1. Third International Conference on Health Informatics
Representation and Efficient Management of Multi-version Clinical Guidelines
Fabio Grandi
Alma Mater Studiorum - Università degli Studi di Bologna
Bologna, Italy
BIOSTEC Valencia, Spain, January 2010

2. Introduction
Clinical Guidelines are definitions of “best practices” encoding and standardizing clinical procedures for a given disease
Several computer systems have been developed for the management of clinical guidelines, and large collections of guidelines are being made available on the Web
Guidelines are subject to continuous development and revision by committees of experts and health authorities and may contain variant parts depending on available resources and execution contexts
As a consequence, multiple versions of guidelines coexist
This work is part of a reasearch activity concerning eGov applications. In recent years, there has been a great development of eGov initiatives pushed by PAs with more and more resources made available on the web. We use TDB and SW techniques to support personalized access and, as resources, we consider multi-version norm documents (laws, decrees, regulations etc) and service specifications in XML format.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

3. Introduction
In this context, we propose to use temporal database and semantic Web techniques to provide personalized access to multi-version clinical guidelines stored in an XML repository (either in textual or executable form)
In particular, we propose to apply powerful techniques we developed for e-Government applications, which have similar versioning and personalization requirements
In this way, individual patient profile, context information (e.g. including available hospital facilities and practitioners’ skills) and multiple temporal perspectives can be used to build a guideline tailored to a specific use case
Extending to the management of clinical guidelines techniques we recently developed for eGov applications, we use TDB and SW techniques to support personalized access. Using patient profile, context information and temporal perspectives, personalization allows users to build a guideline tailored to a specific use case.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

4. A personalization framework for clinical guidelines
PATIENT PROFILE
TEMPORAL PERSPECTIVE
CONTEXT INFO
QUERY
PERSONALIZATION ENGINE
The personalization framework we consider is composed of a multi-version guideline repository which is accessed through a personalization engine. A physician using the system can express queries, which are automatically enriched by the personalization engine with selection conditions concerning the patient, the application context and the temporal perspective of interest to retrieve and reconstruct a personalized version of the desired clinical guideline.
USER
PERSONALIZED GUIDELINE VERSION
MULTI-VERSION GUIDELINE REPOSITORY
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

5. Overview of the contribution
Definition of an XML data model supporting multidimensional temporal and semantic versioning
Design and prototype implementation of a multi-dimensional XML Query Processor which can be used as personalization engine …
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

6. Importance of versioning
Temporal aspects are important in an evolving medical domain
Owing to evolution of medical knowledge and clinical practice, a clinical guideline changes in time due to subsequent modifications, evolving through a sequence of temporal versions
The ability to capture temporal versioning is very important for the management of clinical guidelines in a dynamic environment
it is crucial to reconstruct the current version of a guideline
also past versions are still important! 2
new version
Original guideline 3
new version 1
time
modification
modification
In this context, temporal aspects are very important. As a matter of fact, giudelines are subject to amendments and changes and, thus, starting from an initial text, modifications produce a sequence of versions, the most important of which is the last one, that is (borrowing the term from the legal domain) the so-called consolidated version which takes into account all the effected modifications and represents the state-of-the-art of clinical practice. However, also past versions are still important and are often searched by queries for audit of control purposes.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

7. Importance of versioning
Applicability (semantic) versioning also plays an important role
some guidelines or some of their parts have or acquire a limited applicability
personalized version of the clinical guideline:
A version only containing recommendations which are applicable to a specific use case (patient profile, context information...)
Sec. 1 (myocardial ischemia)
xxy yyx yxyx yyyxx xyyx
Sec. 2 (angina pectoris)
aab bbab abab abba ab
Sec. 3 (myocardial infarction)
qwqq ww wqqw wq ww
Sec. 1 (myocardial ischemia)
xxy yyx yxyx yyyxx xyyx
Sec. 2 (angina pectoris)
aab bbab abab abba ab
Sec. 3 (myocardial infarction)
qwqq ww wqqw wq ww
PATIENT PROFILE:
Another important versioning dimension is semantic and involves the applicability of a guideline or of a guideline portion to a limited set of patients. Hence we can define the personalized version of a guideline as the one composed of the recommendations which are all applicable to an individual patient or to a homogeneous class of patients. The patient is classified according to a profile, which can be automatically derived from medical records through a suitable reasoning service or is explicitly supplied by the physician. Other semantic versioning dimensions can be defined with reference to the context in which the guideline has to be executed.
Angina pectoris
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

8. Temporal versioning
Several time dimensions are involved in the representation and management of clinical guidelines including:
Validity time
Transaction, Efficacy, Event, Availability, Proposal, Acceptance time...
Temporal versioning (along multiple time dimensions) can be added to XML documents via version elements and timestamps
<section num="10"> <title>Drug Therapy</title> <version num="1"> <valid from=" " to=" " > … administration of drug D1 … </version> <version num="2"> <valid from=" " to=" " > … administration of drug D2 … </section>
Guideline version valid from 2006 to 2008:
Drug Therapy
… administration of drug D1 …
… administration of drug D2 …
Guideline version valid from 2009 on:
Drug Therapy
… administration of drug D2 …
Guideline version valid from 2001 to 2005:
Drug Therapy
… administration of drug D1 …
As far as temporal versioning is concerned, several orthogonal time dimensions can be considered (in particular, if we also consider the guideline authoring and approval environment). The most important time dimension is validity , which is the time the guideline can be considered in force by the medical community and, thus, must be applied to patients. All the required time dimensions can be uniformly dealt with in a multi-dimensional data model. For the sake of simplicity, we will only consider validity time in the following.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

9. Semantic versioning
Aimed at providing personalized access to clinical guidelines (in textual and executable form) with respect to their applicability
Applicability can be defined with reference to domain ontologies
Heart disease
Heart valve disease
Myocardial ischemia
Angina pectoris
Unstable angina
Microvascular angina
Myocardial infarction
Rheumatic heart disease
For instance, applicability of multi-version guidelines to individual patients can be defined according to a consensual taxonomy of diseases
Other kinds of context information can be represented according to suitable reference ontologies
In order to support personalized access wrt applicability, we introduce reference ontologies, which capture a classification of use cases wrt to the distinctions introduced by guideline provisions with limitations in their applicability. In this example we can see an ontology concerning a classification of patients suffering form heart diseases. Applicability of guideline recommendations to individual patients can be defined with reference to the ontology classes.
Other kinds of context information can be represented according to suitable reference ontologies, however, in the following we will only consider diseases for simplicity.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

10. Semantic versioning
At this stage of the project, we manage “tree-like” ontologies
class taxonomies induced by the IS-A relationship
we exploit the pre-order and post-order properties of trees
Heart disease
Heart valve disease
Myocardial ischemia
Angina pectoris
Unstable angina
Microvascular angina
Myocardial infarction
Rheumatic heart disease
(1,8)
Applicability of different parts of a guideline to the relevant classes of the reference ontology is used as a new versioning dimension
References to ontology classes can be added to the XML annotation of guidelines in order to encode limited applicability
(2,1)
(3,6)
(8,7)
(4,4)
(7,5)
In this project we consider tree-shaped ontologies, that is taxonomies induced by the ISA relationship and for which codes corresponding to the pre-order and post-order visit of the tree can be assigned to nodes. Pre-order codes are also used as class identifiers. Reference to node Ids can then be added to guidelines as XML annotations to define versions with different applicability.
(5,2)
(6,3)
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

11. Semantic versioning
Applicability is inherited by descendant nodes unless locally redefined
By means of redefinitions we can also introduce, for each part of a document, complex applicability properties
Extensions with respect to ancestors
Restrictions with respect to ancestors
Heart disease
Heart valve disease
Myocardial ischemia
Angina pectoris
Unstable angina
Microvascular angina
Myocardial infarction
Rheumatic heart disease
(1,8)
<section num=“1”>
<ver num=“1”>
<applies to=“C3”/>
[… Temporal attributes …]
<paragraph num=“1”>
<ver num=“1”> [ … Text … ]
<applies to=“C4”/>
</ver>
</paragraph>
<paragraph num=“2”>
<applies also=“C8”/>
</section>
(2,1)
(3,6)
(8,7)
In the XML structure of guidelines, applicability is inherited by descendant nodes unless locally extended or restricted.
For example, section 1 is applicable to class 3, that is myocardial ischemia patients, an to all its subclasses. Its first paragraph is applicable to class 4, which is a restriction, whereas its second paragraph is applicable to class 3 by inheritance but also to class 8 by redefinition, which is an extension.
(4,4)
(7,5)
(5,2)
(6,3)
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

12. Example of complex query
Find all the clinical guidelines ...
... with a section whose title contains the word anticoagulant, ...
... which were valid between 2007 and 2008, ...
... and which are applicable to a patient suffering from unstable angina (disease class 5).
Structural constraint
Textual constraint
Temporal constraint
Applicability constraint
As an example of query that can be executed, we assume a physician using the system is looking for guidelines containing the word “anticoagulant” in a section title, which were valid between 2007 and 2008 and which can be applied to a patient suffering form unsable angina.
4 orthogonal constraints
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

13. Example of complex query
FOR $a IN guidelines
WHERE textConstr ($a//section/title/text(), ’anticoagulant’)
AND tempConstr (’vTime OVERLAPS PERIOD(’ ’,’ ’)’)
AND applConstr (’C5’)
RETURN $a
Structural constraint
Textual constraint
Temporal constraint
Applicability constraint
Such constraints can all be expressed in a XQuery-like statement that can be executed by our personalization engine.
4 orthogonal constraints
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

14. The personalization engine
Based on a Temporal Multi-version XML Query Processor:
provides all the structural, textual, temporal, and applicability query
facilities in a single component
exploits ad-hoc data structures and
algorithms to obtain high efficiency
allows users to retrieve and
reconstruct on-the-fly XML document versions satisfying the four types of constraints
The personalization engine is a powerful multi-version XML query processor implemented in Java, which is able to execute flexible user queries involving structural, textual, temporal and applicability constraints. In order to obtain a high query efficiency, XML guidelines are stored in a partitioned way and the engine exploits ad-hoc structural join algorithms. Giudelines satisfying the query constraints are retrieved and assembled on-the-fly to produce the desired personalized versions.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

15. Prototype implementation
The personalization engine is able to access and retrieve only the strictly necessary data
selection relies on ad-hoc data structures supporting multi-versioning
storage granularity is finer than the entire documents used by standard XML engines
Only the parts which satisfy the temporal and applicability constraints are used for the reconstruction of the retrieved guideline documents
There is no need to retrieve whole XML documents and build space-consuming structures such as DOM trees
An important feature of the XML processor is that it only retrieves the data portions which are strictly necessary to build the desired guideline personalized version, instead of retrieving whole multi-version guidelines and then cutting off unnecessary versions with a further processing in main memory. Hence, with respect to the use of a general purpose off-the-shelf XML database, it has improved efficency in query processing and reduce memory requirements.
Enhanced query processing efficiency
Reduced memory requirements
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

16. Evaluation benchmark Three XML guideline sets Variable guideline size
5000 documents (120MB)
10000 documents (240MB)
20000 documents (480MB)
Variable guideline size
min = 2KB
avg = 24KB
max = 125KB
Five different query types
Queries on keywords (structural + textual constraints)
Q1 – keywords in contents
Q2 – keywords in type and contents
Temporal queries (structural + temporal constraints)
Q3 – conditions on validity and transaction time
Mixed queries (structural + textual + temporal constraints)
Q4, Q5 – with keywords and temporal conditions
Experiments have done in order to evaluate the system performance. Three synthetic guidelines sets of increasing size have been created by means of a parametric generator. Five query types have been used mixing in different ways structural, textual, temporal and semantic applicability constraints.
Query variants with semantic constraints
Qx-A – with additional applicability constraints
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

17. Performance evaluation
Very high personalization query efficiency
The system is able to solve personalization problems by means of simple comparisons involving pre-post encodings
0.5-1% more time for the queries with applicability constraints
3-4% storage space overhead
The results of the experiments are very incouraging, showing a good and uniform behavior of the system. In particular, personalization querises are solved very efficiently, also thanks to the ontology encoding which allows simple number comparisons in order to evaluate applicability. The space and time overhead with respect to queries without personalization is very limited.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

18. Performance evaluation
time
1741 msec
1366 msec
1046 msec
5000 docs
10000 docs
20000 docs
Scalability tests
The computing time grows sublinearly with the number of documents
Good scalability of the system in every type of query context
Furtherore, scalability test have shown very good figures with every type of queries. Sublinear scalability and low storage requirements are desirable features for a system which is likely to bear very high concurrency rates in a working enviroment.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

19. Conclusions
We presented our research work concerning the design and implementation of efficient Web-based information systems for multi-version clinical guidelines management
The approach is based on the extension to clinical guidelines of solutions previously developed for e-Government applications
We introduced a personalized access to multi-version guidelines relying on temporal and semantic versioning
We developed a prototype personalization platform for which a specialized Multi-version XML Query Processor has been designed and implemented
We proved our approach to be very efficient in a large set of experimental situations and showed excellent scale-up figures with varying load configurations
In conclusion, we presented our work concerning the design and implementation of multi-version guidelines management system supporting personalization. The approach is based on the extension of solutions previously adopted for eGov applications. Guideline versioning is based on multiple temporal and semantic dimensions. A prototype personalization engine has been implemented and tested. Experiments have shown good performances in a broad range of operating scenarios.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines

20. Future work Extensions of the current framework
more advanced application requirements may include a more sophisticated ontology definition (e.g. relaxing the constraint of tree-like ontologies)
completion of the personalization engine with advanced auxiliary services (e.g. for automatic classification of the patient wrt disease ontology)
management of multiple reference ontologies (e.g. to support other semantic versioning dimensions - context information)
Assessment of our prototype systems in a concrete working environment
with real users and with a large repository of real guidelines
Future work will be devoted to further extensions of the proposed approach. Extensions will consider advanced application requirements including more structurally complex ontologies, introduction of advanced auxiliary services (e.g. a reasoner for automatic matching with ontology concepts), and the adoption of multiple ontologies corresponding to different semantic versioning dimensions. Further work will also consider the evaluation of the proposed approach in a real working environments, with real multi-version guidelines and real users.
HEALTHINF F. Grandi – Representation and Efficient Management of Multi-version Clinical Guidelines