Embrace domains and concept vocabularies that are DAGs

[artgoldberg]

The LeafDB.app.Concept table represents concept hierarchies in its Id, ParentId, and RootId fields. This data structure works well for concepts that are structured as trees, like ICD-10-CM codes for conditions.

However, some domains and concept vocabularies are not structured as trees. For example, the LOINC hierarchy for measurement provided by the UMLS is structured as a DAG. It has 190,217 records in the table generated from the UMLS that you shared with me @ndobb . Unfortunately, 27,640 of these records have multiple parents. And 4,245 of them are parents themselves.

In fact, one record has 6 parents:

AUI	ParentAUI	UiDisplayName
A18322336	A18243446	Intravascular diastolic (LNC:11377-9-89281-0)
A18322336	A18398312	Intravascular diastolic (LNC:11377-9-89281-0)
A18322336	A18127693	Intravascular diastolic (LNC:11377-9-89281-0)
A18322336	A18282194	Intravascular diastolic (LNC:11377-9-89281-0)
A18322336	A18205078	Intravascular diastolic (LNC:11377-9-89281-0)
A18322336	A18205077	Intravascular diastolic (LNC:11377-9-89281-0)

The CPT4 hierarchy for procedures also has a dozen records with multiple parents.

I suppose that the easiest way to handle this will be to select 1 parent for each record that has multiple parents, and disconnect the record from the remaining ones. All records will still have a path to the root, because each record (except the root) will have at least 1 parent.

But concept domains with vocabularies that are represented as DAGs (instead of as trees) could be represented by breaking LeafDB.app.Concept into two tables, one that represents the concepts and another that represents their parent - child relationships. In this approach the Leaf GUI could continue to represent concept domains as trees, and astute users could note that some concepts could be reached by multiple paths in the tree.