Identifying Patients with Pain in Emergency Departments using Conventional Machine Learning and Deep Learning

Identifying Patients with Pain in Emergency Departments using Conventional Machine Learning and Deep Learning Thanh Vu1, Anthony Nguyen1, Nathan J Brown2,3, James Hughes2,4 1 Australian e-Health Research Centre, CSIRO, Brisbane, Australia 2 Emergency & Trauma Centre, Royal Brisbane & Women’s Hospital, Brisbane, Australia 3 Faculty of Medicine, University of Queensland, Brisbane, Australia 4 School of Nursing, Queensland University of Technology, Brisbane, Australia {thanh.vu, anthony.nguyen}@csiro.au {nathan.brown3, james.hughes}@health.qld.gov.au Abstract Pain is the main symptom that patients present with to the emergency department (ED). Pain management, however, is often poorly done aspect of emergency care and patients with painful conditions can endure long waits be- fore their pain is assessed or treated. To im- prove pain management quality, identifying whether or not an ED patient presents with pain is an important task and allows for further investigation of the quality of care provided. In this paper, machine learning was utilised to handle the task of automatically detecting pa- tients who present at EDs with pain from retro- spective data. Experimental results on a manu- ally annotated dataset show that our proposed machine learning models achieve high perfor- mances, in which the highest accuracy and macro-averaged F1 are 91.00% and 90.96%, respectively. 1 Introduction There are over 8 million presentations to Aus- tralian public hospital emergency departments (EDs) each year (AIHW, 2018). Pain is the most common symptom for patients seeking care in EDs (Karwowski-Souli´e et al., 2006; Hather- ley et al., 2016; Todd, 2017; Varndell et al., 2018). In particular, a study of 726 presentations (Karwowski-Souli´e et al., 2006) showed that 78% of patients presented to EDs with pain. Despite the large number of ED patients with pain, pain is often poorly assessed and treated within the ED (Hatherley et al., 2016; Varndell et al., 2018). This leads to increases in waiting times for patients un- til pain is assessed and pharmacological analgesia is offered (Hatherley et al., 2016; Varndell et al., 2018). In this paper, we propose the task of identifying patients presenting to EDs with pain, with the view of improving care quality and the management of pain. In particular, the identification of patients presenting to EDs with pain allows for the scop- ing of groups who may be receiving poor pain care (Pletcher et al., 2008; Hwang et al., 2014; Todd, 2017). The ease of identification of patients with pain also allows for the evaluation of targeted in- terventions to improve care using large datasets. However, manually handling the identification task at a large scale, such as tens, hundreds of thousands of ED patients is challenging as it requires expensive human effort to determine whether a patient presented to the ED with pain or not. To handle this problem, we propose to use machine learning including both conven- tional feature-based and deep learning models (Scholkopf and Smola, 2001; Liaw et al., 2002; El- man, 1990; LeCun et al., 1998; Chung et al., 2014) to automatically learn the hidden patterns to solve the task. Machine learning research in healthcare has shown …

Content truncated to 500 words