OHDSI Home | Forums | Wiki | Github

Using concept sets to identify non-end-stage chronic liver disease conditions

(Gerald Pulver) #1

In preparing a list of standardized CONDITION_CONCEPT_IDs for chronic non-end-stage liver disease, I decided to see whether using pre-existing concept sets from ATLAS might save me from re-inventing the wheel. (Is this an appropriate usage?)

So, I selected a bunch of sets that sounded promising and downloaded them. Looking through the concept codes listed, I was surprised by how many seemed irrelevant. A look at the ConceptSetExpression table from my download provided an explanation for some.

It appears that there might be some mis-labeling involved.
conceptSetExpressionX.pdf (178.7 KB)

(Christian Reich) #2


Hm. Not sure that works. I think you have to invent a new wheel, here. What you want is to select liver diseases, for example SNOMED 194984 “Disese of liver” and all its descendants, and then exclude end stage liver disease, or chronic liver failure. That is not that straightforward. You could consider exclusion of the following SNOMED concepts and their descendants:

Something like that. You could go a whole lot more into the detail. In fact, this would be a perfect case for a probabilistic phenotype the way @Juan_Banda can build with Aphrodite, because the liver failure can manifest itself in a variety of different symptoms and problems, but none of these are really specific.

(Gerald Pulver) #3

My thought was to combine and de-duplicate the promising concepts and their descendants, then exclude the end-stage & acute codes and their descendants. The only advantage of that approach might be to capture codes inadvertently overlooked. I guess that doing so would be overkill.


(Christian Reich) #4

I totally understand what you are trying to do. Kind of like a minus/except function on liver disease and end-stage liver disease at the vocabulary level. But it won’t work. The hierarchy doesn’t have the precision for that, and worse, the ICD world where the conditions are mapped from don’t have it at all. So, you need to do the delta on the data level.

Makes sense?