James J. Cimino Columbia University MIE ‘02 Budapest, Hungary August 27, 2002 The Challenge of Reuse of Information
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Data Types Text Numeric SignalStructuredCodedStandard Coded NLP Interpretation Image Blobs Symbols
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Information Reuse InformationResearchOther Clinicians Summary Hospital Administration Government Decision Support
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Information Mismatch Form Meaning Language Granularity Semantics Version
Information Mismatch: Form
Information Mismatch: Meaning “Paget’s Disease” “of the bone” Paget’s Disease of the Breast?!?!
Information Mismatch: Language “Tüdőgyulladás” Pneumonia?
Information Mismatch: Granularity “Goodpasture’s Syndrome” Does the patient have lung disease?
Information Mismatch: Semantics AMP Sens. Test = 1:2 Should I prescribe “Ampicillin 250mg Caps”?
Information Mismatch: Version Patient has hantavirus infection “Virus, NEC” 2001 ICD: - Smallpox - Cowpox - Virus, NEC 2002 ICD: - Smallpox - Cowpox - Hantavirus - Virus, NEC
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Terminology Solutions Standards Distribution Semantic Representation
Terminology Solutions: Standards Advantages –Less duplication of work –“Plug and play” compatibility Disadvantages –Cost of adoption –Unresponsive to change –Developers Users
Terminology Solutions: Distribution Media –9-track tape –Floppy disks –CD-ROM –Web Models –ICD: annual –UMLS: change files –HL7: server
Terminology Solutions: Semantic Representation Concept oriented Concept permanence True is-a hierarchies Multiple hierarchies (heterarchy) Semantic relationships Inheritance
Terminology Solutions: Semantic Representation Goodpasture’s Syndrome Kidney Disease Hemoptysis Hematuria Finding Lung Kidney Organ has-site Lung Disease is-a has-finding
Semantic Representation: Galen Structured Meta Knowledge from Pen&Pad Common Reference Terminology Requires terminology server Automated classification Open source terminology
Semantic Representation: Galen Fracture which < hasLocation Bone hasCause Condition> Fracture which < hasLocation (AnatomicalNeck which isDivisionOf Femur) hasCause (Osteoporosis which hasCause PostMenopausalChange)> Can be classified as: Fracture Fracture which hasLocation LongBone. Fracture which hasLocation (AnatomicalNeck which isDivisionOf LongBone). Fracture which hasLocation Thigh. Fracture which hasLocation Hip. Lesion which isCausedBy Osteoporosis. Lesion which isCausedBy PostmenopausalChange.
Semantic Representation: SNOMED-CT Merger of SNOMED and Read Clinical Terms Reference terminology Many domains Heterarchy Semantic relations (roles) Postcoordination >300,000 concepts
Semantic Representation: SNOMED-CT is-a Bacterial Pneumonia Tularemia Pulmonary Tularemia has-causative-agent Francisella tularensis Lung Structure has-finding-site Inflammation associated- morphology
Semantic Representation: LOINC Logical Observations, Identifiers, Names and Codes Codes for observations in HL7 messages Fully-specified names Codes for orderable observations Codes for results
Semantic Representation: LOINC | URINALYSIS PANEL | COLOR | COLOR | PT | UR | NOM | GLUCOSE | SCNC | PT | UR | QN | TEST STRIP Yellow Red Colorless …
Semantic Representation: Drugs Food and Drug Administration Veterans Administration National Library of Medicine Drug knowledge base vendors Common model for Clinical Drug RxNorm
Semantic Representation: Drugs Clinical Drug Ingredient Class Ingredient is-a Chemicals Drug Class Not-Fully-Specified Drug is-a Medications International Package Identifiers Country-Specific Packaged Product is-a Packages Trademark Drug Manufactured Components is-a Composite Clinical Drug is-a Composite Trademark Drug is-a
Semantic Representation: MED Medical Entities Dictionary Data dictionary and controlled terminology Columbia-Presbyterian Medical Center Heterarchy Semantic network Multiple domains >70,000 concepts
Semantic Representation: MED Laboratory Procedure CHEM-7 Plasma Glucose Test Substance Sampled Part of Has Specimen Substance Measured Medical Entity Event Laboratory Test Diagnostic Procedure Substance Bioactive Substance Glucose Chemical Carbo- hydrate Laboratory Specimen Plasma Specimen Plasma Anatomic Substance
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Matching Granularity and Semantics Gentamicin Injectable Gentamicin Gentamicin Sensitivity Test Serum Gentamicin Level is-a Intravascular Gentamicin Tests Gentamicin Toxicity Etiology Measures Sensitivity Substance Measured Has ingredient Summary Reports Decision Rule Expert System Drug Information
Example of Reuse: Summary Reporting Spreadsheets for trends in lab data Defined as concepts in the MED Linked to test classes
Example of Reuse: Summary Reporting Plasma Glucose Test Serum Glucose TestFingerstick Glucose TestLab Test Intravascular Glucose TestChem20 Display Lab Display
Example of Reuse: Summary Reporting
Example of Reuse: Merging Data Merger between Presbyterian Hospital and New York Hospital Separate departmental systems Common repository Merger of terms in MED allows cross- institution data aggregation
Example of Reuse: Merging Data Diazepam 5 mg Tablet CPMC Drug: Diazepam 5 mg Tab CPMC Drug: UD Diazepam 5 mg Tab CPMC Drug: UD Diazepam 5mg Tab CPMC Drug: UD Diazepam 5 mg Tab Cerner Drug: Diazepam Tab 5 mg Diazepam Tablets Diazepam Preparations Benzodiazepine Preparations Drug Enforcement Administration (DEA) Class IV - Drug with Low Abuse Potential Drug Allergy Class: Benzodiazepines Tablet
Example of Reuse: Merging Data Plasma Glucose Measurement Intravascular Glucose Test Plasma Chemistry Test CPMC Laboratory Test: Glucose Tolerance, 6hr CPMC Laboratory Test: Glucose Tolerance, Fasting CPMC Laboratory Test: Glucose, 1/2 Hour CPMC Laboratory Test: Glucose, Fasting NYH Lab Procedure: Glucose, Plasma NYH Lab Procedure: Glucose, 0 H NYH Lab Procedure: Glucose, 2 PP NYH Lab Procedure: Glucose, 0.5 H NYH Lab Procedure: Glucose, 1 H NYH Lab Procedure: Glucose, 2 H NYH Lab Procedure: Glucose, 3 H NYH Lab Procedure: Glucose, 1.5 H NYH Lab Procedure: Glucose, 4 H NYH Lab Procedure: Glucose, 5 H NYH Lab Procedure: Glucose, 6 H NYH Lab Procedure: Ogtt,Gest Screen,(50g) Presbyterian Plasma Glucose Test Presbyterian Plasma Glucose Measurement Allen Plasma Glucose Measurement New CHEM-7 Plasma Glucose Measurement CPMC Laboratory Test: Old Plasma Glucose Measurement CPMC Laboratory Test: Glucose, Challenge CPMC Laboratory Test: Glucose, Fasting CPMC Laboratory Test: Glucose, 1hr Post Prandial CPMC Laboratory Test: Glucose, 2hr Post Prandial CPMC Laboratory Test: Glucose, Random CPMC Laboratory Test: Glucose CPMC Laboratory Test: Glucose Tolerance, 1hr CPMC Laboratory Test: Glucose Tolerance, 2hr CPMC Laboratory Test: Glucose Tolerance, 3hr CPMC Laboratory Test: Glucose Tolerance, 4hr CPMC Laboratory Test: Glucose Tolerance, 5hr
Example of Reuse: Automated Decision Support Data stored in repository reviewed in real time Arden Syntax rules triggered by data Generation of alerts and reminders High-level concepts in rules map to low-level concepts in database
Automated Decision Support: Tuberculosis Monitors for delayed culture results Sends message if result not equal to the code “No growth” One day, dozens of alerts about positive results but no organism was reported What happened?
How the Lab Fooled the Alert Alert looked for results = “No Growth” Lab started reporting “No Growth to Date” “No Growth to Date” “No Growth” Solution: Use the controlled terminology to map all No-Growth-like lab terms into a single class, and have the alert logic refer to the class.
Automated Decision Support: Tuberculosis No Growth Medical Logic Module No Growth to Date
No Growth after... How We Outsmarted the Lab No Growth No Growth after 48 Hours No Growth after 72 Hours “No Growth” Results No Growth after 24 Hours No Growth to Date Medical Logic Module
Example of Reuse: Information Retrieval Understand Information Needs 1 Get Information From EMR 2 Automated Translation 5 Resource Terminology 4 Presentation 7 Querying 6 Resource Selection 3
Example of Reuse: Expert Systems Expert system has high-level concepts Database has quantitative results Semantic mismatch Translation through semantic net traversal
Example of Reuse: Expert Systems L 1600 Gluc32703 Serum Glucose Tests Intravascular Glucose Tests Elevated Abnormal Finding in Body Substance Decreased Abnormal Finding in Body Substance Hyperglycemia 3286 Hypoglycemia Intravascular Specimen Glucose Abnormal Level of Blood Glucose 3286 Hypoglycemia
Expert System: DXplain
Expert System: Lipid Guideline
Lab :1/1/99 Cardiac Enzyme Test Radiology :2/23/99 Chest X Ray Radiology :2/28/96 Head CT Lab :12/28/96 Sickle Cell Test Admission :3/14/96 Stroke Admission :2/14/98 Angina Lab :1/1/99 Blood Type Test Radiology :2/1/97 Knee X Ray Discharge :1/15/99 CHF Medical Record Example of Reuse: Problem-Oriented Views Chest X ray Intravascular CK Test Creatine Kinase Chest X ray 2 View Cardiac Enzyme Congestive Heart Failure Angina Heart Disease Chest Admission :2/14/98 Angina Lab :1/1/99 Cardiac Enzyme Test Radiology :2/23/99 Chest X Ray Discharge :1/15/99 CHF Heart MED
Experience with Information Reuse Summary reporting Merging data Automated decision support Information retrieval Expert systems Problem-oriented views
Overview Data types Information reuse Information mismatch Terminology solutions Experience Conclusions
Information, Then and Now - Cimino JJ, Int J Biomed Comput. 1994; 34: Discharge Diagnoses Radiology Reports Physical Exams Discharge Summaries Patient Histories Medication Lists Uncoded, Unstructured Uncoded, Structured Locally Coded Universally Coded Then (1993)
Information, Then and Now - Cimino JJ, Int J Biomed Comput. 1994; 34: Discharge Diagnoses Radiology Reports Physical Exams Discharge Summaries Patient Histories Medication Lists Uncoded, Unstructured Uncoded, Structured Locally Coded Universally Coded Near Future
Information, Then and Now - Cimino JJ, Int J Biomed Comput. 1994; 34: Discharge Diagnoses Radiology Reports Physical Exams Discharge Summaries Patient Histories Medication Lists Uncoded, Unstructured Uncoded, Structured Locally Coded Universally Coded Far Future
Current Status Discharge Diagnoses Radiology Reports Physical Exams Discharge Summaries Patient Histories Medication Lists Uncoded, Unstructured Uncoded, Structured Locally Coded Universally Coded
Current Status Uncoded, Structured Standard Semantic Terminology Standard Code Set Discharge Diagnoses Radiology Reports Physical Exams Discharge Summaries Patient Histories Medication Lists Uncoded, Unstructured
Current Status Discharge Diagnoses Laboratory Reports Problem Lists Text Reports Text Reports Medication Lists Uncoded, Unstructured Uncoded, Structured Standard Semantic Terminology Standard Code Set
Conclusions Advanced health care means information reuse Semantic-based terminologies support reuse Terminologies are moving in the right direction