What can be improved? Identifying actionable items from patient narratives.

Recent literature suggests the potentiality of patient narratives to evaluate the quality of healthcare services, but little research has investigated the “actionability” of such narratives in driving or reshaping the practices. Moreover, the colloquial and implicit nature of natural language expressions poses challenges to automatic actionable item identification (AII) from large-scale patient narratives. In this paper, we propose a Content Sense and Inference BiLSTM (CSI-BiLSTM) model to address the challenges, and the model includes the element-aware network (EAN) and the aspect indicator network (AIN). The EAN is designed to sense and capture the mentioned degree of the key elements of input sentences to generate effective element-aware sentence representations. The AIN is endowed with reasoning ability for implicit expressions by inferring whether given opinion words are aspect indicators, to further enhance the accuracy and robustness of the model. To the best of our knowledge, this is the first study to automatically identify actionable items from patients’ online narratives. Experiment results on two real-world datasets demonstrate the performance of the proposed model outperforms the competitive baselines and extensive analysis reveals the effectiveness of our model in identifying actionable items.