Arden Syntax and GELLO: HL7 languages for clinical logic and decision support systems

AIHealth

On-premise clinical platform with local LLMs, RAG on FHIR/DICOM data, diagnostic support, remote follow-up. Architecture designed for the MDR pathway.

Discover AIHealth →

Digital Health

Medical software development compliant with CE and MDR regulatory standards. Clinical decision support systems, AI integration in clinical workflows.

Discover →

The problem: encoding clinical knowledge

One of the historic goals of medical informatics is to represent clinical knowledge — guidelines, alert patterns, drug recommendations, monitoring protocols — in a format executable by an information system, separate from the record’s application code. A drug-allergy alert, a paediatric vaccination reminder, an attention threshold on laboratory results are examples of clinical rules that every electronic record must manage.

The simplest way to do it is to implement them in source code (Java, C#, Mumps) within the record system. The problem is that the code is vendor-specific, maintained by the producer company’s programmers, not accessible to the clinician who is the source of that knowledge. Changing a threshold, introducing a new rule, correcting an inaccuracy goes through a commercial software develop-test-release cycle.

The medical informatics community has worked, since the 1980s, on dedicated languages that separate clinical logic from application code, making it readable and maintainable by specialised healthcare staff, portable across vendors, shareable as a commons.

Arden Syntax

The most long-lived and most cited project is Arden Syntax, born from a meeting held in 1989 at Arden House (New York), with participants from Columbia-Presbyterian Medical Center, Stanford, Regenstrief Institute and other leading centres. The name comes from the location of the first meeting.

The coding unit is the Medical Logic Module (MLM) — a structured text file with slots:

• maintenance — metadata: title, maintainer, version, validation, institution
• library — bibliographic citations, keywords, clinical explanation
• knowledge:
  • data — binding with the local system (record query, laboratory read); critical point
  • evoke — events that trigger the MLM (new result, admission, temporal expiration)
  • logic — logical condition with imperative syntax (assignments, if/then, operators)
  • action — what to do if the logic is true (user message, record write, call to another MLM)
  • urgency — alert priority
• resources — data resources, multilingual text

A typical MLM has the form:

maintenance:
  title: Potassium-sparing diuretic with potassium supplement;;
  mlmname: PotassiumWarning;;
  arden: 2.5;;
  version: 1.02;;
  institution: Example Hospital;;
  author: Jane Smith, MD;;
  ...
data:
  active_meds := read last {medications where end_date is null};
  ...
logic:
  if any potassium_sparing_diuretic and any potassium_supplement then
    conclude true;
  endif;
  ...
action:
  write "Warning: patient on potassium-sparing diuretic
         AND potassium supplement — check serum K.";

Arden Syntax 1.0 became an ASTM standard in 1992 and moved under HL7 in 1998, which released version 2.0 in 1999. Version 2.5 was ANSI-approved in 2005 and represents the current reference release.

The curly braces problem

Arden Syntax’s main historic limit is the curly braces problem: data binding (the data slot) is expressed in proprietary syntax enclosed in curly braces, not standardised. Each vendor implements its own queries. An MLM written for Cerner does not run on Eclipsys without rewriting the data slot. The portability promise — the engine of the Arden philosophy — is partially compromised.

An MLM thus shares the clinical logic (portable) but not the data binding (specific). It is one of the reasons Arden Syntax is adopted in closed environments (some US CPOEs, Regenstrief, Columbia, Vanderbilt, Intermountain Healthcare) but has not seen the universal adoption anticipated.

GELLO

Work on an Arden successor began in HL7 in the early 2000s under the name GELLO — Guideline Expression Language, Object-Oriented. GELLO was designed with different goals:

• Integration with the HL7 v3 RIM (Reference Information Model) — data binding to clinical data uses the RIM, not vendor-specific free queries. Formally resolves the curly braces problem
• Object-oriented — based on OMG’s OCL (Object Constraint Language), with declarative expression syntax, similar in spirit to SQL or OCL itself
• Separation between expression language and workflow logic — GELLO is purely expression, workflow is handled separately

GELLO was approved as an HL7 standard in 2005; the community is evaluating interaction with CDA R2, vMR (virtual Medical Record), and use in CDS systems. A working draft of version 2 is in preparation for syntactic ergonomic improvements.

GELLO’s potential is substantial but uptake in 2006 is limited: a mature open source reference engine is missing, and the dependency on the complex HL7 v3 RIM model slows the entry of commercial vendors.

Other approaches

Arden and GELLO are not the only CDS languages. International literature and practice include:

• GLIF (Guideline Interchange Format) — McGill, Stanford, Columbia, founded in the late 1990s. Model for representing clinical guidelines as decision flowcharts. XML format. Associated with reference implementations like GLIF interpreters
• PROforma — Cancer Research UK (later Oxford, OpenClinical), logic-argumentative language based on a calculus of tasks (enquiry, decision, action, plan). Tallis runtime
• Asbru — TU Vienna, specialised in guidelines with temporal components (oncology protocols, treatment sequences)
• SAGE (Standards-Based Active Guideline Environment) — GE, IDX, Mayo, Intermountain, University of Nebraska consortium, to implement guidelines on HL7 foundations
• EON — Stanford, SAGE’s precursor model

None of these languages has reached Arden Syntax’s diffusion; none has fully solved the data-binding problem.

CDS in practice in 2006

Despite the richness of languages and standards, operational CDS in 2006 clinical practice most often relies on proprietary vendor EHR engines (Epic, Cerner, Eclipsys, McKesson, AllScripts):

• Drug-drug interaction engines (First Databank, Medi-Span as knowledge sources)
• Alert on order entry engines — appropriateness criteria, dosages, allergies
• Reminder system engines — vaccinations, screening, follow-up
• Order sets — pre-packaged prescribing pathways for specific conditions

Vendor-internal rules, sometimes coded in Arden-like dialects, sometimes in proprietary Java/XML. Cross-system portability remains low.

Open source implementations

The open source CDS runtime ecosystem is, as of 2006, still limited:

• Arden2ByteCode and experimental academic tools for MLM execution
• PROforma / Tallis by OpenClinical — Java runtime available in a free version
• Drools (JBoss, discussed separately) — general-purpose, not medically specific but applicable rule engine
• Jess (Rete in Java) — academic rule engine from Sandia National Labs

Missing: a complete production-ready open source Arden engine; a reference GELLO engine; an MLM authoring environment accessible to non-programmer clinicians.

The Italian landscape

In Italy, as of 2006, adoption of standardised CDS languages is very limited. Italian hospital clinical records have alert rules, but coded proprietarily by the vendor. Some academic projects (University of Pavia, University of Messina, Mario Negri Institute) explore GLIF/Asbru for specific scenarios — oncology, epidemiology. Entry of Arden/GELLO into public procurement specifications is still far off.

HL7 Italia includes a CDS working group, more oriented to monitoring international work than to producing specific Italian profiles. The concentration of the national FSE programme (launched after 2007-2008) on the documentary part makes CDS a relatively minor priority in the medium term.

Outlook

The future of Arden/GELLO — and of standardised CDS in general — will depend on a few factors:

• Availability of mature open source reference engines allowing uptake by centres without commercial partners
• Integration with emerging clinical information models (HL7 v3, CDA R2, and what HL7 is starting to discuss as an alternative — a new web-oriented specification is in exploratory phase)
• Clinical demand — pressure on healthcare systems to reduce medical errors (the famous IOM report “To Err Is Human” of 1999) increases the call for active and rigorous decision support
• Public incentive policies — some countries introduce financial incentives for EHRs with certified CDS functionality

The most promising generational leap will probably be a web-service-based standardisation — where the record calls a remote CDS service with the patient context and receives a structured recommendation. Such an evolution is expected in the coming years. Arden Syntax and GELLO retain their relevance as knowledge representation languages, regardless of the execution architecture that prevails.

References: Arden Syntax for Medical Logic Modules, HL7 v2.5 (ANSI 2005). GELLO — Guideline Expression Language, Object-Oriented, HL7 standard (2005). ASTM E1460 (Arden Syntax 1.0, 1992). GLIF, PROforma, Asbru, SAGE, EON — academic literature. OpenClinical (www.openclinical.org). Jess — Sandia National Laboratories. IOM — To Err Is Human, 1999.