Notes

Outline

NIST Barriers to Electronic Medical Record Systems – the Babel of code systems August 1, 2002 Gaithersburg, MD Clement J. McDonald, M.D.

Why banking is not a good example Banks  track one fungible quantity – dollars. Health care deals with tens of thousands of quantities Health care tracks many thousands of quantities. Banking only worries about the account - - no need to link the many accounts of the account holder. Banking uses “only” numbers.  Health care uses, numbers, narrative, drawings, images, tracings, video.

Why electronic medical records are hard The patient’s whole picture is divided into many isolated parts The parts are separate by organization and and computer system vendors. Variations in what is available in electronic form by time and place (institution) Variation in the degree of structure and coding – (and thus the functions that can be automated)

Slide 4

What separates the parts Entropy– : Each institution invents their own patient identifiers. Each institution (and often) each system within an institutions invents its own provider numbers- so hard to verify rights of access, to deliver information to right providers Different conceptualization of data over time and place Different codes for same information over time and place

Where we have to overcome entropy Patient identifiers –so we can cross institutions Provider identifiers – so we can deliver information, study health care system and control access Messages/models – So we can move data to where it is needed. Observation and concept identifiers- so we can do more than display data

Why hard-more Data can be represented in different levels of structure Coded & structured (like a UB 92) Narrative ( dictation) Scanned document (like a FAX) Different uses have different requirements for structure

Misperceptions Medical Records produce no patient data They are just containers. Other systems/Sources produce every thing First we need standards for individual clinical chunks ( we have that ) The medical record is subservient to the message.

Mis assumptions The medical record (paper or electronic) is not a cornucopia. It does not (can not) carry every variable anyone would ever want. (Much of what people can imagine they would like is not every recorded) That is why God made prospective research. Data collection has to vary by patient, context, visit Each data item costs. We have to choose ( Ottawa knee rule) We have almost zero information about the relative value of different pieces of information  under different contexts The need to collect time costs of data collection and measure the value (predictive etc) is not on anyone’s agenda.

Why Hard – still more Different users have different (and fairly unbending ) conceptualizations of the data with different degrees of abstraction. * Flat versus Stacked data

Flat Data Set Analytic Conceptualization

Stacked Data Set Application Conceptualization Kind of structure used for laboratory, pharmacy, billing and electronic medical record system

Focus on coding Remember that what is a data (a code in a data) base and what is structure ( a field in the data base definition)-can change depending on the data structure Need to firm up the data structure to decide about code requirements Recommend an HL7-world view. Will LOINC as an illustration

Major Code Systems ICD9 (ICD-O) - http://www.mcis.duke.edu/standards/termcode/icd9/1tabular.html IDC10 - http://www.who.int/whosis/icd10/ CPT - http://www.ama-assn.org/ama/pub/category/ 3113.html LOINC - http://www.Regenstrief.org/loinc/ SNOMED - http://www.snomed.org/main.html NDC (NDC - new) - http://www.fda.gov/cder/ndc/

Other Codes Medispan GPI Code - http://www.mcis.duke.edu/standards/HL7/pubs/version2.3/html/ch700005.htm MEDCIN - http://www.accuchart.net/Medcin.html MedDRA - http://www.meddramsso.com/ Clinical drug - http://www.fda.gov/cder/regulatory/ersr/ECGdata.htm

Kinds of codes Question codes (variables) Glucose concentrate? Glascow Coma score? Discharge dx? Answer codes (concepts) Numeric Codes Free text Package codes Lytes, ESRD, Study 1325

LOINC – codes for variables/Questions LOINC provides a universal identifier for  observations, to eliminate the entropy in in identifying variables Provides a universal ID for HL7 OBX field #3 (Observation ID)

LOINC codes batteries *, and Questions but not  Answers Batteries - (Packages ) (OBR-4) SF 36 survey Comprehensive metabolic panel Questions  (OBX-3) (variables/observations ) Serum glucose Blood culture result  Values (OBX-5) numbers  * codes (eg E coli)

What is the LOINC database? Database of 30,000 observations & some batteries Six part formal name LOINC code with check digit Mapping/browsing program (RELMA) Copyright: LOINC is free for all uses Available on internet: http://www.regenstrief.org/loinc

Important facts In 1995 LOINC 1.0  contained 4000 Clinical/26000 Lab LOINC codes In 2002 LOINC contained 31,544 entries LOINC list servers at www.hl7.org Download files and tools from Regenstrief Web Site www.regenstrief.org/loinc RELMA – a tool for mapping local codes to LOINC

Where does LOINC fit in HL7? HL7 OBX message structure Where LOINC Fits in OBX-3 ID & name = Troponin-I OBX-5 Value = 5 OBX-6 Units = ng/nl OBX-7  Normal range = 0-1.3 OBX- 8 Normal flag = H OBX-15 Producer  = ACME lab

Where does LOINC fit in HL7? OBX for serum Troponin-I and a CK MB

LOINC Copyright No Cost in perpetuity Prevents multiple variants Those who distribute in a database must include the copyright notice, all six parts of the name and the short name, and the copyright notice.

LOINC Media Formats – on CD & Web Site PDF – report format ASCII Tab delimited AccessTM

Success: Government and Miscellaneous Centers for Disease Control and Prevention Communicable disease reporting + NEDSS DEEDS emergency database Tumor registries HIPAA claims attachment transaction Veterans Administration Medical Record (all labs) Big Pharma: CDISC vs 2.0 for  FDA submissions

Success: Big Laboratories Endorsed by American Clinical Laboratory Association (ACLA) Laboratory Corporation of America Quest Diagnostics Incorporated Focus laboratories All 26 US Veterinary Labs

Success: Care Systems Kaiser Foundation Health Plan, Inc. Intermountain Health Care, Utah Partners of Boston Care Group of Boston Clarian/Indiana University Columbia Presbyterian, New York Hospital for Sick Children, Toronto

Success: HMOs Aetna Empire Blue Cross Independence Blue Cross in Philadelphia Harvard Pilgrim Health Care

New trend: Instrument Vendors VA – may  require instrument messages to contain LOINC codes * Dade Microscan (antibiotic susceptibilities) Beckman/Coulter Roche Diagnostics (Boehringer Mannheim) More than 15 vendors in CAP Today surveya

International Australia Brazil Estonia Germany - adopted by DIN New Zealand Ontario & British Columbia, Province-wide Korea Switzerland’s laboratory quality assurance project (CUMUL)

Slide 31

LOINC Language Translations EXISTING For commonest tests Swiss (CUMUL) Geneva English to German French Italian Working on CEMIC School of Medicine, Argent Spanish ABNT/CB-36 Comitê Brasileiro de Análises Clínicas e Diagnóstico In Vitro Portuguese

Areas of Development Laboratory Clinical HIPAA Attachment

LOINC Lab Scope

Clinical LOINC Subject Areas Vital Signs Hemodynamic measures Fluid Intake/Output Body Measurements Emergency Department Respiratory Therapy Document sections Standard survey instruments Ophthalmology measurements EKG (ECG) Cardiac Ultrasound Obstetrical Ultrasound Discharge Summary History & Physical Pathology Findings Colonoscopy/Endoscopy Radiology reports Clinical Documents Tumor Registry

RELMA Regenstrief Manual mapping assistant Same free use as LOINC Used to browse the LOINC file Used to link local observation codes to LOINC codes for transmission

Slide 37

RELMA – Collapsed grid by component system and more

Slide 39

What\to Do next Use what we have Separate infrastructure from messages Concentrate on the data chunks- Not the ultimate application (may or may not be able to standardize applications)-Have to standardize the interface/messages Force semantic conformance. Values go in value field. Units go in units field, etc Deal with the four content  barriers

What to do next

What to do next Find – simplest , secure messaging (not tied to one industries message Universal person, (provider), place, organization identifiers Patient identifier ? Converge and support  , non encumbered, universal codes for major categories of clinical variables, concepts.