OHDSI Home | Forums | Wiki | Github

Condition type for French discharge diagnosis codes


#1

We are currently working on the ETL of the French National Healthcare System database (SNDS), which merges, among others, claims data and hospital discharge summaries.

Four different types of condition can be found:

  • Primary diagnosis [hospital settings]: the health problem that motivated the admission in the hospital.

  • Related diagnosis [hospital settings]: exists only if the primary diagnosis is a care procedure with a code Z of the ICD-10 classification (e.g. chemotherapy session). It indicates the pathology at the origin of the care procedure

  • Associated diagnoses [hospital settings]: specified if they justify the use of specific healthcare resources. They are mainly underlying chronic diseases, making patient management more complex (and more expensive).

  • Long term disease registration: mainly concerned costly chronic diseases, obtained at the request of a patient’s practitioner and validated by the health insurance system physician. Once registered, patients receive full reimbursement for expenditure related to the disease.

We will use 44786627 “Primary Condition” for Primary diagnosis. Associated diagnoses could fit 44786629 “Secondary Condition”, but what about the others?

Thank you !


(Vojtech Huser) #2

your type 4 (Long term disease registration) is truly a new and unique type of condition entry not covered well. You can create a local code for it (2B+ range; not preferred by me) or request a new condition type (I would prefer this).
From analytical point of view I can see it as quite useful for cohort definition. (we will have to support country specific phenytope variants one day anyway) (translate this French phenotype concept to USA claims data or Japan data)

your type 2 is also interesting. As informatician I always wish for reason for drugX or reason for procedureY and it seems like French data gives me that so I would want to not loose it. (so local concept_id or new network wide type)


#3

I agree!

Who should I bother to request this new condition type concept_id?


(Christian Reich) #4

@nthurin:

This is a good time to bring this up. We are currently revising all the Type Concepts and making them more international.

44786627 “Primary Condition” sounds about right. But we also have two more:

  • 38000183 “Inpatient detail - primary”
  • 38000199 “Inpatient header - primary”

The distinction between detail and header is a US claims artifact, and we will clean that up.

You probably don’t need that. Any diagnosis can be related. What you do need is the explicit “Related to” relationship to a procedure. Right now, there is no explicit cross-link between the CONDITION_OCCURRENCE and PROCEDURE tables. Funny thing is in an older CDM we used to have a field relevant_condition_concept_id to point to the related diagnosis (but not the to a record in the table), but we abandoned it for lack of utility and lack of data.

So, now your only choice is to write records into the FACT_RELATIONSHIP table linking those together. There is no need for a Type Concept.

Is there an analytical use cases for you to know this? If so, what should the type be called?

Same question. Is there a need for knowings in some analytic you are planning?


#5

Thank you for your feedback @Christian_Reich
I agree on Primary diagnosis, 38000183 “Inpatient detail - primary” could also be a good option.

Regarding this point, the issue is not only to conserve the relationship but also to consider the condition. So far these codes are very useful in our studies. Usually, we almost always consider them as principal diagnoses.
I think, alongside using FACT_RELATIONSHIP table, we could also put them into CONDITION_OCCURENCE table with the same Type Concept than the Primary diagnosis (“Inpatient detail - primary” or “Primary Condition”)

As explained above they are mainly underlying chronic diseases coded by hospitals to justify an increased cost of stay. They are not the motive for the hospital admission, but they are still very useful to characterize patient chronic conditions, since in the SNDS we do not have outpatients diagnoses.

What about “Significant associated condition” ?

Yes! They are eminently important! They are very specific with a very high positive predictive value. It is the major way to identify chronic conditions for out/inpatients in the database (e.g. diabetes, parkinson disease, cancer…).

We could call them “Long term condition”.


(Dmytry Dymshyts) #6

Once you agree on the list of new concepts, please post the issue here


t