OHDSI Home | Forums | Wiki | Github

One to Many relationships for mapping a non standard to OMOP standard codes

If I am standardising an ICD10CN code to OMOP (SNOMED), where there is a one-to-many relationship, for example

S21.331 (ICD10CN code) maps to SNOMED 210410008,10850301000119105,735765006

What should I expect after our automated ETL, for example should the patient have (i) one OMOP SNOMED code or the patient has (ii) multiple OMOP SNOMED codes.

If (ii) multiple, does this create artefacts in the dataset where it looks like a patient now has multiple conditions? On the analytics side how do you distinguish between multiple conditions as an artefact of the etl (i.e. from a single diagnosis event) to real cases of multiple diagnoses

1 Like

Don’t worry, @gruber. It’s (ii). You don’t “count” conditions. So, making 3 out of 1 is not a problem, as long as all three are actually true.

And they are: There is no one SNOMED concept for S21.331 “Puncture wound without foreign body of right front wall of thorax with penetration into thoracic cavity”. The attributes are (i) puncture wound, (ii) no foreign body, (iii) right front wall of thorax and (iv) penetration. Collectively, the three SNOMEDs cover the space.

This system works, but has side effects. The better way would be creating an OMOP Extension concept to SNOMED which would pre-coordinate everything into one. But who has time to build all those?

1 Like

Thanks for your response. I understand that you are saying that all 3 terms collectively define one condition. My issue is that if you need to get a count of the number of conditions in a hospital for a certain period of time or the number of conditions per patient for the case that a hospital needs to provide this statistics for planning or funding reasons those conditions will be counted as 3 conditions and will not give the correct statistic. How would you suggest to overcome this issue?

1 Like

Understood. But there is no such a thing as a “number of conditions”, because conditions are not independent from each other facts. For example, renovascular hypertension means hypertension caused by kidney disease. Is this one disease or two? Same is true for diabetic retinopathy. Chronic hepatitis with hepatic coma. I am not even talking about hard to define things like long Covid.

In addition, OMOP is not a model for hospital reporting or optimization. It is a patient-centric model, allowing to research on the natural history of disease and health outcomes of interventions.

You count what you got in CONDITION_OCCURRENCE. Most likely, this will be a different number from what some BI report gives you in your EHR. The number will not be drastically different. But it will be different. There is no exact number possible in this world.