CLIPS: An interactive retrieval system for clinical trial studies with context-dependent protocol elements (Preprint)

BACKGROUND Clinical trial protocol includes all the steps that should be conducted in clinical trial and all clinical trial start with development of the protocol. And the protocol is kinds of procedure manual that consists of appropriate combination of related elements, so one element selection generally affects the next element selection. To development the protocol, researchers refer a published guides or electronic protocol templates. The detail elements of the protocol are designed based on the previous data which is retrieved separately in public database of clinical trial. Specifically, researchers retrieve for preliminary information using text based user input keywords, then they extract the elements. However, the current retrieve approaches do not provide a method to interactively select for a combination of the elements. OBJECTIVE The purpose of our study is to provide context-dependent protocol element selection system for the development of objective and successful clinical trial protocols. In detail, we construct a database that can retrieve protocols by combined analysis of elements. Moreover, we develop a web-based interactive protocol element wise selection application using constructed database. METHODS We have constructed a key-value type database for searching element combinations. To build the database, we have defined structure of protocol within five factors; design, subject, variables, statistical issue and descriptions. In this study, five factors are defined as a subset of protocol elements and we have manually classified them from collected information in public database. Then, we have developed a web application to implement interactive method for retrieving selected combination of protocol element. The application in the form of a connected tree provides options to select the next element according to the decision of previous element. Therefore, a researcher can retrieve the structure of the combined protocol. Also, the application supports a function for retrieve various selected protocol structure at the same time. RESULTS We have developed a database and search application for protocol structure retrieval. The database is built on individual protocol information extracted from previous 184,634 clinical trials and provides 13,210 integrated structural information. Furthermore, the database contains various semantic information of the protocols to filter protocols in search application. We did technical validation for evaluating of the database with ‘Cancer and Other Neoplasms’ category of clinicaltrials.gov. By comparing with the clinicaltrials.gov our method has better performance in predicting phenotypic features. Our F1 score was 0.515, while F1 score of clinicaltrials.gov was 0.377. Finally, we developed web application, CLinical trial protocol database System(CLIPS), which provides users with search our database interactively based on protocol elements. CLIPS is available at http://corus.kaist.edu/clips. CONCLUSIONS We have developed database and application to interactively search clinical trial protocols efficiently. We believe that our system is helpful for conducting a new clinical trial. Furthermore, we expect that the database would be utilized as meta-analysis in clinical trials for various purposes. CLINICALTRIAL None