gms | German Medical Science

Background: Computer-interpretable representation of clinical practice guidelines (CPG) and interpretation thereof remains, despite the various approaches within the last decades, an open topic. In [1] we have proposed an architectural concept for the structuring of a computer-interpretable guideline (CIG), which involves separating the procedural aspects of a CPG from medical knowledge. For the representation of procedural aspects of a CPG we utilize workflow control patterns (WCPs) [2] and BPMN. For both there are various hints that they can be of use in the domain of CPGs (e.g. [3], [4]. We have presented the WCP-concept at the 30th Medical Informatics Europe conference (MIE). Adhering to the principle of exploiting established standards (as we did with BPMN) instead of creating new proprietary systems, we utilized the rule-based system Drools to evaluate a CIG modelled in BPMN and based on WCPs.

Methods: The procedural aspects of CPGs were modelled using BPMN. To have a clear definition of the behavior (i.e. semantics) of the modeled guideline, the formal definition of the underlying WCPs was used. We used the business rules management system Drools to define how the individual workflow control patterns have to be interpreted. The system consists of a preprocessor creating an internal model from a given BPMN file and set of Drools rules that specify how such an internal model can be combined with patient data to infer guideline recommendations. The rules, so to speak, formally describe the semantics of the underlying WCPs.

Results: The rule system was embedded in a Spring Boot Application together with a software library that handles the medical criteria knowledge aspects of the CIG. When given a CIG and patient data, the system can infer guideline recommendations using the Drools system. Currently, only the five most basic WCPs are supported, since the rule-set only describes the interpretation rules for those. However, the current implementation outlines how the set can be extended to support additional WCPs. By defining unit tests that test each individual WCP, we were able to apply best practices for software development.

Conclusion: We have created an inference engine for guideline recommendations that is based on a set of business rules that were specified using Drools.

Using an open-source software framework such as Drools instead of a more formal framework (such as e.g. Prolog) has shown to have several distinct advantages:

1.

It was easy to be embedded into a Java-based Application.

2.

Rich documentation and various developer tools are available.

3.

The framework is time-proven (since it is an established component of various professional business softwares).

4.

The rules remain intuitively comprehensible (even for non-insiders).

Additionally, using a rule-based approach has the advantage of being validatable using first-order logic. To do so, we plan to

1.

Map the rule system to a first-order logic (FOL) system

2.

Map the colored petri-nets that define the WCPs also to a FOL system

3.

Prove that the two FOL Systems are equivalent.

The authors declare that they have no competing interests.

The authors declare that an ethics committee vote is not required.