Automated clinical concept mapping using SNOMED
A graph-based clinical concept mapping algorithm maps ICD-9 (International Classification of Disease, Revision 9) and ICD-10 (International Classification of Disease, Revision 10) codes to unified Systematized Nomenclature of Medicine (SNOMED) clinical concepts to normalize longitudinal healthcare data to thereby improve tracking and the use of such data for research and commercial purposes. The graph-based clinical concept mapping algorithm advantageously combines a novel graph-based search algorithm and natural language processing to map orphan ICD codes (those without equivalents across codebases) by finding optimally relevant shared SNOMED concepts. The graph-based clinical concept mapping algorithm is further advantageously utilized to group ICD-9/10 codes into higher order, more prevalent SNOMED concepts to support clinical interpretation.
1. A computing device comprising:
one or more processors; and
one or more hardware-based computer-readable memory devices storing instructions which, when executed by the one or more processors, cause the computing device to:
map International Classification of Disease, Revision 9 (ICD-9) codes and International Classification of Disease, Revision 10 (ICD-10) codes to ICD-Systematized Nomenclature of Medicine Clinical Terminology (SNOMED CT) concepts by obtaining and using external SNOMED CT mappings;
generate a directed acyclic graph (DAG) of the SNOMED CT concepts comprising the ICD-9 and ICD-10 codes mapped thereon:
identify one or more of the ICD-9 codes and ICD-10 codes unpaired with a shared SNOMED CT concept on the DAG;
identify, for each of the unpaired ICD-9 codes and ICD-10 codes, a common SNOMED CT concept on the DAG;
determine, for each of the unpaired ICD-9 codes and ICD-10 codes having a common SNOMED CT concept, a semantic similarity of clinical descriptions therebetween; and
pair one or more of the identified unpaired ICD-9 codes and one or more of the identified unpaired ICD-10 codes based on their respective common SNOMED CT concept and their respective semantic similarity.
2. The computing device of claim 1 in which the instructions further cause the computing device to utilize a neural network to identify pairs of the ICD-9 and ICD-10 codes according to semantic similarity of clinical descriptions, wherein the ICD-9 codes include 5 characters and the ICD-10 codes include 7 characters.
3. The computing device of claim 1 in which the instructions further cause the computing device to obtain the ICD code to SNOMED mappings from an external system.
4. The computing device of claim 1 in which the external system provides Observational Medical Outcomes Partnership (OMOP) ICD-9 to SNOMED and ICD-10 to SNOMED mapping data and SNOMED directed acyclic graph (DAG) data.
5. The computing device of claim 1 in which the instructions further cause the computing device to produce mappings of the ICD-9 and ICD-10 codes to SNOMED concepts whereby multiple ICD codes are rolled up into a single SNOMED concept.
6. The computing device of claim 5 in which the instructions further cause the computing device to rollup the SNOMED concepts using the directed acyclic graph (DAG).
7. The computing device of claim 6 in which the rollup comprises rolling up each ICD code that is below a predetermined threshold until each ICD code is aggregated in a SNOMED concept that is above the predetermined threshold.
8. The computing device of claim 7 in which the instructions further cause the computing device to delete rolled-up SNOMED codes.
9. The computing device of claim 8 in which the instructions further cause the computing device to keep redundant rolled-up SNOMED codes.
10. The computing device of claim 1 , the clinical descriptions of the ICD-10 codes semantically match the clinical descriptions of the respective ICD-9 codes.
11. The computing device of claim 1 , wherein identifying the one of the one or more ICD-10 codes comprises searching a predetermined number of the ICD-10 codes.
12. The computing device of claim 1 , wherein identifying the one of the one or more ICD-10 codes having clinical descriptions semantically similar to the respective ICD-9 code comprises identifying the one of the one or more ICD-10 codes having a cosine similarity to the respective ICD-9 code.
13. The device of claim 1 , wherein each of the unpaired one or more ICD-9 codes is mapped to a SNOMED CT concept lacking a corresponding mapped ICD-10 code, and wherein each of the unpaired one or more ICD-10 codes is mapped to a SNOMED CT concept lacking a corresponding mapped ICD-9 code.
14. A method for automated clinical concept mapping using SNOMED CT (Systematized Nomenclature of Medicine Clinical Terms), comprising:
obtaining mappings between ICD (International Classification of Disease) codebases and a SNOMED CT ontology that is organized as a directed acyclic graph (DAG), and using the mappings to map International Classification of Disease, Revision 9 (ICD-9) codes and International Classification of Disease, Revision 10 (ICD-10) codes to SNOMED CT concepts on the DAG;
identifying one or more of the ICD-9 codes and ICD-10 codes unpaired with a shared SNOMED CT concept on the DAG;
identifying, for each of the unpaired ICD-9 codes and ICD-10 codes, a common SNOMED CT concept on the DAG;
determining, for each of the unpaired ICD-9 codes and ICD-10 codes having a common SNOMED CT concept, a semantic similarity of clinical descriptions therebetween; and
pairing one or more of the identified unpaired ICD-9 codes and one or more of the identified unpaired ICD-10 codes based on their respective common SNOMED CT concept and their respective semantic similarity.
15. The method of claim 14 in which the pairing of codes comprises for queries, identifying all ancestors in the DAG as pivot nodes,
for all target nodes, identifying all possible ancestors,
identifying target node ancestors that are pivots to queries such that queries are linked to targets by the pivot nodes, and
utilizing search paths by linking to pivots having lower positions in the DAG.
16. The method of claim 15 in which the identified ancestors for the target nodes include ancestors that are within a predetermined minimum separation distance on the DAG.
17. The method of claim 14 in which each of the previous and subsequent ICD codebases include one of ICD-9 (International Classification of Disease, Revision 9), ICD-10 (International Classification of Disease, Revision 10) or ICD-X (International Classification of Disease, Revision 10) in which X is a variable having an integer value.
18. One or more computer-readable media storing instructions which, when executed by one or more processors in a computing device cause the computing device to:
obtain mappings between ICD (International Classification of Disease) codebases and a Systematized Nomenclature of Medicine Clinical Terminology (SNOMED CT) ontology that is represented as a directed acyclic graph (DAG), and using the mappings to map International Classification of Disease, Revision 9 (ICD-9) codes and International Classification of Disease, Revision 10 (ICD-10) codes to SNOMED CT concepts on the DAG;
identify one or more of the ICD-9 codes and ICD-10 codes unpaired with a shared SNOMED CT concept on the DAG;
identify, for each of the unpaired ICD-9 codes and ICD-10 codes, a common SNOMED CT concept on the DAG;
determine, for each of the identified one or more unpaired ICD-9 codes and ICD-10 codes having a common SNOMED CT concept, a semantic similarity of clinical descriptions therebetween; and
pair one or more of the identified unpaired ICD-9 code codes and one or more of the identified unpaired ICD-10 codes based on their respective common SNOMED CT concept and their respective semantic similarity.
19. The one or more computer-readable media of claim 18 in which vocabulary concepts are provided by an Observational Medical Outcomes Partnership (OMOP).
20. The one or more computer-readable media of claim 18 in which the instructions further cause the computing device to search for pairings in both directions using query=ICD-9, target=ICD-10 and query=ICD-10, target=ICD-9.
21. The one or more computer-readable media of claim 20 in which the instructions further cause the computing device to keep pairings that are within a predetermined radius on the DAG.
22. The one or more computer-readable media of claim 18 in which the instructions further cause the computing device to reduce entries in a table of pairings by keeping pairings having a shortest distance on the DAG or by keeping pairings that are determined to be semantically similar.
23. The one or more computer-readable media of claim 18 , wherein the ICD-10 codes are a predetermined number of nodes within a fixed search radius.
24. The one or more computer-readable media of claim 23 in which the instructions further cause the computing device to produce an output table of ICD code pairs that supports one of ETL (extract, transform, and load) functionality by including ICD→SNOMED mappings for both ICD-10 and ICD-9 codebases that are usable during ETL to convert the ICD codes to SNOMED, or non-ETL functionality.