A guideline representation language for pervasive healthcare
Due to the COVID-19 crisis the PhD defence of Nick Fung will take place online.
The PhD defence can be followed by a live stream.
Nick Fung is a PhD student in the research group Biomedical Signals and Systems (BSS). His supervisors are prof.dr.ir. H.J. Hermens and dr.ir. M.J. van Sinderen from the Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS).
Chronic diseases represent one of the main health challenges in the 21st century. Their increasing prevalence, coupled with factors such as an ageing population and a growing lack of healthcare resources, have necessitated a shift from the traditional, episodic and responsive model of healthcare to a chronic and proactive model that emphasises patient empowerment. As a result, to support this shift, pervasive healthcare systems have been researched and developed as early as the 2000s with the aim to provide "healthcare to anyone, at anytime, and anywhere". While these systems are traditionally focused on the continuous monitoring of patients, the continual development of mobile hardware technologies has enabled the emergence of intelligent systems that can support patients autonomously with minimal manual intervention from care providers.
At the same time, patient care in the traditional healthcare setting is increasingly supported by the use of clinical practice guidelines, which document the current best clinical practice as supported by the latest scientific evidence. These guidelines aim to improve and ensure the quality of patient care by facilitating adherence to proven best practice; to support their adoption, computer-interpretable, guideline representation languages have been developed to formalise clinical guidelines such that they can be executed automatically by guideline-based computer systems. In this way, support can be seamlessly provided to clinicians in making the best possible, evidence-based decisions for the patient.
This research aims to extend evidence-based healthcare beyond the traditional healthcare setting by bringing computerised clinical practice guidelines to the free-living setting to provide pervasive and guideline-based decision support to patients. In general, guideline languages capture the control flow between the different tasks in a guideline and thereby assume a centralised controller for executing them. However, for pervasive healthcare, such centralised system architectures may not be the most appropriate since system components may require dynamic reconfiguration in response to factors such as changing patient requirements and unreliable communications environments. Therefore, the main contribution of this research is a new guideline language that focuses on the data flow in guidelines, whereby tasks are modelled as processes that execute in parallel. By parallelising and dynamically distributing guideline knowledge, each device that constitutes the patient’s pervasive healthcare system can be adapted in real-time and provide decision support independently of each other, thereby avoiding a single point of failure.
The new guideline language is developed by using formal methods and by following a model-driven methodology. Thus as part of this research, a formal and generic data flow model of disease management in pervasive healthcare is created; it comprises four types of processes, namely Monitoring (M), Analysis (A), Decision (D) and Effectuation (E), and six types of data flowing between them, namely Measurement, Observation, Abstraction, Action Plan, Action Instruction and Control Instruction. This model is given a precise mathematical interpretation using axiomatic set theory, the result of which is divided into two complementary models. The first is a reference information model for representing the data flow, which comprises 32 set definitions, while the second is a guideline model for representing the MADE processes, which comprises 28 set definitions as well as 13 function signatures and 38 logical invariants to specify their behaviour.
From the reference information model and guideline model, the syntax and semantics of the new guideline language are derived for representing clinical guidelines. To support the verification and validation of the formalised guidelines, the syntax and semantics of an accompanying archetype language were also developed for specifying MADE archetypes (i.e. well-formedness constraints on MADE data items). Furthermore, a reference implementation for these two languages is developed which comprises a set of libraries implemented on top of Rosette. Since Rosette provides support for not only executing the languages but also verifying them using off-the-shelf constraint solvers, the reference implementation is formally verified to comply with the 38 logical invariants of the guideline model. More specifically, these constraint solvers help ensure that no patient data will ever cause the reference implementation to violate its invariants during execution.
Validation of the MADE models and languages is conducted by formalising, verifying and testing a complete clinical guideline for gestational diabetes mellitus, i.e. diabetes that is first developed or first recognised during pregnancy. The guideline comprises 13 semi-formal workflows, such as for managing blood glucose levels and urinary ketone levels, and the result is a MADE network comprising 55 processes, specifically 0 Monitoring, 4 Analysis, 22 Decision and 29 Effectuation processes, all of which are connected together by the flow of 50 types of MADE data, specifically 0 Measurement, 8 Observation, 8 Abstraction, 14 Action Plan, 3 Action Plan and 17 Control Instruction archetypes. The 0 Monitoring processes and 0 Measurement archetypes were a result of the fact that all measurement tasks in the guideline were to be performed manually by the patient.
In the future, the clinical relevance of the MADE languages should be evaluated by developing and performing clinical studies on pervasive healthcare systems that implement the MADE languages. Usability of the languages should also be evaluated in collaboration with clinicians, patients and other stakeholders; improvements may include adding support for personalising guideline knowledge and for partial specifications of manual processes. A knowledge acquisition tool suite may also be developed from the reference implementation to provide support for formalising clinical guidelines, developing and executing test data as well as visualising the test and verification results.