DICE: Data-Efficient Clinical Event Extraction with Generative Models

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 15898–15917 July 9-14, 2023 ©2023 Association for Computational Linguistics DICE: Data-Efficient Clinical Event Extraction with Generative Models Mingyu Derek Ma∗ Alexander K. Taylor∗ Wei Wang Nanyun Peng Computer Science Department University of California, Los Angeles {ma, ataylor2, weiwang, violetpeng}@cs.ucla.edu Abstract Event extraction for the clinical domain is an under-explored research area. The lack of train- ing data along with the high volume of domain- specific terminologies with vague entity bound- aries makes the task especially challenging. In this paper, we introduce DICE, a robust and data-efficient generative model for clinical event extraction. DICE frames event extraction as a conditional generation problem and intro- duces a contrastive learning objective to accu- rately decide the boundaries of biomedical men- tions. DICE also trains an auxiliary mention identification task jointly with event extraction tasks to better identify entity mention bound- aries, and further introduces special markers to incorporate identified entity mentions as trigger and argument candidates for their respective tasks. To benchmark clinical event extraction, we compose MACCROBAT-EE, the first clinical event extraction dataset with argument anno- tation, based on an existing clinical informa- tion extraction dataset, MACCROBAT (Caufield et al., 2019). Our experiments demonstrate state-of-the-art performances of DICE for clin- ical and news domain event extraction, espe- cially under low data settings. 1 Introduction Event extraction (EE) is an information extraction task that aims to identify event triggers and argu- ments from unstructured texts (Ahn, 2006). The EE task consists of two subtasks: 1) event detec- tion, in which the model extracts trigger text and predicts the event type; and 2) event argument ex- traction, in which the model extracts argument text and predicts the role of each argument given an event trigger and associated event type. Clinical EE aims to extract clinical events, which are occurrences at specific points in time during a clinical process, such as diagnostic procedures, symptoms, etc. The arguments for such events are ∗Equal contribution. A 45 - year - old lady sought dermatology consultation for severely tender erythematous vesicles and bullae over back , chest and arms . Sign symptom Detailed description Texture Biological structure Biological structure Biological structure Event trigger nodule Event type Sign_symptom Detailed description abnormal Area 0.8 x 1.5 cm Biological structure left upper lung lobe Event trigger computed tomography Event type Diagnostic_ procedure Biological structure chest Sign_symptom Diagnostic_procedure A man presented with an abnormal nodule measuring 0.8 x 1.5 cm in the left upper lung lobe imaged through chest computed tomography scanning. Figure 1: Illustration of a SIGN_SYMPTOM event triggered by “nodule” with multiple arguments including an AREA argument “0.8x1.5cm”, and a DIAGNOSTIC_PROCEDURE event whose predicate is “computed tomography” described by argument “chest” of role BIOLOGICAL_STRUCTURE. entities that modify or describe properties of these events (Caufield et al., 2019). Figure 1 shows an example sentence with two clinical events. The overwhelming volume and details of clinical in- formation necessitate clinical EE, which benefits many downstream …

Content truncated to 500 words