Evidence-based clinical guidelines are being more and more adopted and applied by healthcare structures within and outside Italy. The main goal of clinical guidelines is to capture medical evidence and turn it into directions, principles and recommendations that support care practitioners in the appropriate diagnosis, therapy and execution of clinical procedures for specific medical circumstances.
On the one hand, the elicitation and unambiguous modeling of medical guidelines is a complex task, which requires to properly tune the level of abstraction (e.g., which part of the whole medical knowledge must be explicitly mentioned in the guideline description) as well as finding a balance between different, sometimes contrasting requirements (medical, juridical, social, economic, …).
On the other hand, the guidelines-driven development and execution of corresponding clinical pathways and processes inside a concrete healthcare structure is also an extremely complex and critical task in many respects.
As far as modeling is concerned, the recommendations provided by guidelines must in this phase be suitably combined with administrative and resource constraints, and more in general with specific “local” requirements, to produce executable, fine-grained healthcare processes that refine the guidelines. In this light, clinical processes can be considered as a notable example of “knowledge-intensive process”, because the execution of their activities mostly depend on the data acquired for the patient so far, and on a consequent decision-making task in which the involved professionals combine such data with the guideline’s recommendations and their own background medical knowledge. This, in turn, poses many challenges for computer scientists: from the development of suitable clinical process modeling languages, able to mediate between simplicity/usability and expressiveness, to the verification of their correctness and compliance with the guidelines recommendations and other rules of interest.
As far as technology is concerned, support is required to properly interconnect the abstract, conceptual data model used in the guideline’s description with the concrete data sources that store such information inside the information system of the organization. Typically, conceptual information regarding patients and their EPRs, medical terminologies, drugs, diseases, medical devices, …, must be properly reconciled with the relational tables stored inside one or more databases.
This problem obviously impacts on the execution of clinical processes: on the one hand, during the execution they will typically manipulate the concrete data stored inside the healthcare information system, but on the other hand healthcare professionals (would like to) understand the evolution of such data at the conceptual level. Reconcile these two different levels of abstraction is far from trivial.
In the context of the PRIN Project, the FUB Unit (constituted by the KRDB group of the Free University of Bozen-Bolzano) will investigate these relevant, open challenges, tackling in particular the following three main issues:
1. Combined modeling of clinical processes and data (“data-centric clinical processes”), focusing not only on the control-flow among activities, but also on how their execution leads to manipulate data. We will consider both the case in which data are represented at the conceptual level (by means of description logics ontologies), and the case in which the data model is fixed by the sanitary organization (typically following the relational model).
2. Verification of rich temporal properties over such data-centric processes, checking whether the produced model is correct and meets some properties of interest. Properties can represent specific requirements and rules that must be ensured within the organizational boundaries (such as internal policies or administrative constraints), but can also represent (a portion of) a clinical guideline. In this latter case, verification is exploited to check the adherence of the process to the recommendations of the guideline for which it has been developed.
3. Investigation of ontology-based data access (OBDA) techniques to reconcile the representation of knowledge at the conceptual level and the underlying data sources. On one side, this requires a study of different ontology languages and their balance between expressivity and tractability. Expressivity is required to properly capture the (complexity of the) medical knowledge, while tractability is needed to guarantee the feasibility of OBDA in accessing real-world amounts of data. On the other side, we will study how such OBDA techniques can be profitably exploited to “understand” the evolution of data in view of the process execution, and to ultimately query, analyze and govern the running system.
The FUB Unit has extensive experience in bridging the gap between the areas of databases and knowledge representation, and in the development of techniques and tools for the efficient management of large amounts of data with a complex structure and complex interrelationships, taking also into account the processes that manipulate such data. In the context of a technology transfer grant, and then during his PhD and Post-Doc, the principal investigator of the unit studied and applied languages, techniques and tools developed in the area of knowledge representation and reasoning to the formalization, verification and operational support of clinical guidelines and processes. Furthermore, the FUB unit is currently involved in the EU Project ACSI (Artifact-Centric Service Interoperation), in which the unit is investigating the formal modeling, verification and governance of artifact- and data-centric dynamic systems (where both processes and data are considered as first-class citizens).
The experience developed by the FUB unit in ACSI will constitute a good starting point for the PRIN research activity. We will check if, how, and to what extent the techniques developed in ACSI can be applied, adapted and further improved for the medical setting.
The three aforementioned major challenges will be extensively studied by the FUB unit within the five work packages of the project, and in particular WP1 and WP2. The unit will also participate to the analysis of the state of the art (WP0) and to WP4, dedicated to the development of concrete prototypes.
In the remainder of this Section, we provide a more detailed description of the tasks and sub-tasks that will involve the FUB unit, following the work package structure of the project.
WP0: State of the art analysis of the clinical domains
FUB will collaborate with all the other research units in the fundamental task
“0.1 - Analysis of the state of the art and cross-dissemination between Units”. The two major contributions by FUB will cover:
• A literature review of the current process modeling approaches that consider data as first-class citizens, such as case-handling systems and artifact-centric languages. In parallel, we will focus on careflow and clinical guidelines/process specification languages, in order to classify their data-support not only at the level of editing tools, but also at the language/foundational level.
• An analysis of the main approaches that propose the use of ontologies and conceptual data models in the medical setting, especially those focusing on the information of interest during the execution of clinical processes (e.g. patients’ EPRs). This study will also reveal which specific ontology and description logic languages are mainly used in this setting
This literature review will be used to actively contribute to task “0.3 – Selection of case studies”, in particular to select the case studies that require a deep investigation of the interplay between the process and data dimensions. This selection will be also driven by the presence of real data, to be used for the concrete experimentation of the medical OBDA techniques studied in WP1-WP2.
WP1: Integrated modeling of data, clinical processes and healthcare processes
In the context of WP1, the efforts of the FUB unit will be directed towards tasks “1.1 - Medical ontologies and clinical processes” and “1.3 - Modeling clinical Pathway (workflow)”. In this respect, FUB will cooperate with other units of the project, in particular UNIBO.
For what concerns task 1.1, we will first focus on the analysis of ontology languages and their computational properties. Starting from the literature study carried out in WP0, we will evaluate several ontology languages (in particular description logics), focusing on the two contrasting dimensions of expressivity and tractability.
The first dimension will help us to better understand which languages can be effectively used for expressing typical conceptual requirements posed by the medical domain. The evaluation will rely on the existing literature and medical ontologies, but also on the analysis of the relevant case studies selected in WP0, by modeling fragments of their structural knowledge.
The second dimension will instead help us in the assessment of which ontology languages can be exploited during the clinical process execution, especially for what concerns ontology-based data access (see below). In this respect, we will focus on data complexity, since real-world healthcare structures typically rely on very large databases.
We will conclude this analysis by investigating whether lightweight but tractable ontology languages can be satisfactorily employed to “approximate” complex medical descriptions [Pan&Thomas, 2007] [Botoeva et al, 2010], thus providing a good trade-off between expressivity and tractability.
Alongside the investigation of ontology languages for modeling relevant medical information at the conceptual level, we will also study how to connect such a semantic level to concrete data sources (medical OBDA). On the one hand, this will enable the possibility of posing queries at the conceptual level, and obtain answers calculated using the real data maintained in a healthcare information system. On the other hand, this will provide the basis for integrating multiple, possible heterogeneous data sources, facilitating their interoperation and understandability. Technically, the link between ontologies and underlying data sources is typically established via mappings [Lenzerini, 2002], logical rules that relate queries posed to the ontology with queries posed to the data sources. In this respect, we will investigate which kinds of mappings are of interest in the project’s setting, and more in general study data integration patterns in the medical domain.
With respect to task 1.3, we will investigate specification languages, and corresponding formalisms, which allow for modeling the process and data perspective together. As we already pointed out, this is of utmost importance in the medical setting, where processes are inherently “knowledge-intensive”.
The majority of guideline/clinical process specification languages, as well as process languages in the business process management setting, mainly focus on the control-flow perspective. Some degree of data-support is provided but only at the “tool level”, i.e., the development tool supports the possibility of connecting activities to data, but this connection is not represented in the underlying formal representation language. For example, Petri Nets are often used to define the semantics of process specification languages. However, (standard) Petri Nets only capture the control-flow perspective, completely abstracting away from the data component. To overcome this issue, artifact- and data-centric approaches [Nigam&Caswell, 2003] have been increasingly studied in the last years. They do not only provide the possibility of truly modeling how the process manipulates the data component, but they also provide new conceptual abstractions (such as “artifacts”) that put the main focus on the key relevant entities of the domain. In the healthcare setting, typical artifacts could be the EPR of a patient, the notion of medical examination, of prescription, and so on.
It is our intention to investigate the use of these data-centric approaches for modeling clinical behavioral patterns and pathways as well as entire clinical processes. At the same time, we will study the possibility of extending currently existing guidelines/careflow specification languages (such as GLARE [Terenziani et al, 2003]) to provide a more tight connection with the data/knowledge component.
By leveraging on the ontology-related task 1.1, we will specifically consider two distinct settings for the combined modeling of clinical pathways and data (see figure):
• Specification of “abstract” pathways, whose data component is specified using an ontology language. This facilitates the re-use of pathways in different healthcare structures, because data are modeled at the conceptual level, abstracting away from implementation details.
• Specification of “concrete” pathways, whose process component manipulates data that are modeled by the concrete schema of a particular organization. In this case, OBDA techniques can be used to understand the execution of such pathways at the conceptual level (see task 2.2).
To provide the underlying formalization, we will start from some recent works in the area, in particular [Bagheri Hariri et al, 2011] for abstract pathways, and [Bagheri Hariri et al, 2012] for concrete ones.
WP2: Design time and Run-time verification of clinical processes
By leveraging on the research results achieved in WP1, we will investigate corresponding verification and reasoning techniques for design time verification (task 2.1) and run-time verification (task 2.2). To do so, we will tightly collaborate with UNIVR, UNIBA, UNITO, UNIPR and UNIPMN.
For what concerns task 2.1, we will focus on the verification (model checking) of rich temporal properties over clinical pathways modeled in a data-centric way.
We will consider the adoption of sophisticated first-order temporal logics such as CTL, LTL and mu-calculus [Clarke et al, 1999], which allow us to specify properties about the (un)desired evolutions of the data component (such as for example that “for each patient, if an indication of diabetes risk is detected, then in every possible future evolution of the process the patient will be eventually informed of the risk”).
The verification of such expressive formulae over data-centric clinical pathways is a challenging problem. In fact, due to the presence of data, the transition system representing the possible pathway evolutions is in general infinite-state. In order to gain verifiability, suitable restrictions on the temporal formalism and/or the shape of the pathway must be investigated, in such a way that a corresponding faithful finite-state abstraction of the system can be constructed, consequently guaranteeing decidability of verification.
Some recent works in this context show that undecidability of verification holds even for very simple cases, but also that relevant interesting decidable fragments can be isolated [Bagheri Hariri, 2011] [Bagheri Hariri, 2012].
Starting from these results, we will study the decidability of verification of interesting medical properties over clinical pathways, taking into account the projects’ case studies to check whether the necessary restrictions on the shape of the process component are practically reasonable, and how to improve them.
With respect to task 2.2, we will apply the medical OBDA techniques investigated in WP1 to the aim of providing run-time governance and analysis facilities. We will in particular consider the typical setting in which “concrete” clinical pathways and processes directly manipulate the data stored into the healthcare information system of a sanitary organization. In such a situation, the execution of an activity has the effect of modifying the data maintained into the concrete data sources. Thanks to OBDA and mappings, it is possible to “project” the different database instances resulting from a clinical pathway’s execution at the conceptual level, enabling to understand the evolution of the system through the filter of the ontology. On top of this mechanism, we will therefore provide to managers and healthcare professionals a unified and conceptual view of the whole information maintained in the (possibly heterogeneous and distributed) data sources of the organization’s information systems. We will then elaborate on how to support stakeholders in the task of querying such information at the ontological level, exploiting the obtained answers for reporting and analysis.
WP4: Design and implementation of a software suite for the integrated management and analysis of clinical guidelines and clinical data.
FUB will collaborate with the other research groups towards the development of a suite for the integrated management and analysis of clinical guidelines and data. In particular, FUB will concentrate on the adaptation, extension and customization of OBDA techniques to cope with the medical setting and to tackle the interplay with the process component. To this aim, the OBDA technology developed by the FUB unit will be exploited (see http://obda.inf.unibz.it/). In particular, we will rely on the Quest Java-based reasoned, on the Protégé OBDA plug-in and on the Quelo ontology-driven query interface to provide a suitable semantic dashboard for the run-time analysis and governance of clinical pathways.
In conclusion, the expected results are:
• A critical evaluation of ontology languages for modeling the data component of clinical pathways, assessing their suitability w.r.t. expressivity and tractability;
• The elicitation of suitable clinical pathways data-centric modeling languages;
• The definition of suitable abstraction techniques to enable the (static) verification of data-centric clinical pathways w.r.t. expressive temporal properties;
• The development of a semantic dashboard for querying and analyzing the running executions of clinical pathways and their manipulated data at the conceptual level.
13 - Descrizione delle attrezzature già disponibili ed utilizzabili per la ricerca proposta
14 - Elenco dei partecipanti all'Unità di Ricerca
14.1 Personale dipendente dall'Ateneo/Ente cui afferisce l'Unità di ricerca