Joan Bartlett
Martin Dawes (PI)
Roland Grad
Lorie Kloda
Ann McKibbon
Jian-Yun Nie
Pierre Pluye
Laura Shea


While clinicians may need the latest experimental evidence published in the format of a research article to make a decision about the selection of an appropriate therapy, students often will have questions answered about background information answered more effectively from textbooks (Dawes and Uchechukwu 2003). Student questions can often be answered using single word entries in a textbook such as “asthma”.

Finding a specific research article requires a more complex structure. This phenomenon is also seen in other areas, with domain knowledge and expertise affecting the type of information required, as well as the information-seeking process (Marchionini, Dwiggins et al. 1993). Even during the early stages of electronic resources such as MEDLINE it was recognized that making questions more precise reduced the number of articles retrieved (Richardson, Wilson et al. 1995). Consequently over the last 10 years health professionals using electronic information resources have been encouraged to make their questions more explicit (Dawes, Pluye et al. 2007; Schardt, Adams et al. 2007).

They have been taught to identify the key elements of Patient/Population/Problem, Exposure/Intervention, Comparison, Outcome, Duration, Results  (PECODR) in questions. However most searches by clinicians are still single word with only 12% containing a Boolean operator such as “AND” (Meats, Brassey et al. 2007). One possible explanation for this might be that while the use of the structured question does improve search specificity (Schardt, Adams et al. 2007) it does not enable searchers to see the individual PECODR elements clearly in the results of their search.

PECODR Elements,	Example of structured question
P Patient/Population/Problem,	In a 56 year old man with hypertension
E Exposure,	Does Atenolol
C Comparison,	Compared with Placebo
O Outcome,	Reduce Cardiovascular events
D Duration,	of exposure/follow up Over 5 years
R Results,	With a number needed to treat of 25

What is evident from the research to date is that a search engine that could match the PECODR elements in the questions to the PECODR elements in the abstracts requires a controlled vocabulary, developed from taxonomies, to be produced for each of the PECODR elements. Much of this work has already been undertaken. However it also clear that there are missing taxonomy particularly in the exposure elements that will need to be addressed.
Once taxonomies are identified the next step is to automatically index these to the electronic information resources. This cannot rely purely on a simple word identification process. Instead, our approach will also identify the PECODR element related to each index term. For example in our previous work we found words that in conjunction with an “exposure” term indicated that this was more likely to be the comparison element than an exposure element. As indicated above the proximity of a negative word such as “not” or “excluded” needs to be considered. Thus, the resulting index terms will support retrieval based not simply on the presence of specific concepts, but also on the context within which each topic is discussed. This is a difficult area of our project. However, with our work so far and the composition of our research team including researchers from many different fields we feel that we can enhance the current indexing process using these taxonomies. In addition, we possess the expertise to evaluate the impact of such a system.

While online generic search engines such as Google aim to meet the needs of lay users, indexing systems satisfy the needs of traditional users, i.e., researchers and librarians. Both these systems rarely satisfy the needs of clinical (physicians, nurses, and other health professional) users. The present research  aims to improve the performance of information retrieval technology for health professionals, and ultimately enhance traditional indexing systems by providing a new conceptual approach applicable across not only health but also all information sources. To provide a more user specific search, a form of semantic tagging or understanding, corresponding to an established structured framework for questions, is required, so that we can recognize more precisely what is described in the user query and a document. This research project aims to use, and where necessary add to, established taxonomies to develop a new controlled vocabulary for clinical questions. We will then use this vocabulary to enhance the current indexing of a selected subset of articles and test the effectiveness of such an enhanced index.

Research Plan