Distributed Knowledge-Based Temporal-Abstraction Mediation Yuval Shahar M D

Distributed, Knowledge-Based Temporal-Abstraction Mediation Yuval Shahar, M. D. , Ph. D. Medical Informatics Research Center Department of Information Systems Engineering Ben Gurion University, Beer Sheva, Israel

The Need for Intelligent Integration of Multiple Time-Oriented Clinical Data • Many medical tasks, especially those involving chronic patients, require extraction of clinically meaningful concepts from multiple sources of raw, longitudinal, time-oriented data – Example: “Modify the standard dose of the drug, if during treatment, the patient experiences a second episode of liver toxicity (Grade II or more) that has persisted for at least two weeks” • Examples of clinical tasks: – Diagnosis • Searching for “a gradual increase of fasting blood-glucose level” – Therapy • Following a treatment plan based on a clinical guideline – Quality assessment • Comparing observed treatments with those recommended by a guideline – Research • Detection of hidden dependencies over time between clinical parameters

The Need for Intelligent Mediation: The Gap Between Raw Clinical Data and Clinically Meaningful Concepts • Clinical databases store raw, time-stamped data • Care providers and decision-support applications reason about patients in terms of abstract, clinically meaningful concepts, typically over significant time periods • A system that automatically answers queries or detects patterns regarding either raw clinical data or concepts derivable from them over time, is crucial for effectively supporting multiple clinical tasks

The Temporal-Abstraction Task • Input: time-stamped clinical data and relevant events (interventions) • Output: interval-based abstractions • Identifies past and present trends and states • Supports decisions based on temporal patterns, such as: “modify therapy if the patient has a second episode of Grade II bone-marrow toxicity lasting more than 3 weeks” • Focuses on interpretation, rather than on forecasting

A Clinical Temporal-Abstraction Example: The Bone-Marrow Transplantation Domain PAZ protocol BMT Expected CGVHD M[0] Platelet counts ² ² • • ( • ) 150 K 100 K 0 50 100 M[1] M[2] M[3] . M[1] ²² ² ² ² • • • 200 Time (days) ² • M[0] Granulocyte counts ² ²² • • • (²) 2000 1000 400

The Bone-Marrow Transplantation Example, Revisited

Uses of Temporal Abstractions: Examples in Bio. Medical Domains • Therapy planning and patient monitoring; E. g. , the EON and De. Gel projects (modular architectures to support guideline-based care) • Creating high-level summaries of time-oriented medical records • Supporting explanation modules for a medical DSS • Representing goals of therapy guidelines for quality assurance at runtime and quality assessment retrospectively; E. g. , the Asgaard project: Guideline intentions regarding both process and outcomes are captured as temporal patterns to be achieved or avoided • Recent use in Italy for detecting patterns in gene expression levels • Visualization of time-oriented clinical data: the KNAVE project

Knowledge-Based Temporal Abstraction (KBTA)

The KBTA Ontology • Events (interventions) (e. g. , insulin therapy) - part-of, is-a relations • Parameters (measured raw data and derived concepts) (e. g. , hemoglobin values; anemia levels) - abstracted-into, is-a relations • Patterns (e. g. , crescendo angina; quiescent-onset GVHD) - component-of, is-a relations • Abstraction goals (user views)(e. g. , therapy of diabetes) - is-a relations • Interpretation contexts (effect of regular insulin) - subcontext, is-a relations • Interpretation contexts are induced by all other entities

Temporal-Abstraction Output Types • • State abstractions (LOW, HIGH) Gradient abstractions (INC, DEC) Rate Abstractions (SLOW, FAST) Pattern Abstractions (CRESCENDO) - Linear patterns - Periodic patterns

Temporal-Abstraction Knowledge Types • Structural (e. g. , part-of, is-a relations) - mainly declarative/relational • Classification (e. g. , value ranges; patterns) - mainly functional • Temporal-semantic (e. g. , “concatenable” property) - mainly logical • Temporal-dynamic (e. g. , interpolation functions) - mainly probabilistic

Dynamic Induction of Contexts: Temporal Constraints Between Inducing Proposition and Induced Context (Shahar, AMAI 1998) ee ss es se

Induction of Interpretation Contexts

The Meaning of Interpretation Contexts • Context intervals serve as a frame of reference for interpretation: Abstractions are meaningful only in a context (e. g. , “anemia in a pregnant woman”) • Context intervals focus and limit the computations to only those relevant to a particular context (thus, knowledge is brought to bear only when relevant) • Contexts enable the use of context-specific knowledge, thus increasing accuracy of resultant abstractions

Advantages of Explicit Contexts • Any temporal relation (e. g. , overlaps) can hold between a context and its inducing proposition; contexts can be induced before and after the inducing proposition (thus enabling a certain type of hindsight and foresight) + Note: Forming contexts is a finite process • The same context-forming proposition can induce multiple context intervals • The same interpretation context might be induced by different propositions • Explicit contexts support maintenance of several concurrent views (or interpretations) of the data, in which the same parameter has different values at the same time, each within a different context + Note: No contradiction--values are in different contexts

Local and Global Persistence Functions: Exponential-Decay Local Belief Functions (Shahar, JETAI 1999) t j 1 j 2 I 1 Bel(j) I 2 1 jth 0 Time

Abstraction of Periodic Patterns Periodic Pattern Linear Component Fever Anemia Week 1 Temperature Hemoglobin Level Fever Linear Component Fever Anemia Week 2 Anemia Fever Anemia Week 3

The RÉSUMÉ System Architecture. Temporal-abstraction mechanisms Domain TA knowledge base Temporal fact base E v e n ts Event ontology C o n te x ts Context ontology A b s tr a c te d in te r v a ls Parameter ontology P r im itiv e d a ta External patient database Events Primitive data • + + • • +

Application Domains for the KBTA Method (Shahar & Musen, 1993, 1996; Shahar & Molina 1999; Boaz and Shahar 2005; Shabtai, Shahar, and Elovic, 2006) • Medical domains: – Guideline-based care • AIDS therapy • Oncology – Monitoring of children’s growth – Therapy of insulin-dependent diabetes patients • Non-medical domains: – – Evaluation of traffic-controllers actions summarization of meteorological data Integration of intelligence data over time Monitoring electronic security threats in computers and communication networks

Monitoring of Children’s growth: The Parameter Ontology

Monitoring of Children’s growth: Temporal Abstraction of the Height Standard Deviation Score (HTSDS)

The Diabetes Parameter Ontology = PROPERTY-OF relation; = IS-A relation; = ABSTRACTED_INTO relation

The Diabetes Event Ontology = PART-OF relation; = IS-A relation

The Diabetes Context Ontology = SUB-CONTEXT relation; = IS-A relation

Forming Contexts in Diabetes

Acquisition of Temporal-Abstraction Knowledge (Shahar et al. , JAMIA, 1999)

Evaluation of Automated Knowledge Entry • Formal evaluation performed, using – 3 experts, 3 knowledge engineers, 3 clinical domains – a gold standard of data, knowledge and output abstractions • Domains: – monitoring of children’s growth – care of diabetes patients – protocol-based care in oncology and AIDS • The study evaluated the usability of the KA tool solely for entry of previously elicited knowledge

KA Tool Evaluation: Results • Understanding RÉSUMÉ required 6 to 20 hours (median: 15 to 20 hours) • Learning to use the KA tool required 2 to 6 hours (median: 3 to 4 hours) • Acquisition times for physicians varied by domain: 2 to 20 hours for growth monitoring (median: 3 hours), 6 and 12 hours for diabetes care, and 5 to 60 hours for protocol-based care (median: 10 hours) • A speedup of up to 25 times (median: 3 times) was demonstrated for all participants when the KA process was repeated • On their first attempt at using the tool to enter the knowledge, the knowledge engineers recorded entry times similar to those of the second attempt of the expert physicians entering the same knowledge • In all cases, RÉSUMÉ, using knowledge entered via the KA tool, generated abstractions that were almost identical to those generated using the same knowledge, when entered manually

Editing The KBTA Ontology in Protégé 2000

Temporal Reasoning and Temporal Maintenance • Temporal reasoning supports inference tasks involving time-oriented data; often connected with artificial-intelligence methods • Temporal data maintenance deals with storage and retrieval of data that has multiple temporal dimensions; often connected with database systems • Both require temporal data modelling

Examples of Temporal-Maintenance Systems • TSQL 2, a bitemporal-database query language (Snodgrass et al. , Arizona) • TNET and the TQuery language (Kahn, Stanford/UCSF) • The Chronus/Chronus 2 projects (Stanford)

Examples of Temporal-Reasoning Systems • RÉSUMÉ • M-HTP • TOPAZ • Tren. Dx

Temporal Data Manager • Performs – - Temporal abstraction of time-oriented data – - Temporal maintenance • Used for tasks such as finding in a patient database which patients fulfils the guideline eligibility conditions (expressed as temporal patterns), assessing the quality of care by comparison to predefined timeoriented goals, or visualization of temporal patterns in the patient’s record

Two Possible Implementation Strategies 1) Extend the DBMS 2) Extend the Application

Problems in Extending The DBMS Temporal data management methods implemented in a DBMS: § are limited to producing very simple abstractions § are often database-specific

Problems in Extending the Application Temporal data management methods implemented in applications: § duplicate some of the functions of the DBMS § are application-specific

Our Strategy • Separates data management methods from the application and the database • Decomposes temporal data management into two general tasks: Application Temporal Abstraction Temporal Querying – temporal abstraction – temporal maintenance Database

The Tzolkin Temporal-Mediator Architecture [Nguyen, Shahar et al. , 1999] Application Query Results Tzolkin Knowledge Base Temporal Abstraction Module Temporal. Querying Module Abstraction Knowledge Database

The IDAN Temporal-Abstraction Mediator (Boaz and Shahar, 2003, 2005) Knowledge Service Knowledgeacquisition tool Medical Expert Standard Medical Vocabularies Service Data Access Service Temporal. Abstraction Controller KNAVE-II Clinical User Temporal Abstraction Service (ALMA)

Adding a New Clinical Database to The IDAN Mediator Architecture • Due to local variations in terminology and data structure, linking to a new clinical database requires creation of – A schema-mapping table – A term-mapping table – A unit-mapping table • The mapping tools use a vocabulary-server search engine that organizes and searches within several standard controlled medical vocabularies (ICD-9 -CM , LOINC, CPT, SNOMED, NDF) • Clinical databases are mapped into the standard terms and structure that are used by the clinical knowledge base, thus making the knowledge base(s) highly generic and reusable • The overall mapping methodology has been implemented within the Medical Database Adaptor (MEIDA) system [German, 2006]

The LOINC Server Search Engine

LOINC Search Results

Accessing Local Data Sources Local data source site Term mapping table 2: get local term and unit (Std. Term ) 3: Local. Term, Local. Unit 1: Data request ( 4: Data request( Patient, Std. Term, Out. Unit ) 5: Data 9: Result (DAM) ) Virtual schema ? adaptor access module Patient, Local. Term Transformation 6: get transformation function( Local. Unit, Out. Unit ) functions library Unknown 7: Trans. Func schema 8: Result = transform (Data, Trans. Func )

Summary: Knowledge-Based Abstraction of Time-Oriented Data • Temporal abstraction of time-oriented data can employ reusable domain-independent computational mechanisms that access a domain-specific temporal-abstraction ontology • Temporal abstraction is useful for monitoring, therapy planning, data summarization and visualization, explanation, and quality assessment • The IDAN distributed temporal mediator mediates and coordinates queries to the knowledge base and to the database • Current and future work: – Continuous temporal abstraction - The Momentum architecture [Spokoiny and Shahar, 2004, in press] – Probabilistic temporal abstraction (PTA) [Ramati and Shahar, 2005]