OHDSI Home | Forums | Wiki | Github

Voijtech's question: Conventions for iological relationships in FACT_RELATIONSHIP table

CDM Builders
#1

I would like to provide an example to represent 222 is child of 111 in the fact_relationship table.
What is the domain code I should use for the person table? (and what vocabulary ID should I look into when I search for it?)

domain_concept_id_1 and …id_2.

See my draft proposal below.

In fact, what are domain codes for all tables? (e.g., relationship between diagnosis and procedure) (the CDM v5.01 specs may want to list those)

Vojtech

We have some genomic data where biological parenthood is certain (the same is true for 99.9% of mothers delivering a baby in hospital
(except for rare cases of surrogacy).
I would like to add some more clarifying rules to the CDM v5 specifications.

Consider the following fact:

person_id 111 is “biological parent of” person_id 222

bilogical parent is SNOMED:13646006
CDMV (CDM Vocabulary) CID (concept_id) for this snomed code is: tbd

In final v5 CDM specs, the person_realtionship table was removed and we wisely decided to use fact_relationship

FACT_RELATIONSHIP has the following table fields

domain_concept _id_1 The concept representing the domain of fact one, from which the corresponding table can be inferred.
fact_id_1 The unique identifier in the table corresponding to the domain of fact one.
domain_concept_id_2 The concept representing the domain of fact two, from which the corresponding table can be inferred.
fact_id_2 The unique identifier in the table corresponding to the domain of fact two.
relationship_concept_id A foreign key to a standard concept identifier of relationship in the Standardized Vocabularies.

Hence to represent the person to person relationship
the table would contain the following two rows

fact_id_1: 111
fact_id_2: 222
relationship_concept_id: SNOMED:13646006

and

fact_id_1: 222
fact_id_2: 111
relationship_concept_id: SNOMED:75226009

domain concepts would be in both cases: (tbd)

We need to design a binding set of rules so that the same SQL code runds on many CDM instances.

I would like to propose an initial suggestion usinga simple appraoch and the following codes:

SIMPLE ANCESTOR APPROACH
“biological parent SNOMED:13646006”
reverse would be
“biological child SNOMED:75226009”

MORE COMPLEX GENDER BASED APPROACH (SNOMED and CDMV codes are not provided)

X is mother of Y (Y is male) reverse: Y is son of X
X is mother of Y (Y is female) reverse: Y is daughter of X
Z is father of Y (Y is male) reverse: Y is son of Z
Z is father of Y (Y is female) reverse: Y is daughter of Z

See the URL below to see more SNOMED codes
http://schemes.caregraf.info/snomed#!125678001

#2

Voijtech.

You are putting the finger into a wound here. Several points.

  • The domain codes you get by querying select * from concept where vocabulary_id=‘Domain’;
  • The problem is there is no domain ‘Person’. There is only ‘Gender’, ‘Race’ and ‘Ethnicity’. We have to figure out what we do in terms of a convention here. Neither of the 3 are very natural. But in the absence of any, let’s use ‘Gender’, at least we all have one (as opposed to Race and Ethnicity). What do you think?
  • All relationships have the domain_id = ‘Relationship’. You can find anybody in your family there. :smile:
  • However, we should establish and add to the Conventions section of the CDM which concepts we actually use for relationships. 13646006 - 75226009 makes sense to me. We could add ‘Natural sibling’ 82101005, ‘Fraternal twin’ 313416000, ‘Identical twin’ 313415001. We should think of:
  • Adoptive child 48673000
  • Adoptive parent 41953004
  • Biological grandparent 78272007
  • Foster child 39062003 (maybe it’s the same as Adoptive)
  • Foster parent 90921004
  • Stepchild 86764008 (maybe it’s the same as Adoptive)
  • Stepparent 74128007
  • Spouse 127848009
  • Not sure what to do about grandparents and grandchildren, and then maternal and paternal versions thereof. Also, not sure I know about cousins, brother-in-laws and all those wonderful people arriving at Thanksgiving Dinners.
  • I don’t like ‘mother’, ‘father’, ‘sister’, ‘brother’. We already know what gender the Persosn have. If we combine the fact relationship with gender we will run into the danger of creating a contradiction. Also, there are cousins. In English, there is no distinction between male and female cousins (there is in French and English, and I think also in Chech).
#3

Christian,

I’m very interested in the familial relationships. We will begin ETL of parent / child trios next week.

Bill

#4

I came across this and was wondering if either of you would know the meaning of the SNOMED-CT codes found within OMOP regarding ‘Mother delivered’ and ‘Mother not delivered’
I was specifically wondering why these codes are separate from ‘full term normal delivery’ or ‘single live birth’ or any normal delivery code for ‘Mother delivered’ and why ‘Mother not delivered’ isn’t merged with say pre-eclampsia or some other condition of a pregnant woman close to delivery?
Was hoping some of you may have some insights into this
Thanks!

#5

@Mary_Regina_Boland:

The short answer is: OMOP doesn’t add anything to the SNOMED concepts, it just uses them and the relationship.

With respect to your questions:

  • “Mother delivered” and “Mother not delivered” is a condition of the nother, while “Single live birth” is about the child.
  • “Preeclampsia” is a disease or a pathological condition. “Mother not delivered” is just a state in the pregnancy. Why would you want to merge them?

What are you trying to do?

#6

thanks @Christian_Reich very helpful.

Regarding my purpose, I was trying to ascertain if there is any added value to these codes. Perhaps they are just descriptors of the patient state that are sometimes billed.
What I think is a little strange is that they don’t seem to be consistently applied throughout the pregnancy, which makes me think there is probably a condition that warrants the coding. But perhaps its a weird thing where if a doctor thinks to use the code then he or she will use it.

Regarding ‘single live birth’ being a state of the child, i find that to be fascinating because it seems to be very similar to ‘full-term normal delivery’ in our records systems and to be more indicative of the state of the mother and not the child.

My ultimate interest in these codes and when/how they are applied is because of the birth month-disease association study that I hope to expand to OHDSI shortly (working on finishing up some final details with this to make it fully OHDSI-compatible). Previously, I was treating each of these codes as ‘conditions’ although from your comment it would seem some are descriptors of the patient state (which may hint at an underlying condition)

Does that make sense?

#7

All that depends on the data source you are working with. If that is claims, then yes, the record will be there to justify something. In case of pregancies these are probably the check ups visits. So, I would work closely with the data.

Regarding conditions: All conditions are equal. OMOP doesn’t make a difference whether it is a pathological or physiological condition (like pregnancy). I was just responding to your idea of merging different conditions into one.

#8

Thanks.

Yeah i was considering merging the conditions, but given your comments it might be better not to and perhaps to consider conditions that indicate the patient state in a separate manner, i’ll have to think about it.

Thanks so much for your insights!

t