An Automated Method for Gender Information Identification from Clinical Trial Texts

Gender is fundamental and essential information for eligibility criteria electronic prescreening aiming for recruiting appropriate target population for human studies. Current commonly applied gender architecture contains the problems of incompleteness and ambiguity particularly on transgender. This study designs a flexible and extensible virtual population gender architecture for enhancing trial recruitment. We also propose an automated method for high accurate transgender identification and validation. The method defines and identifies transgender features from free clinical trial text. After that, we apply a context-based strategy to obtain final gender summary. The experiments are based on clinical trials from ClinicalTrials.gov, and results present that the method achieves a True Positive Rate of 0.917 and a True Negative Rate of 1.0 on the clinical trial text, demonstrating its effectiveness in transgender identification.