OHDSI Home | Forums | Wiki | Github

Condition_occurrence, Death diagnoses

(Christian Reich) #21


I am not sure why this is necessary. Look:

  1. The source data (the physician declaring death of a patient) define the cause of death. We don’t have to second guess.
  2. Only where we have more than one mapping from one of these choices we have a problem, which is very rarely. because we cannot have mroe than one cause of death, but suddenly we got two or three. And for those cases we need a mechanism to choose.

So, if you could be so kind, take all the multiple mappings of ICD codes and pick one as a potential cause of death and we are done.

(Dmytry Dymshyts) #22

@nitishkjha sent me an example where
were mentioned as a cause of death
@nitishkjha, Am I right?

(Christian Reich) #23


Were they mapped from a single code? Then it is our problem. But if Nitish’s source gives him three reasons than he has to figure that out on his end.

To be honest with you, neither atrial fibrillation, nor pulmonary fibrosis, nor hypotension are acute causes of death. They are the foundation for an acute phenomenon (like a stroke). If anything, the fibrosis could wear you out.

(Dmytry Dymshyts) #24


probably in his data there are underlying conditions, but not an actual cause of death.

(Oksana Babich) #25

@Dymshyts @Christian_Reich

I have an example. We have one project where cause of death is identified by a special flag (specific for this particular project only). For one event date and one patient there are 4 causes of death:

G3…13 (IHD - Ischaemic heart disease)
B22z.11 (Lung cancer)
G581.00 (Left ventricular failure)
H25…00 (Bronchopneumonia due to unspecified organism)


  1. should we choose just one cause of death or create 4 records in cdm death table for that patient?
  2. if we have to choose one cause of death, how should we take the right cause of death?

(Christian Reich) #26


That’s an interesting problem. Right now, you have only one cause of death. Clinically, this case sounds like the pneumonia did the patient in, but clearly the other diseases have contributed. Can we do more than one death record?

(Dmytry Dymshyts) #27

According to this list every condition from the list will be cause of death.
And if we talk about actual cause of death, it we’ll be not a pneumonia but a respiratory failure.
So if in a raw data we have underlying conditions leading to death instead of “actual cause of death”,
thus if we display this in a CDM, we need to allow to put there several causes of death.

(Oksana Babich) #28

@Christian_Reich, @Dymshyts

I am not a medical expert, so just from my understanding: if this is according to medical standards to record several causes of death when a patient dies (actual, main, leading, contributing etc causes of death), in other words if multiple causes of death for a patient in source data is not a mistake, then ETL folks should not try to guess which one should go to cdm table. I would create multiple records in cdm death table for each cause of death.

Another question is multiple death dates for a patient. Here, in my opinion, it makes sense to choose one death date “up to ETL”, “the best approximation” etc. Because it is humanly impossible to die in different dates.

However, according to CDM V.5 conventions a person can have only one record in death table: “A person can have up to one record if the source system contains evidence about the Death”, “Each Person may have more than one record of death in the source data. It is the task of the ETL to pick the most plausible or most accurate records to be aggregated and stored as a single record in the DEATH table”.

And, as I see, it was not a case in CDM V.4 conventions for death table “Each Person may have more than one record in the Death table if there is available information for multiple contributing causes of death”.

I can be wrong in my assumptions, so I just want to make it clear for me about “only one record in CDM V.5 death table for a patient”.

Thank you very much for making it clear!

(Chris Knoll) #29

Just out of curiosity, the causes of death (as it relates to diagnosis codes) will be recorded in CONDITION_OCCURRENCE, correct? After thinking on what people are writing about the cause of death, we don’t need to look both in the DEATH table and CONDITION_OCCURRENCE table to find diagnosis codes for a patient, do we? IE: a person who dies of sudden heart failure (with no prior history of it), would the diagnosis code for that condition only be found in DEATH because it was a cause of death, or would it be found in CONDITION_OCCURENCE on the same date as they died?


(Christian Reich) #30


That could be a solution. We would have to officially change the Data Model back to how it was in V4, where you could have more than one record. The CDM WG does these kind of things.

However, you should not have more than one date. You got to figure out which date is the right one.

Yes, all those Conditions, including the actual killer, are in the Condition table. But only one should go into the Death table. But Oxana is saying she can’t pick and there is more than one.

(Irina Yabbarova) #31

Can we start creating multiple records in cdm death table for each cause of death now or we should wait for new version of omop spec?

(Christian Reich) #32

I’d say yes. The documentation can be fixed momentarily. But only one death_date please.

(Patrick Ryan) #33

thats a breaking change to our model conventions, so i think we should
discuss it on the cdm workgroup to make sure we have community consensus.

(Christian Reich) #34

Are you checking on us? :smile:

(Melanie Philofsky) #35

Our source data contains up to 11 fields for acme codes. These acme codes are ICD-10-CM and ICD-9 codes with the decimal removed. I read about acme codes here “ACME Automates the underlying cause-of-death coding rules. The input to ACME is the multiple cause-of-death codes (ICD) assigned to each entity (e.g., disease condition, accident, or injury) listed on cause-of-death certifications, preserving the location and order as reported by the certifier. ACME then applies the World Health Organization (WHO) rules to the ICD codes and selects an underlying cause of death.”

Has anyone used this code system and tool? Is it accurate? It seems to be inline with what the WHO expects for mortality reporting. Would this help solve the multiple cause of death issue?

It looks like we will still have use cases for the preservation of the underlying cause/s of death code/s. But the above may help determine which code should be used for the cause of death.

(Eldar Allakhverdiiev) #36

Hi all,
And what if suicide is reported as cause of death?
I’ve found concept

440925	Suicide	Observation	SNOMED	Clinical Finding	S	44301001


I’d like to change domain of Suicide from Observation to Condition.

(Dmytry Dymshyts) #37

Well, that’s interesting

  1. Suicide (obdservation) leads to Brain Injury (Condition) for example

  2. Depression leads to Suicide.
    and if somebody makes a study on Depression, they wouldn’t really care about classical Causes of death like “Brain Injury” or “loss of blood” or whatever people can make to kill themselves.
    In this case they need to know about Suicide itself.

So I see two options here:

  • put Observations as a cause of Death
  • make Suicide a Condition


(Christian Reich) #38

Yeah, suicide is an interesting one. Let me think.

(Pavel Grafkin) #39

Could you point me whether there was any further discussion and a decision written down? Storage of several records per a single death just because of different causes, as @Christian_Reich proposes, goes against simplest forms of DB normalization and therefore against good database design.

(Clair Blacketer) #40

Hi @pavgra as far as I know there was no decision made here. There was no proposal created so we haven’t discussed it in the CDM workgroup. For now, the DEATH table can only have one record per person.