Extracting Intrauterine Device Usage from Clinical Texts Using Natural Language Processing

Abstract

Intrauterine devices (IUDs) are highly-effective contraceptive methods for preventing unintended pregnancy and related adverse outcomes. Clinical Decision Support (CDS) systems could aid care providers in identifying patients at risk for pregnancy due to lack of contraceptive use. However, research suggests that this information is not reliably documented in structured data fields for query, but rather in the clinical notes. As a first step towards developing a robust CDS tool to identify high-risk patients for contraceptive counseling, we developed a clinical information extraction tool, EasyCIE, that readily identifies mentions of IUD usage and classifies whether a note contains evidence that an IUD is present or not for review by domain experts. In this preliminary study, EasyCIE produced high recall and excellent precision distinguishing notes of patients with current IUD usage from notes of patients with historical or no usage.