OHDSI Home | Forums | Wiki | Github

Data quality metric around death (discussion; location of death)

(Vojtech Huser) #1

Death is a VERY important outcome to track. But one with a lot of issues around it. Often, there can be a big delay in learning about death outside hospital or outside other institution (outside nursing home).

For a sensitivity analysis, it may be important to know the breakdown of death-within-hospital-or-institution and death-at-home (or perhaps death-outside-an-institution). From data quality point of view, we anticipate missing data for death-outside-an-institution. Knowing that death-within-institutions agree with the expectation provides insights into data quality. (are there still missing deaths if only considering death-within-institution?)

In v6, Death Types were made Condition Types. (see query here for relevant condition types: http://athena.ohdsi.org/search-terms/terms?standardConcept=Standard&conceptClass=Death+Type&page=1&pageSize=15&query= )

In absence of a specific field that would say LOCATION _OF_DEATH, we hope to only ascertain somehow death-within-hospital-or-institution from claims data. (if that is possible at all; perhaps using Condition Code in the Header or Detail information of claims). In other words, the OMOP query should replicate what is stated here https://www.ncbi.nlm.nih.gov/pubmed/515338 and here https://palliative.stanford.edu/home-hospice-home-care-of-the-dying-patient/where-do-americans-die/.

Here is our US-based question for medical coders or users of claims data out there:
What is the in-hospital death code we can search for in OMOPed data? (I am afraid it is not a simple ICD or CPT code; or is it?)

(relevant death discussion here https://github.com/OHDSI/CommonDataModel/issues/210)

There seem to be no ICD9CM codes mapped to THE DEATH SCT CODE (http://athena.ohdsi.org/search-terms/terms/4306655). From where comes the existing list of ICD10CM codes and is it complete? (think “validation of death-via-claims phenotype”)

(Vojtech Huser) #2

Tagging @Christian_Reich @Patrick_Ryan @aostropolets for their thoughts?

I made some more progress in looking at the ICD9CM and ICD10CM sets of codes indicating death. By using Included Source Codes tab, one can see 115 ICD10CM codes and only 9 ICD9CM codes.

The link to this concept set is http://www.ohdsi.org/web/atlas/#/conceptset/1799115/included-sourcecodes

I took “THE CODE” for death (Concept ID 4306655) as the concept of which I took descendants (and Mapped to concepts).

If our decomposition of ICD9CM and ICD10CM codes into relevant SNOMED codes is correct (in Athena vocabulary), we should retrieve all of them.

Again the goal is to show that deaths in hospital for 70-79 year old patients in a given dataset are agreeing with some expected number. If that is not high as expected, perhaps we may be missing some inpatient hospital deaths as well. (which would be a data quality indicator).

(Dmytry Dymshyts) #3

This approach definitely works.
But I found 3 valid ICD9CM or ICD10CM concepts that mapped to some concepts not having
death as an ancestor.
348.82 - Brain death (ICD9CM)
798.0 - Sudden infant death syndrome (ICD9CM)
G93.82 - Brain death (ICD10CM)
So, we’ll add additional mappings to death in the next ICD9/10CM update.

(Anna Ostropolets) #4

Regarding ICD10CM:

Here we discussed what the possible ICD10 codes for death might be. This list has never been validated, so maybe that’s a good chance for us to figure out if those codes actually represent death.

ICD10CM concepts representing death