Search

Based on previous experience from working on a task-based search engine, we present a list of suggestions and ideas for an Information Retrieval (IR) framework that could inform the development of next generation professional search systems. The specific task that we start from is the clinicians' information need in finding rare disease diagnostic hypotheses at the time and place where medical decisions are made. Our experience from the development of a search engine focused on supporting clinicians in completing this task has provided us valuable insights in what aspects should be considered by the developers of vertical search engines.

Task-based search addresses situations where standard off-the-shelf Information Retrieval (IR) technology may not suffice to satisfy users in their tasks. In these situations, IR systems should be tailored to the user’s task-specific needs and requirements. One such task is searching for rare disease diagnostic hypotheses in the domain of medical IR. In this work, we build upon an existing vertical medical search engine, Zebra, that is focused on rare disease diagnosis. In previous work, Zebra has been evaluated using real-life medical cases of rare and difficult diseases, and has been found to be a useful and competitive tool for clinicians. In this work, we extend Zebra’s functionalities to optimise the task of medical diagnosis through search as follows: we add the option of grouping retrieved documents into clusters based on disease name occurrence, and we offer a ‘disease-ranking’ option, in addition to the standard ‘document-ranking’ option. This paper presents and discusses these functionalities.

Increasingly more clinicians use web Information Retrieval (IR) systems to assist them in diagnosing difficult medical cases, for instance rare diseases that they may not be familiar with. However, web IR systems are not necessarily optimised for this task. For instance, clinicians’ queries tend to be long lists of symptoms, often containing phrases, whereas web IR systems typically expect very short keyword-based queries. Motivated by such differences, this work uses a preliminary study of 30 clinical cases to reflect on rare disease retrieval as an IR task. Initial experiments using both Google web search and offline retrieval from a rare disease collection indicate that the retrieval of rare diseases is an open problem with room for improvement.

Increasingly more clinicians use web Information Retrieval (IR) systems to assist them in diagnosing difficult medical cases, for instance rare diseases that they may not be familiar with. However, web IR systems are not necessarily optimised for this task. For instance, clinicians’ queries tend to be long lists of symptoms, often containing phrases, whereas web IR systems typically expect very short keywordbased queries. Motivated by such differences, this work uses a preliminary study of 30 clinical cases to reflect on rare disease retrieval as an IR task. Initial experiments using both Google web search and offline retrieval from a rare disease collection indicate that the retrieval of rare diseases is an open problem with room for improvement.