e13073 Background: Incorrect assumptions and poor feasibility are often cited as the major cause of study delay and outcome resolution. Trial placement and recruitment can be optimized by monitoring and regular assessment of trial registries. However, just a snapshot analysis does not take into account retrospective editing of records. This paper evaluates the effectiveness of a robotic, semantic analysis of registered oncology trials to assess individual changes in trial records over time. Methods: Evaluating individual trial records from 2008 to 2012 with state-of-the art natural language processing and semantic linking of data fields within clinical trial registries and literature databases to add increased resolution to erroneous or inadequate registry fields. Results: Substantial modifications are made to registration records over the life cycle of a clinical trial. Quantifiable fields as recruitment, site numbers and end dates all deviate significantly from the study roll-out. Analysis of oncology trials recruiting over 200 patients between 2008 and 2012 shows two thirds of trials deviate from the initial record. Most frequently study end date is delayed (about 50% of all trials) or enrollment targets are amended (at least 25%). Over all oncology indications, enrollment increases are twice more frequent as decreases. However, there are notable exceptions such as lung cancer, where enrollment is frequently decreased. A differential analysis of countries and sites in commercial oncology phase III trials over the period revealed a marked shift of research to sites in France, Spain, the UK and China. However, when sites were added to ongoing trials this general trend was not entirely followed as Poland, Japan, Belgium, Argentina and Switzerland were frequently chosen as additional countries above the normal distribution, suggesting these are preferred countries for trial “rescue”. Conclusions: This paper demonstrates the effectiveness of automated semantic web-mining to identify incorrect clinical trial assumptions and subsequent remediation.
Semantic Web Mining using RDF Data