Automated Mapping of Pharmacy Orders from Two Electronic Health Record Systems within the STRIDE Clinical Data Warehouse Penni Hernandez, N.D., R.N., Tanya Podchiyska Susan Weber, Ph.D., Todd Ferris, M.D., M.S. Henry Lowe, M.D. Center for Clinical Informatics, Stanford University, Stanford CA © 2009. All rights reserved. Outline STRIDE Overview Problem Definition Strategy Terminology Standard Selection Methodology Matching Algorithm Applied Examples Use of SNOMED-CT® for drug classification Evaluation Methodology Results Conclusion 11/06/09 Stanford University Medical Center A complex research-intensive organization Lucile Packard Children’s Hospital (LPCH) Stanford Hospitals & Clinics (SHC) Stanford School of Medicine Two different EHR systems Cerner at LPCH, Epic at SHC Clinical researchers expect an integrated and coherent environment that supports their needs 11/06/09 STRIDE Overview Stanford Translational Research Integrated Database Environment (STRIDE) Standards-based research and development project supporting clinical and translational research Clinical data warehouse (CDW) containing clinical information on over 1.3 million patients 11/06/09 STRIDE Overview 11/06/09 Problem Definition Users expect an integrated environment and the ability to search medications by Generic Ingredient, Brand Name and Drug Classification Hospitals operate different EHR systems with different drug information providers SUMC hospitals are cooperating to share content but their pharmacy data is not interoperable Solution needed to be rapidly implemented, 11/06/09 Strategy Define research use cases and requirements for data retrieval and interoperability Merge the pharmacy order data into a single standards-based model that will support integrated representation Goal was to achieve complete mapping of each hospital pharmacy order to a RxNorm Ingredient (IN) 11/06/09 Terminology Standards Selection Selected RxNorm for drug representation Full model with explicit relationships Interlingua between source vocabularies (e.g. SNOMED-CT®) Robust coverage of medications in U.S. No license fees and actively maintained by NLM Selected SNOMED-CT® for drug classification Selected FDA Structured Product Labeling (SPL) data standard for Route of 11/06/09 Drug Representation Strategy 11/06/09 Methodology Selected HL7 v2.3 SUMC Pharmacy Order (RDE) messages as the initial source of data Targeted segments: Pharmacy Encoded Order (RXE) segment Pharmacy Component (RXC) segment Pharmacy Route (RXR) segment Developed an algorithm using PL/SQL to match message segments to RxNorm atoms of type “IN” 11/06/09 Data Elements Header Data HL7 v2.x Pharmacy Order Message MSH|^~\&|EPIC|PHARM|STRIDE|SOM|20050403001656|S3820351 |RDE^O01|4830577|P|2.4|||^M PID|||9999999^^^SHC^SMRN||HERNANDEZ^PENNI^M||19451120| F|||8472 Dartmouth^Boulder^CO^80305^^^^||(303)5555555||E|MINXX|03822739-203|999-99-999||||^M PV1|||POINT OF CARE^ROOM^BED| ORC|NW|8392009287^|||||^EVERY 3 HOURS PRN^^ 2080726000^^ Norm^PRN||20080724|||||||||^Hernandez^Penni^M^^^^^SH C^^^^MSPV||(303)555-5555 |||||||STANFORD HOSPITAL|||^M RXE|^ONCE&^X1^200901210030^200901210004^Fax^|1834^IBUP ROFEN 100 MG/5ML PO SUSP^ADS|95||3^mg|Shake well. Give with food or milk.|||1|||BV9999999^ Hernandez^Penni^^^||||||||||||||||||^ADVIL, MOTRIN^M RXR||PO^Oral|||^M RXC||1834^IBUPROFEN 100 MG/5ML PO SUSP^ADS|5mL|||^Advil, MOTRIN^M^M 11/06/09 Matching Algorithm Algorithm Input RXE and RXC data fields Give Code – local identifier for the drug order Give Text – drug name, form and strength Give Alt Text – alternate representation of “Give Text” Multi-ingredient drug orders are split by a series of delimiters before a match is attempted 11/06/09 Matching Algorithm Ignoring case, compare only alphanumeric characters to the “STR” column in the RxNorm concept table “RXNCONSO” If a match is found, then the RXNREL table is used to navigate to the RXCUIs for the clinically active ingredients If a match is not found, then the last word in the original string is stripped and another attempt is made for matching 11/06/09 Branded Pack Example Hospital Input String: LO OVRAL-28 TABLET loovral28tablet (no match) loovral28 (match for RXCUI = 749787) Identified BPCK: {21 (Ethinyl Estradiol 0.03 MG / Norgestrel 0.3 MG Oral Tablet) / 7 (Inert Ingredients 1 MG Oral Tablet) } Pack [Lo/Ovral 28 Day] "has_tradename" GPCK: {21 (Ethinyl Estradiol 0.03 MG / Norgestrel 0.3 MG Oral Tablet) / 7 (Inert Ingredients 1 MG Oral Tablet) } Pack 11/06/09 Branded Pack Example "contained_in" SCDs: Ethinyl Estradiol 0.03 MG / Norgestrel 0.3 MG Oral Tablet Inert Ingredients 1 MG Oral Tablet "constitutes" SCDCs: Ethinyl Estradiol 0.03 MG Inert Ingredients 1 MG Norgestrel 0.3 MG 11/06/09 Branded Pack Example "ingredient_of" INs: Ethinyl Estradiol (rxcui= 4124, SNOMED CT concept_id=15432003) Inert Ingredients (rxcui=748794) Norgestrel (rxcui=7518, concept_id=82240008) SNOMED drug classes: IS-A Progestin preparation (product) Oral contraceptive preparation (product) Contraceptives (product) Reproductive system drug (product) Estradiol preparation (product) Sex hormone product (product) Hormone preparation (product) Hormones, synthetic substitutes and antagonists (product) 11/06/09 Use RXCUIs to Traverse to SNOMED-CT® •RXCUI unifies all strings representing Hydrocodone no matter what the content source •Leverage mapping to navigate from RxNorm to SNOMED-CT® 11/06/09 Evaluation Methodology •Found ingredient, verified correct = TP •Found ingredient, verified incorrect = FP •Found no ingredient, verified correct = TN •Found no ingredient, verified incorrect = FN 11/06/09 Evaluating Multiple Ingredient Drugs •Found both ingredients, verified correct = TP •Found only one ingredient, verified incorrect = FP •Found both ingredients but one was incorrect verified incorrect= FP •Found neither ingredient, verified incorrect = FN 11/06/09 Results LPCH % Total Number 8895 93.28 True Negative 316 False Positive 305 True Positive Total Number •Algorithm correctly False 20 Negative Pharmacy Orders SHC % Total Number % 6665 92.70 2230 95.06 3.31 270 3.76 46 1.96 3.20 240 3.34 65 2.77 mapped approximately 0.21 15 0.21 93% of5SUMC 0.21 •No suitable RxNorm concept could be found for 3% of SUMC Pharmacy Orders •Approximately 4% required algorithm adjustments 11/06/09 Characterization of True Negatives Ambiguity in Original Message Outside the Scope Limitation of of RxNorm Algorithm Non Drug Orders •Seasonal vaccine •Total parenteral nutrition •Peritoneal dialysis solution •Iron, no salt •Vitamins (> 4 ingredients) •Over the counter (OTC) product •Investigational Drug •Non-US Drug •Dummy Order •Devices •Supplies •Non Drug Kit •Valid Abbreviation •Fragment, beyond machine readability •Local name for a custom compound 11/06/09 New Pharmacy Orders by Month 10400 •Algorithm is in production 10300 10200 10100 New Mapped Orders New Unmapped Orders Reviewed Orders 10000 9900 9800 9700 9600 July August September 11/06/09 •Ongoing review of new Pharmacy Orders takes less than 4 hours/month Value Proposition to STRIDE Utilization of linkages within RxNorm Brand Name (BN) to Ingredient (IN) Ingredient (IN) to SNOMED-CT® Product Aligned with emerging standards Flexibility to incorporate other drug information sources (e.g. Clinical documents) Support search by drug concept Potential interoperability with other data 11/06/09 More Information Stanford Center for Clinical Informatics http://clinicalinformatics.stanford.edu Contact information: Penni.Hernandez@stanford.edu RxNorm http://www.nlm.nih.gov/research/umls/rxnorm/ RxNav http://mor.nlm.nih.gov/download/rxnav/ 11/06/09