Decentralised Clinical Guidelines Modelling with Lightweight Coordination Calculus

Bo Hu, Srinandan Dasmahapatra, David Robertson, Paul H. LewisLBM (Short Papers) 2007

Abstract

Clinical practice guidelines (CPGs) are vital for standardising patient care, yet traditional computerised models often assume a centralised execution environment. This assumption is increasingly incompatible with modern healthcare, which is characterised by distributed teams and multi-institutional care pathways. This paper explores the application of Lightweight Coordination Calculus (LCC)—a message-based process calculus from the field of multi-agent systems—to model and execute CPGs across decentralised networks. By formalising clinical workflows as peer-to-peer interactions, LCC provides a robust framework for secure, unambiguous task allocation among medical professionals and software agents.

Introduction

The healthcare industry is undergoing a paradigm shift toward pervasive, decentralised care. Patients with complex comorbidities, such as concurrent heart disease and diabetes, are frequently managed by multidisciplinary teams across different hospitals and geographical regions. To ensure safety and efficacy, these distributed teams must adhere to established Clinical Practice Guidelines (CPGs).

However, existing CPG modelling languages—including PROforma, Asbru, and GLARE—were primarily designed for centralised execution. These languages typically operate as single-thread workflows and lack native mechanisms for task partitioning, cross-institutional role allocation, or managing concurrent execution threads. Currently, the distribution of these tasks relies on informal, ad hoc agreements, which introduces risks of inconsistency, reduces repeatability, and raises significant quality assurance concerns. This research addresses these gaps by formalising decentralised CPG execution.

Methodology: Lightweight Coordination Calculus (LCC)

Rather than relying on the conventional "task-network model," this approach utilizes declarative interaction specification. The authors adopt Lightweight Coordination Calculus (LCC), an executable process calculus originally developed for peer-to-peer networks and multi-agent systems.

In the LCC framework, coordination is achieved by defining specific roles for participants and the messages they exchange. The authors evaluated LCC against four primary criteria for clinical modelling: Formalisation, Conceptualisation, Expressiveness, and Readability. Using a Prolog-based implementation, clinical workflows are mapped into logical clauses:

  • Roles: Participants are explicitly defined (e.g., a(radiologist, Rad)).
  • Workflows: Constructed using LCC operators such as then (sequential execution), par (parallel execution), and or (conditional/alternative paths).
  • Constraints: Logic-based checks (e.g., verifying if a patient's heart rate exceeds a specific threshold) are managed by the underlying Prolog logic program.

Case Studies and Evaluation

To validate the system, the authors applied LCC to two real-world clinical guidelines from the UK National Health Service (NHS):

  1. Adult Chronic Cough Guideline (2006): A text-heavy guideline intended for general practitioners and nurses.
  2. Breast Cancer Triple Assessment (TA) Guideline (2001): A complex guideline involving text, flowcharts, and tables, requiring coordination across various clinical specialists.

Key Insights

  • Healthcare as a Multi-Agent System: Modelling clinical professionals and software systems as independent "agents" aligns closely with the reality of modern healthcare. LCC defines exactly what data or requests an agent must receive before they can act, ensuring clear accountability.
  • Implicit Access Control: Unlike centralised systems that require external security layers, LCC-based models offer inherent security. Clinicians only access information contained within messages specifically routed to their role, naturally enforcing patient confidentiality.
  • Natural Mapping: LCC’s constructs mirror medical logic. The par operator handles simultaneous lab tests; or manages alternative treatment options; and recursive calls facilitate continuous patient monitoring until specific clinical thresholds are met.

Results

The qualitative analysis demonstrates that LCC is highly capable of representing complex, real-world CPGs:

  • Breast Cancer TA Model: The authors successfully modelled the ultrasound recommendation process. They demonstrated how LCC uses cyclical calls to evaluate patient status iteratively, update numeric scores, and trigger recommendations once termination conditions are satisfied.
  • Chronic Cough Model: The study showcased LCC’s strength in parallel data acquisition. The model coordinated the simultaneous gathering of demographic data, smoking history, and medical records from disparate sources, updating the Electronic Health Record (EHR) only after all parallel tasks were verified by the patient.

Limitations and Future Work

While LCC is logically rigorous, its Prolog-like syntax presents a barrier to entry for domain experts like clinicians. Future work must focus on improving readability and developing user-friendly graphical tools. Additionally, the authors suggest extending LCC to include:

  • Typed variables for stricter data validation.
  • Temporal constraints to handle time-sensitive medical interventions directly within the calculus.
  • Probabilistic assertions to better manage medical uncertainty.
  • Ontological integration with standards like SNOMED to ensure semantic interoperability across different healthcare systems.

Summary updated

BibTeX
@inproceedings{hu2007decentralised,
  author = {Bo Hu and Srinandan Dasmahapatra and David Robertson and Paul H. Lewis},
  title = {{Decentralised Clinical Guidelines Modelling with Lightweight Coordination Calculus}},
  booktitle = {LBM (Short Papers)},
  year = {2007},
}
Figure