Automated mapping of Electronic Health Records to document off-label drug use. There are thousands of off-label medications that are being prescribed to patients on a daily basis. For this project, we have developed an automated method to detect and analyze off-label prescription pattern by comparing diagnosis codes and approved indications for prescribed medications within the MIMICIV dataset.
We utilize OMOP vocabulary to map nonstandard diagnosis codes to their standard SNOMED equivalent. For medications, we map NDC codes to their RxNorm equivalent, then to SNOMED to ensure that we are capturing the drug ingredient. We leverage the hierarchical vocabulary structure to compare indication matchings between:
1. One-to-one Nonstandard ICD to standard SNOMED diagnosis mapping,
2. Horizontal mapping from Standard SNOMED to other non-standard SNOMED, and
3. Hierarchical mapping from Standard SNOMED to its parents and children vocabulary.
-
Drug Central : A database that includes formal, government-approved indication information, as well as known off-label and contraindications.
-
AACT : Clinical trial database that contains information on all clinical trials and their underlying conditions that use a particular drug as intervention / treatment.
-
NDF-RT : Included in OMOP vocabulary. Developed by the Veterans Health Association, and includes indication relationships within the vocabulary.
MIMICIV is a de-identified, publicly available dataset originated from intensive care units at the Beth Israel Deaconess Medical Center (BIDMC). MIMICIV uses ICD 9 / 10 (CM) as their diagnosis codes, and NDC codes for medication prescriptions. Each patient is given a unique subject_id, and a unique hadm_id for each visit. We extract all ICD diagnoses and NDC medications for each subject_id and hadm_id, and map them both to SNOMED for convenience of vocabulary traversal.
MIMICIV also includes patient demographic and insurance information. We use this information to conduct preliminary implication analysis based on our results.
off_label/mapping
off_label.py : Extracts relevant data from MIMICIV publicly available clinical data, which can be accessed at https://physionet.org/content/mimiciv/0.4/{.uri}.
vocab_mappings.py : Python script that defines functions that extract standard to non-standard terms or vice versa. Used to convert from ICD to SNOMED, and to MeSH codes.\
off_label/Results
Finalized summary of Drug Central, AACT, and NDFRT analysis. Includes a powerpoint presentation that summarizes our project. figures/ include visualizations of our result, and our initial implication analysis.