Domain Specific Multi-stage Query Language for Medical Document Repositories

Transcription

1 Domain Specific Multi-stage Query Language for Medical Document Repositories S. Bhalla Model Driven Querying 9/23/2013 VLDB Phd Workshop

2 Specialized Domains Introduction Biomedical, agriculture, medical/healthcare Require Effective search and query mechanisms Insufficient Search-engine like Key-words based searches Medical domain Complex Example, Query Information about general AIDS information medical search tool, such as PubMed Result 1000s of documents Different aspects of AIDS are displayed Such as, treatment, drug therapy, transmission, diagnosis, and history Medical professionals Specific technical articles (particular topic ) General public General information (disease or medicine). How to retrieve medical query related information 9/23/2013 VLDB Phd Workshop

3 Medical Information Introduction (1) Knowledge Evolved over 10s of years Contains Well defined terms and processes Available on the Web Patient Specific Information Knowledge-based Information Web Documents Medical Literature Patient-encounter Recordings 9/23/2013 VLDB Phd Workshop

4 Introduction (2) Complexity Knowledge-based Resources Heterogeneous End-user groups Patients, researchers, doctors and other experts Variation Information Requirements Patient-treatment, self -diagnosis, general health information Structure Medical Documents Scientific papers, encyclopedias and other literature Unique, well-defined 9/23/2013 VLDB Phd Workshop

5 Introduction (3): Specialized Documents Case of medical encyclopedias Comprehensive medical guide Patients and clinicians Authoritative source NLM (National Library of Medicine) Paper based resources Electronic format Frequently referred Medical domain users Example, MedlinePlus, WebMD, ADAMS, Merriam-Webster Medical Dictionary 9/23/2013 VLDB Phd Workshop

6 Introduction (4): Why Query External knowledge base Clinicians Evidence based medicine During different stages of point-of-care Assessment plan of treatment Patient diagnosis Improve Quality of Care Authoritative information required Self Diagnosis During Early appearance of symptoms Personal Knowledge Patients and their relatives 9/23/2013 VLDB Phd Workshop

7 The Hierarchical Structure The Underlying Structure Topic of the Document Subtopics Miscellaneous/Related Content Subtopic 1 Subtopic 2 Subtopic n Content topic 1 Content topic 2 Content topic n Content Content Content Content Content Content Flow of Contents Organized as stages of point-of-care 9/23/2013 VLDB Phd Workshop

8 Introduction (5): The End Users Variable a. Demographical Characteristics b. Tasks/Purpose c. Computer/Domain Expertise Practitioners and Researchers Well-versed Domain knowledge and terminologies Require Precise, complete, accurate and timely results Patients and their relatives NOT Well-versed Domain knowledge and terminologies Require General information Healthcare Workers Specialized Researchers Patients, their relatives 9/23/2013 VLDB Phd Workshop

9 The Medical Queries Evidence-based Queries Intent: Diagnostic Raised by: Clinicians/Experts Target resources: Online Medical Repositories (e.g. medical encyclopedia) Example: Cases where helicobacter bacteria causes peptic ulcer Hypothesis-directed Queries Intent: Non-diagnostic Raised by: Novice users/patients Target resources: Online Medical Repositories (e.g. medical encyclopedia) Example: Treatment in case of high fever and dizziness 9/23/2013 VLDB Phd Workshop

10 The Query Flows Occurrence Evidence-based and hypothesis-directed queries Represent Stages of information seeking Comprise Varying levels of query complexity /23/2013 VLDB Phd Workshop

11 The Research Gap Healthcare Expert Paper-based resources Help Needed Query: Find chances of "Cancer Risk" in patients showing symptom "Sleep Deprivation" and have been exposed to "Radiation" (but not "Environmental Toxins" and does not have "Genetic Disorder"). Results Large in number, irrelevant Failure Keyword search, domain-specific search tools Require Precise and easy-to-use database style query methods Key steps: 1. Schema understandable by users 2. Identify Resources to query 3. Identify Granularity of results 9/23/2013 VLDB Phd Workshop

12 Bridging the Gap Aim: Effective Online Medical Information Transform Document Repository User-Level Schema Enable High-level Query Language Target Audience Skilled and semi-skilled users Utilize Query capabilities of a database query language Assist Domain Experts Using Query language Facilitate In-depth Queries Granular Results 9/23/2013 VLDB Phd Workshop

13 Resource Querying the New Way Results returned - Lack specificity - Long list of full documents - Trustworthiness of resources unknown User-level Schema Resource Search Method High-level Query language Query Method Results returned - Specific, granular - Segments of documents query criteria - Trustworthy/Authoritative sources only Medical Expert Traditional Method Proposed Method 9/23/2013 VLDB Phd Workshop Medical Expert

14 Proposed Approach 9/23/2013 VLDB Phd Workshop

15 Key Features User-Level Schema Universal, concept-level schema Attributes Understandable Domain experts and novice users Query-able Granular results Multi-stage Query Language Map multi-stage diagnostic process Step-by-step Query Flow Interactive Querying View Results Add concept Continuous query refinement Supported Queries Simple, Medium, Complex, Recursive 9/23/2013 VLDB Phd Workshop

16 Outline Two-step framework Offline Process Create User-level schema Online Process Enable Multi-stage Query Language Offline Process Use Web segmentation algorithm, Domain concepts Result Automatic creation of a User-level schema Online Process Enable Multi-stage Query Language Use User-level schema Results Granular, segment-level, context-based 9/23/2013 VLDB Phd Workshop

17 The Method Two-step Framework 9/23/2013 VLDB Phd Workshop

18 Data Model 9/23/2013 VLDB Phd Workshop

19 Data Model Tree Structured Repository H1 f1 f2 Example: MedlinePlus Medical Encyclopedia 9/23/2013 VLDB Phd Workshop

20 Data Model (1) 9/23/2013 VLDB Phd Workshop

21 Data Model (2):The Schema Attributes Diagnostic concepts/terms Understandable by expert and novice users Do not change frequently 9/23/2013 VLDB Phd Workshop

22 Data Model (3): A XML Document Document corresponding to Aarskog Syndrome Title Causes Symptoms Treatment Example: MedlinePlus Medical Encyclopedia 9/23/2013 VLDB Phd Workshop

23 Data Model (4): Query Effort Query: Find if "Oxygen therapy" work for the treatment of "Chronic Respiratory Failure" and symptoms are "Lethargy" OR "Shortness of breath. Not Possible Easy-to-Use Result segment Advanced search: MedlinePlus Proposed Method SELECT attribute = Disease_name WHERE Attribute Disease_name = Chronic Respiratory Failure AND Attribute Treatment = Oxygen therapy AND Attribute Symptoms = Lethargy OR Attribute Symptoms = Shortness of breath Context of user-query 9/23/2013 VLDB Phd Workshop

24 Data Model (5): Granular Results Each result is a segment, combination of Concept/context in query Item of concern (content enclosed in a segment) Context Queried Attributes/Segments Granular Query Results 9/23/2013 VLDB Phd Workshop

25 An Example Query: Find other symptoms where chronic kidney failure is caused by anemia Queried segment Symptoms Segments in Query Causes = anemia and Disease_name = Chronic kidney failure Result Segment Symptoms Context disease_name = chronic kidney failure & causes = anemia (SYMPTOM - segment) 9/23/2013 VLDB Phd Workshop

26 Next Step Multi-stage Query Language 9/23/2013 VLDB Phd Workshop

27 Proposed Query Language (1) XQBE Medical document repositories Create queries Drag and drop interface Query : Find cases where a person is inflicted with peptic ulcer due to helicobacter pylon bacteria Attributes understandable by end users 1. Case = disease_name Value =?? 2. Due to = Causes Value = helicobacter pylon bacteria 3. Inflicted with = Symptoms Value = peptic ulcer Query Effort Minimal learning curve Computer-expertise not required 9/23/2013 VLDB Phd Workshop

28 Proposed Query Language (2) Multi-stage Query-by-Concept Concept Query-able attribute Topic, sub-topic, medical concept Query Effort Dynamic selection of attributes No computer expertise Query Process An Example: Cases where fever is caused due to infliction of Pneumonia and Tuberculosis 9/23/2013 VLDB Phd Workshop

29 Another Example Query: Find cases where 3 clinical concepts ( cough, no sore-throat, and had no sterol injection ) occur in context of symptoms occur along with a concept having sub-key (i.e. non sterol injection at the left side ) Possible XQBE on Specialized Medical Repositories Possible Multi-stage Query-by-concept Query Language 9/23/2013 VLDB Phd Workshop

30 Data Sets Evaluation Plan MedlinePlus document repository Health topics (900+), encyclopedia (4000+), drugs (12000+) Set of Queries 50 test queries (multi-staged) Using diseases and medication etc. Quantitative Studies Evaluation Metrics Accuracy of segment extraction (schema creation) Precision and Recall Reduction in search space Qualitative Studies Usability Studies Actual End-users Query Performance 9/23/2013 VLDB Phd Workshop

31 Initial Achievements HTML XML as per proposed model XQuery on XML Integration with XQBE Query by concept Enumeration using paper and pencil 9/23/2013 VLDB Phd Workshop

32 Challenges Scalability Similarly structured repositories List Query operations needed Implementation Above query operations Query Language User Interface 9/23/2013 VLDB Phd Workshop

33 Related Work Domain-specific Information Retrieval Similarity and popularity based models Insufficient for domain experts Information granulation needs to be considered in huge document repositories Form-based Query Interfaces Easy-to-use Limited access to the database Complex queries large number of forms Varying medical concepts large number of fields in forms Beyond single page web search results Provide granular results for user s search Return segments from multiple or related web documents as results High-level Graphical Query Languages Easy-to-use and understand Little or no programming effort required by the user Common languages QBE, XQBE 9/23/2013 VLDB Phd Workshop

34 Summary and Conclusions Proposed Multi-stage Query Language 1. Aim Effective online medical resources 2. Key feature User-Level Schema 3. Facilitates Granular/Context-based Results 4. Support Healthcare Experts 5. Minimize Learning curve for novice users 6. Reduce Dependency on general-purpose search engines Provide Web user level activity no or little programming effort Healthcare experts 9/23/2013 VLDB Phd Workshop

35 References (1) [1] D. Braga, A. Campi, and S. Ceri. Xqbe (xquery by example): A visual interface to the standard xml query language. ACM Trans. Database Syst., 30(2): , June [2] D. Cai, S. Yu, J.-R. Wen, and W.-Y. Ma. Extracting content structure for web pages based on visual representation. In Proceedings of the 5th APWeb, pages Springer-Verlag, [3] M.-A. Cartright, R. W. White, and E. Horvitz. Intentions and attention in exploratory health search. In Proceedings of the 34th Intl. ACM SIGIR conference, pages 65 74, New York, NY, USA, 2011.ACM. [4] S. Cohen, Y. Kanza, Y. Kogan, W. Nutt, Y. Sagiv, and A. Serebrenik. Equix-a search and query language for xml. Journal of the American Society for Information Science and Technology, 53:2002, [5] S. M. Freire, E. Sundvall, D. Karlsson, and P. Lambrix. Performance of XML Databases for Epidemiological Queries in Archetype-Based EHRs. In Proceedings Scandinavian Conference on Health Informatics 2012, volume 70 of Linkping Electronic Conference Proceedings, pages Linkping University Electronic Press, [6] M. Gschwandtner, M. Kritz, and C. Boyer. Requirements of the health professional research. In Technical Report D Khresmoi Project, [7] A. Hanbury. Medical information retrieval, an instance of domain. In SIGIR'12. ACM, August [8] S. Hunt, J. J. Cimino, and D. E. Koziol. A comparison of clinicians s access to online knowledge resources using two types of information retrieval applications in an academic hospital setting. J Med Libr Assoc, 101(1):26 31, [9] [10] M. Jayapandian and H. V. Jagadish. Automating the design and construction of query forms. ICDE, page 125, [11] F. Li and H. V. Jagadish. Usability, databases, and hci. IEEE Data Eng. Bull., 35(3):37 45, [12] [13] A. Marian and W. Wang. Flexible querying of personal information. IEEE Data Eng. Bull., 32(2):20 27, [14] [15] [16] Choice-OpenEHR-persistence-layer S ?qid=208adbca-fc26-4ada-bf02-7efe5a9e5661&trk=group_most_recent_rich-0-b-ttl&goback=%2Egmr_144276, /23/2013 VLDB Phd Workshop

36 References (2) [17] [18] S. A. Rahman, S. Bhalla, and T. Hashimoto. Query-by-object interface for information requirement elicitation in m-commerce. Int. J. Hum. Comput. Interaction, 20(2): , [19] X. Y. Raymond, Y. Lau, D. Song, X. Li, and J. Ma. Toward a semantic granularity model for domain-specific information retrieval. ACM Trans. On Information Systems., 29(3), July [20] S. Sachdeva and S. Bhalla. Implementing high-level query language interfaces for archetype-based electronic health records database. In COMAD, [21] [22] R. Varadarajan, V. Hristidis, and T. Li. Beyond single-page web search results. IEEE Transactions on Knowledge and Data Engineering, 20(3): , [23] A. Yasir, M. Kumara Swamy, P. Krishna Reddy, and S. Bhalla. Enhanced query-by-object approach for information requirement elicitation in large databases. In Big Data Analytics, volume 7678 of Lecture Notes in Computer Science, pages Springer, /23/2013 VLDB Phd Workshop

37 Questions 9/23/2013 VLDB Phd Workshop