240 likes | 682 Views
Virtual Medical Record. HL7 UK 2003 Peter Johnson. What is a vMR?. It’s a standard INTERFACE Safe mediation interface between HL7 compliant clinical systems and decision support systems Purpose is DSS task-specific communication NOT a persistent record – on the fly translation
E N D
Virtual Medical Record HL7 UK 2003 Peter Johnson
What is a vMR? • It’s a standard INTERFACE • Safe mediation interface between HL7 compliant clinical systems and decision support systems • Purpose is DSS task-specific communication • NOT a persistent record – on the fly translation • NOT an EHR standard • Standard names for categories of things found in records (medical record classes) – e.g. “Investigation” • NOT a vocabulary – it has < 20 classes • In HL7 form a set of messages from a DMIM • Performance guarantees - QoS
vMR definition • define a shared model of "what structures are possible to query" in a way that abstract away from idiosyncrasies of a particular institution's or vendor's EHR • plus the messages/events to facilitate that
DSS KB Uses VMR terms CDSS engine Queries Writes/updates Action requests Events API or messages - using VMR objects/attributes/terms Terminology Mediation Server VMR <> host Clinical system adaptor Clinical System
Why do we need one? • Expense of DSS; maximise reuse • No human in the loop • Impedance mismatch • One to Many systems interface issue • Performance guarantees
Why? – Expense of DSS • DSS Knowledge base production very resource intensive, therefore expensive • Much of the resource use is in creating logical expressions about the patient • e.g. “if diabetic or hasLVF and systolicBP > 135 or disatolicBP > 85 then…” • Need to standardise the way these are written • Write once, reuse on all systems
Why? – No humans • Machine to machine communication here • No intelligence in AI • Can’t fudge issues expecting a human to resolve them • Everything explicit, minimise ambiguity
Why? - Impedance mismatch problem • DSS to Care Record System has impedance mismatch • Record structure mismatch • Complexity vs. Simplicity • Terminology mismatch • Specific vs. General
Why? - differing systems problem One DSS system >> many clinical systems • Differing underlying medical record models • Differing terminology • Surely HL7 messages makes this go away? • Subtle problems remain – differing institutions use same records/terms to mean different things
Differing medical record models • Many different schemata in vendor medical records • Most of their distinctions not relevant to clinical DSS • Drop administrative, audit trail, security distinctions • Define a simplified schemata – a ‘view’ vMR contains a standard simplification of record data-model • Only distinctions important to clinical decision support • Minimum number of patient-data classes for this task • Minimum number of attributes for this task • Surprising success across domains and countries • UK primary care • US ambulatory care • US secondary care • Based on real experience on real systems • UK implementation on Torex; US on VA systems
List of relational tables in one UK clinical system (145; 3-20 fields)
Terminology problem • Problem 1. Level of specificity • DSS KB mainly written in generalised/abstraction of term found in record • Coded records may be very specific, e.g. ‘prolonged ST interval on ECG’ • DSS may need only more general ‘risk of heart block’ – satisfiable by presence of hundreds of coded entries. • Mapping from specific -> general • Lessens risk of changed meaning • vMR defines messages to Terminology Server for these services • Problem 2. Blurred/variable boundary between vocab and medical record structure • e.g. Blood pressure. Is it 3 distinct records with relationships, each with terms, or is it one term with parts and values • Other problems e.g. physical units mismatch etc
Why? -performance problem • Interactive DSS requires sub second responses • Prodigy guideline system can generate several hundred queries to bring up one set of choices between actions – need to be sub-second • Exacting and technically hard to achieve, and has to influence the approach taken • Will HL7 v3 messages work? • Even harder in WAN environment • Differences between close-coupled and loosely-coupled systems • Quality of Service guarantees required
vMR Aims: • Unambiguous communication between DSS and EHR by: • Standard model of record classes • Standard names of record classes • Standard attribute names • Allow reusable logical expressions to be written ‘…observation.code == ‘SBP’ AND observation.value > 140 AND … assessment.code == ‘Diabetes’ OR assessment.code == ‘LVF’ OR…’ • Standard set of messages/functions • Standard process achieving terminology mediation • Performance/QoS guarantees
Work in HL7 • Most work so far on standardising the simplified record model for queries • Similarities with GP2GP EHRStatement noted • Effort at convergence with GP2GP • While the models may be very similar; vMR != GP2GP • Messages different • QoS guarantees
DSS KB Uses VMR terms CDSS module Queries Writes/updates Action requests Events API - uses VMR objects/attributes/terms Terminology Mediation KB VMR <> host Clinical system adaptor Clinical System
In a hypertensive patient on no antihypertensive medication initiate antihypertensive treatment at BP > 140/85 if diagnosis of diabetes OR diagnosis of LVF OR 10 year CHD risk > 30%. OR CVS complications present Aim to keep BP < 130/85. Assessment Intervention Plan Observation Assessment Assessment Assessment(DSS) Assessment Goal • i.e.‘…observation.code == ‘SBP’ AND observation.value > 140 AND … assessment.code == ‘Diabetes’ OR assessment.code == ‘LVF’ OR…’
Latest HL7 work • Reformulating same approach but based on other TC’s existing messages rather than G2GP (although this was already occurring) • Agreement to release as DSTU - draft standards for trial use rather than to ballot • But: vMR more than just the model • Terminology mediation, QoS standards • Where is the EHR framework in HL7? “rim light” • Where is the boundary between model and terminology?
Wider application • Problems with DSS and vMR are a microcosm of any (machine) agent to agent communication – this is just one early example • More exacting requirements than most of HL7 contributors appear used to: Document, human centric view of records vs. machine to machine, no human interaction where machines may be in different cultures