Formulating Queries for Assessing Clinical Trial Eligibility

This paper introduces a system that processes clinical trials using a combination of natural language processing and database techniques. We process web-based clinical trial recruitment pages to extract semantic information reflecting eligibility criteria for potential participants. From this information we then formulate a query that can match criteria against medical data in patient records. The resulting system reflects a tight coupling of web-based information extraction, natural language processing, medical informatic approaches to clinical knowledge representation, and large-scale database technologies. We present an evaluation of the system and future directions for further system development. 1 Background and overview Researchers design information extraction systems to perform various tasks, and these tasks require various levels of linguistic processing. Some systems are only concerned with parsing out the extracted information and therefore only require the use of a syntactic parser. Others need more in-depth processing and include a semantic component that can give some meaning to the extracted information. Yet other systems are dependent on real-world knowledge and require a pragmatic component to relate the data gathered from the system to outside information. One area receiving recent attention is the medical domain. Much of the natural language processing (NLP) research done with medical literature has involved developing systems that extract different types of relationships from text. For example, NLP techniques have been used on Medline abstracts to extract information on genes, proteins, acronyms, and molecular binding relationships. For its part, the field of medical informatics has produced large-scale resources, largely in database format, that specify the vast knowledge required for medical research and patient services. Highly specialized tools for representing clinical information and patient data have also been developed. Unfortunately, there has been only a modest amount of crossover between the NLP and medical informatics fields. The topic of information extraction is a salient one for demonstrating how applications can leverage the developments from both fields. This paper describes our approach to identification, extraction, and query formulation of information regarding medical clinical trials. Figure 1 shows an overview of the system. In Step 1, extraction and formula generation, we extract patient criteria from a web-based natural language description of qualifications for clinical 1 See http://www.medlineplus.gov. 2 This work was partially funded under National Science Foundataion Information and Intelligent Systems grant IIS-0083127. See also www.deg.byu.edu.