MedicInfoSys: An Architecture for an Evidence-Based Medical Information Research and Delivery System

Authors: 

Pif Edwards
Vlado Keselj

Author Addresses: 

Faculty of Computer Science
Dalhousie University
6050 University Ave.
PO Box 15000
Halifax, Nova Scotia, Canada
B3H 4R2

Abstract: 

Medical information is growing at an exponential rate. The majority of physicians information needs are not being met. Present information systems are insufficient. Knowledge-based methods and resources, once brittle and unreliable, have matured. Resources such as the UMLS open promising new avenues for experimentation, new implementations and better relevance performance. This paper explores information systems in the context of Evidence-Based Medicine (EBM) and the information needs of physicians.

In this work we identify 3 primary problems specific to this domain, and propose a solution in the form of an architecture. The first problem is time; physicians spend on the average 2-8 minutes per question and it takes on the average 10-45 minutes to answer all but the most simple clinical query. The answer to this problem is delegation. Just as in the primary care context, physicians often delegate tasks to specialists, the same must be done in the information context: physicians must delegate the information finding tasks to 'informationists' (a.k.a. medical librarians). Studies show questions answered from a central location can be done at an average cost $27.50 per question with an average wait time of 6 hours (via FAX). This fast, inexpensive medical test is shown to increase the average quality of care 47%. The second problem is the average length of queries is 2-3 keywords, which is insufficient for medical question answering. The answer that we propose, an Evidence-Based Medicine (EBM) style "Well-made Question" approach to: a) structure the query for the user; and b) contextualize the query for the system. Furthermore, a structured query prompts the user to first form the question in their mind and thus form better queries. The third problem is the sheer volume of results: 1000s of results for even moderately specific well-formed queries is the norm. We propose a hierarchical categorization of search results. The maturity of knowledge-based resources in the medical domain allows a speedy, trustworthy and customizable categorization.

Our 3-layer architecture details a delivery system that is fashioned around the studied information needs of physicians, and it is not a simple adaptation of existing systems. The end-user layer structures the query, interacts with the Informationist layer and shows the results after passing through each layer. The informationist layer is where the medical information specialist uses information need to skillfully query, browse and form a solution with interaction with the system layer. The system layer filters, categorizes, extracts and presents the information from multiple sources to the informationist layer.

Tech Report Number: 
CS-2009-05
Report Date: 
October 1, 2009
AttachmentSize
PDF icon CS-2009-05.pdf1.34 MB