None listed
Conditions
Brief summary
Background Patients who arrive at the emergency department are triaged by a trained triage nurse, but then may wait hours before being seen by an emergency doctor who decides if they need to be admitted to hospital. Early identification of patients requiring admission from the emergency department may help improve hospital efficiency. Previous research has suggested that machine learning may be able to be applied to triage data to predict if a patient will be admitted to hospital, however no research has been conducted in Western Australia. Objective Use machine learning to predict disposition for patients presenting to the emergency department based on data available at the time of triage. Project plan We will develop our dataset using retrospective triage data from Western Australian Emergency Departments. We will then use a portion of this dataset to train a machine learning model to predict emergency department disposition (such as admitted to ward, intensive care, or discharged). We will test the performance of our machine learning model on the remainder of the dataset and compare the prediction of the best performing machine learning model to the predictions of emergency doctors.
Interventions
We will use historical Western Australia Health Emergency Department Information System (EDIS) data from 1st Jan 2010 to 1st September 2021 to train a machine learning model. We will adapt existing state of the art natural language processing models such as BERT (Bidirectional Encoder Representations from Transformers) for this project.[1] We will also test a trained model developed by Tahayori et al. on our data.[2] We will also apply other pre-exisiting machine learning algorithms such as XGB
We will use historical Western Australia Health Emergency Department Information System (EDIS) data from 1st Jan 2010 to 1st September 2021 to train a machine learning model. We will adapt existing state of the art natural language processing models such as BERT (Bidirectional Encoder Representations from Transformers) for this project.[1] We will also test a trained model developed by Tahayori et al. on our data.[2] We will also apply other pre-exisiting machine learning algorithms such as XGBoost. Historical EDIS data will be split into a training and test group and validated in line with current best practices. Input variables will include all information collected at the time of triage. This includes patient age, time of presentation, mode of arrival, type of residence, Australasian Triage Scale (ATS) category, injury surveillance data, and free text triage notes. Outcome data will include disposition from the ED (such as admitted to ward, intensive care, or discharged). References 1. Devlin J, Chang M, Lee K, Toutanova K. BERT: Pre-training of deep bidirectional transformers for language understanding. Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics; 2019 Conference of the North American Chapter of the Association for Computational Linguistics; June 2-7, 2019; Minneapolis, MN. 2019. Jun, pp. 4171–4186. 2. Tahayori B, Chini-Foroush N, Akhlaghi H. Advanced natural language processing technique to predict patient disposition based on emergency triage notes [published online ahead of print, 2020 Oct 11]. Emerg Med Australas. 2020;10.1111/1742-6723.13656. doi:10.1111/1742-6723.13656
Sponsors
Dr Jonathon Stewart
Eligibility
Sex/Gender
All
Healthy volunteers
No
Inclusion criteria
All patients who presented to an emergency department and were triaged will be eligible for inclusion.
Exclusion criteria
Participants will be excluded if triage information is incomplete, or if they left the emergency department without being seen (“Did Not Wait”).
Outcome results
None listed