Drools: open source rule engine and applications in clinical decision support

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 →

A general-purpose rule engine

Work on languages and rule engines born inside the healthcare domain — Arden Syntax, GELLO, GLIF, PROforma, Asbru — has produced technically ambitious specifications but with operational uptake limited by ecosystem factors: fragmented runtimes, difficult authoring, non-trivial integration with record systems. In parallel, the general IT industry has developed domain-agnostic rule engines with a much wider user base and mature tooling.

The most widespread in the open source world is Drools — Java project started in 2001 by Bob McWhirter and later absorbed by the JBoss community in 2005 (with the temporary branding JBoss Rules). With Red Hat’s acquisition of JBoss in 2006, Drools officially enters the enterprise open source offering of a top-tier vendor.

Drools 4

Drools 4.0, released in May 2007 and still stable in the 4.0.x series, is the project’s biggest evolutionary leap since inception. The new features:

• Rete-OO — rewrite of the Rete algorithm (Forgy, 1974) with object-oriented adaptation, developed by Mark Proctor (current project lead). Preserves sub-linear average complexity with respect to the number of facts, with efficiency improvements on common patterns
• DRL (Drools Rule Language) — declarative language for rules, with more readable syntax than version 3
• DSL (Domain Specific Language) — mechanism to define domain-specific declarative languages on top of DRL, writable by business analysts (or by clinicians, in the healthcare domain)
• Decision Tables — tabular representation of rule families in Excel or OpenDocument, useful to model decision matrices
• Drools Flow — ruleflow workflow engine describing sequences of rule invocations as processes
• Drools Solver — constraint optimisation problem solver, an extension that will become an independent sub-project in coming years
• Drools Guvnor — web-based rule repository, business rules management system (BRMS) for artefact lifecycle

The licence is Apache 2.0. The code is hosted in the JBoss public repository.

The rule model

A Drools rule in DRL has the form:

rule "Warfarin and aspirin interaction"
    when
        $patient : Patient()
        $warfarin : ActiveMedication(drug == "warfarin", patient == $patient)
        $aspirin  : ActiveMedication(drug == "aspirin",  patient == $patient)
    then
        insert(new Alert(
            $patient,
            "Interaction: warfarin + aspirin — bleeding risk",
            AlertLevel.HIGH
        ));
end

The when block (also called LHS, Left-Hand Side) is a sequence of patterns the engine tries to match against the working memory — the set of facts injected by the calling application. The then block (RHS) executes actions — create new facts, modify existing ones, call external services.

The Rete algorithm implemented by Drools builds a discrimination network across rule conditions, in which facts are propagated and filtered incrementally: adding or removing a fact does not require re-evaluating every rule from scratch, only the affected part of the network. This incrementality is what makes Rete suited to scenarios with many rules and frequent working-memory updates — the case of an active clinical record.

Why it matters to healthcare

Drools is not designed for healthcare. But the combination of its characteristics makes it attractive for CDS scenarios:

• Performance — Rete-OO handles thousands of rules over working memories of tens of thousands of facts with sub-second response times
• Open licence — Apache 2.0 allows any use, including embedding in products certified as medical devices (with the qualification work that entails)
• Enterprise ecosystem — natural integration with Java stacks (JBoss AS, Spring), JTA transactional support, execution in J2EE containers
• DSL — ability to build domain-specific declarative languages for clinicians, with syntax readable by non-programmers
• Authoring tools — Drools Guvnor, Eclipse plugin, while Arden MLM authoring remains mostly textual editors

Some concrete adoption directions:

• Clinical trial eligibility — rules on inclusion/exclusion criteria expressed as Drools rules, with automatic evaluation at enrolment
• Drug-drug interaction — interaction databases expressed as rules, with matching at prescription time
• Order set appropriateness — verification of the appropriateness of a clinical order set against a primary diagnosis
• Preventive reminders — temporal rules on screenings, vaccinations, follow-ups
• Pathway compliance — monitoring of adherence to diagnostic-therapeutic pathways for chronic conditions (diabetes, hypertension, heart failure)

Some academic projects — NHS in the UK, Veterans Affairs in the US, European e-Health groups — are experimenting with Drools as an execution engine for clinical guidelines encoded in more abstract models (GLIF, PROforma); in these cases Drools does not replace Arden/GELLO but is the engine that runs them.

The data binding issue

The recurring CDS question — how the rule engine accesses patient data — resurfaces with Drools. Unlike Arden (which prescribes a specific data slot), Drools is entirely agnostic: facts in working memory are plain Java objects (POJOs), which the calling application must populate from clinical sources.

Patterns used in experimental projects include:

• HL7 vMR virtual model (virtual Medical Record) — standard Java objects representing patient demographics, active medications, diagnoses, lab results. The adapter populates the vMR from the record
• Record database binding via JPA/Hibernate — record entities directly become Drools facts
• FHIR as intermediate format — here the medical informatics world starts looking at HL7’s early discussions of a new standard oriented to resources and REST APIs, under incubation in international working groups

The problem is technically solvable but requires a coherent architecture — which cannot be delegated entirely to the rule engine.

Commercialisation and support

Red Hat offers JBoss BRMS (Business Rule Management System) as commercial product, with support subscriptions on top of open source Drools. The model is the typical Red Hat one: identical code to the open version, plus certifications, QA, patch backports, phone and on-site support. For clients with enterprise production needs (large hospitals, pharmaceutical companies with regulatory applications) this model provides a stability anchor that the pure open source version may not offer.

Italian adoption

In Italy, in 2007, Drools is known in enterprise Java contexts — banks, telco, public administration — but its uptake in healthcare is still exploratory. Some research groups (Universities of Pavia, Milan, Pisa, Messina) explore Drools for experimental CDS projects. Integration with Italian commercial clinical records is a non-trivial topic — requires custom adapters — and the record vendor business model does not always incentivise such integration.

From the public administration perspective, the Red Hat/JBoss model with enterprise support is already present in tenders for application stacks; extending it to healthcare CDS scenarios is a plausible but not yet activated trajectory.

Alternatives

Drools is not the only rule engine available:

• Jess — Sandia National Labs, Rete algorithm in Java, historically academic. No longer under intensive active development
• IBM ILOG JRules — commercial, de facto standard in banking-insurance
• OpenRules — commercial Decision Table engine with a reduced free version
• Prova — reactive rule engine with event support, academic
• ClassicJade — CLIPS-derived expert systems

In the open source ecosystem, Drools is the dominant JVM choice. For other languages, the ecosystem is fragmented (NRules for .NET, pyke/rete-py for Python, various Ruby DSLs).

Outlook

Drools development directions in the coming years will include:

• Complex Event Processing (CEP) — handling of temporal event streams, with operators over sliding windows. Extremely relevant in clinical contexts (vital sign time series, continuous monitoring)
• Integration with BPMN 2.0 — the new business process modelling standard published by OMG in 2007 — via jBPM
• Expansion of Drools Guvnor towards a full BRMS with versioning, approval flows, release policies
• Integration with standardised data models — HL7 vMR, prospectively the models that will emerge from HL7’s new post-CDA discussions

Drools’ role in healthcare in coming years will be determined by: availability of shared clinical rule libraries (the good most lacking in the CDS ecosystem); maturity of adapters to real clinical records; emergence of a truly adopted exchange standard for communication between record and external CDS service. Drools is technically ready; the other pieces are starting to move.

References: Drools 4.0 (May 2007), JBoss Rules / Red Hat (www.jboss.org/drools). Rete algorithm (Charles Forgy, 1974). Rete-OO (Mark Proctor). Apache 2.0 licence. HL7 vMR — virtual Medical Record. OMG BPMN 2.0 (2007).