only states one record is allowed, “A person can have up to one record if the source system contains evidence about the Death”.
So, based on the most recent documentation, these are the current conventions. The documentation @QI_omop linked has been superseded by the above.
Agreed.
Correct, this is not currently documented in the v5.4 conventions. Once documented, researchers will know to look for other causes of death WHERE condition_status = 32891.
And technically, Hypoxia is the cause of death and acute on chronic respiratory failure is a contributing factor. And whatever caused the acute respiratory failure, pneumonia, smoke inhalation, traumatic injury, etc., is the contributing factor, too.
Next steps:
The Themis issue tracker is here and the issue template contains all the attributes necessary for a complete issue submission.Thoughts?
The THEMIS Issue is here: Multiple Death Records Due to Death_type or Cause - Version 5.4 · Issue #79 · OHDSI/Themis · GitHub
Please see the following thread for additional discussion