Analysis Normal Form A Standard Model for Aggregating Clinical Data - - PowerPoint PPT Presentation

Keith E. Campbell, MD, PhD1, Tim Williams, BS2, Raja A. Cholan, MS2

(1) Veterans Health Administration (2) Deloitte Consulting LLP

Analysis Normal Form

A Standard Model for Aggregating Clinical Data for Analysis

The Current Health IT Landscape

A major contributing factor to high medical errors is inaccuracy in patient health care data in electronic health records (EHRs) due in part to a variety of clinical data models that exist.

Source: Debourgh & Prion and HealthIT.gov

This inaccuracy in health care data has led to erroneous clinical decisions and patient safety issues. More than 250,000 deaths in the US are attributed to medical errors every year. There is an urgent need for widespread standard and consistent quality of health care.

What is a Clinical Statement?

• A definite and clear representation of a clinically-significant fact or

situation that was observed to exist or happened, or that a particular procedure was requested

• Expressed as a narrative that provides a written account that can be

naturally read by humans, as well as a normal form which is a machine- processable representation

Example of Variation in Current Clinical Data Models

Variation in FHIR Observation Resources

Value Code Body Site Method Device

FHIR Observation Resource (EHR 1)

Patient identifier: [Identifier] ... category: CodeableConcept code: CodeableConcept <<Brachial artery

• left Systolic blood pressure LOINC

[8546-4]>> ... value: Type[] << Quantity | CodeableConcept| 140 >> interpretation: CodeableConcept << ObservationInterpretationCodes+ >> ... bodySite: CodeableConcept << SNOMED CT brachial artery: laterality = left >> ... method: CodeableConcept << SNOMED CT Vital Signs Monitor>> ... Device: Reference[] Device | DeviceMetric

Value Code Body Site Method Device

FHIR Observation Resource (EHR 2)

Patient identifier: [Identifier] ... category: CodeableConcept code: CodeableConcept << Brachial artery Systolic Blood Pressure LOINC [8508-4]>> ... value: Type[] << Quantity | CodeableConcept| 140 >> interpretation: CodeableConcept << ObservationInterpretationCodes+ >> ... bodySite: CodeableConcept << SNOMED CT left brachial artery >> ... method: CodeableConcept << SNOMED CT Vital Signs Monitor>> ... Device: Reference[] Device | DeviceMetric

There are currently many ways to represent the same clinical statement

• Simplicity: Simplicity of the entire system instead of just one area of the system
• Consistency: Consistent representation of information that is commonly shared

across models

• Reusability: Architectural patterns should encourage class reusability where
• possible. Reusability may further refine encapsulation when composition is

considered

• Reducing Variability: Minimize arbitrary classification rules to eliminate ambiguities

to allow a single representation of clinical data

Analysis Normal Form (ANF) Design Principles

ANF’s Design Principles help standardize statement representation so that it is understandable, reproducible, and useful

• Applies a clean separation between

statement and terminology concerns

• Helps structure clinical information

so that it is consistently represented when exchanged among systems

• Normalizes a variety of clinical

statements to a common normal form

• HL7 ANF Informative Ballot was

successfully published in 2019

Analysis Normal Form (ANF)

• Topic
• Defines the action (being performed or requested) or the result of that action
• Needs to be able to exist on its own and still retain original intent and clarity of meaning
• Each clinical statement may only have one topic (but the topic is a comprehensive

expression)

• Circumstance
• HOW, WHY, WHEN, and with WHAT “result” a requested or performed action will be or

was carried out

• ANF promotes a normalized representation of observation or intervention results where all results are

represented to a "measure"

Major Components of an ANF Statement

ANF Statement as a FHIR Observation Extension

Value Code

FHIR Observation Resource (ANF)

Patient identifier: [Identifier] ... category: CodeableConcept code: CodeableConcept <<[Observation procedure] – (Method) [Examination – action]- (Has focus)[Sitting systolic blood pressure]- [Procedure site – Direct] [Structure of left brachial artery]- (Using device) [Blood pressure cuff, adult size] >> ... value: Type[] << Quantity | CodeableConcept| 140 >> interpretation: CodeableConcept << ObservationInterpretationCodes+ >>

• Data elements for ‘body site’,

‘method’, and ‘device’ may no longer be required

• Using ANF’s Topic component, this

contextual information can be modeled in FHIR’s ‘code’ data element

ANF may reduce the variability of clinical statement representations, by shifting the complexity to terminology expressions

• Clinical Decision Support
• Provide a standards-based, normalized representation of clinical statements
• Use an objective measure to help evaluate the result, presence, and magnitude of a

specific finding, request or observation

• Classify the topic of a statement using standard terminology expressions
• Data Analytics / Mining
• Create normalized data natively
• Transform representations of clinical statements to normal form (e.g. transform C-CDA

templates, FHIR profiles to ANF)

What Can ANF Be Used For?

The ability to measure and improve outcomes relies on consistent, high-quality data that is aggregated from a variety of systems

• Make ANF accessible so that statements can be represented as new FHIR

extensions to existing resources for clinical decision support

Next Steps

Thank you

• Read a copy of the HL7 ANF Informative Ballot Specification here:

http://solor.io/anf

• Contact Information
• rajkumar.cholan@va.gov
• keith.campbell@va.gov
• http://solor.io/