Clinical Information Retrieval using Document and PICO Structure

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 822–830, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Clinical Information Retrieval using Document and PICO Structure Florian Boudin and Jian-Yun Nie DIRO, Universit´e de Montr´eal CP. 6128, succursale Centre-ville Montr´eal, H3C 3J7 Qu´ebec, Canada {boudinfl,nie}@iro.umontreal.ca Martin Dawes Department of Family Medicine McGill University, 515 Pine Ave W Montr´eal, H2W 1S4 Qu´ebec, Canada martin.dawes@mcgill.ca Abstract In evidence-based medicine, clinical questions involve four aspects: Patient/Problem (P), In- tervention (I), Comparison (C) and Outcome (O), known as PICO elements. In this pa- per we present a method that extends the lan- guage modeling approach to incorporate both document structure and PICO query formu- lation. We present an analysis of the distri- bution of PICO elements in medical abstracts that motivates the use of a location-based weighting strategy. In experiments carried out on a collection of 1.5 million abstracts, the method was found to lead to an improvement of roughly 60% in MAP and 70% in P@10 as compared to state-of-the-art methods. 1 Introduction As the volume of published medical literature con- tinues to grow exponentially, there is more and more research for physicians to assess and evaluate and less time to do so. Evidence-based medicine (EBM) (Sackett et al., 1996) is a widely accepted paradigm in medical practice that relies on evidence from patient-centered clinical research to make decisions. Taking an evidence-based approach to searching means doing a systematic search of all the available literature, individually critically appraising each re- search study and then applying the findings in clini- cal practice. However, this is a time consuming ac- tivity. One way to facilitate searching for a precise answer is to formulate a well-focused and structured question (Schardt et al., 2007). Physicians are educated to formulate their clinical questions according to several well defined aspects in EBM: Patient/Problem (P), Intervention (I), Comparison (C) and Outcome (O), which are called PICO elements. In many documents in medical lit- erature (e.g. MEDLINE), one can find the elements of the PICO structure, but rarely explicitly anno- tated (Dawes et al., 2007). To identify documents corresponding to a patient’s state, physicians also construct their queries according to the PICO struc- ture. For example, in the question “In children with pain and fever how does paracetamol compared with ibuprofen affect levels of pain and fever?” one can identify the following PICO elements: Patient/Problem: children/pain and fever Intervention: paracetamol Comparison: ibuprofen Outcome: levels of pain and fever Very little work, if any, has been carried out on the use of these elements in the Information Retrieval (IR) process. There are several reasons for that. It is not easy to identify PICO elements in documents, as well as in the question if these are not explicitly separated in it. Several studies have been performed on identifying PICO elements in abstracts (Demner- Fushman and Lin, 2007; Hansen et al., 2008; Chung, 2009). However, all of them are reporting coarse- grain (sentence-level) tagging methods …

Content truncated to 500 words