Reducing Waiting Time for Time-Critical Patients in the Emergency Department Through Machine Learning-Based Triage
(2026) BMEM01 20261Division for Biomedical Engineering
- Abstract
- Emergency department (ED) wait times can be long, and not all patients can safely wait for treatment. Therefore, an effective emergency triage system is required to prioritize patients according to the urgency of their condition upon arrival at the ED. Assessing a patient’s level of urgency without access to complete physiological measurements is a complex task that is currently performed by nurses.
This study investigated the possibility and suitability of implementing machine learning models for emergency triage, as well as identifying the important features for model performance. Previous work by André Johansson developed a Triage Effectiveness metric, enabling quantitative comparison between machine learning-based triage and... (More) - Emergency department (ED) wait times can be long, and not all patients can safely wait for treatment. Therefore, an effective emergency triage system is required to prioritize patients according to the urgency of their condition upon arrival at the ED. Assessing a patient’s level of urgency without access to complete physiological measurements is a complex task that is currently performed by nurses.
This study investigated the possibility and suitability of implementing machine learning models for emergency triage, as well as identifying the important features for model performance. Previous work by André Johansson developed a Triage Effectiveness metric, enabling quantitative comparison between machine learning-based triage and nurse-assigned priority levels. The models in this study were trained using data from the Skåne Emergency Medicine (SEM) cohort and the Regional Healthcare Information Platform in Halland (RHIPH). While the datasets contained partially different features, the primary distinction was the available physiological measurements: the SEM cohort included blood gas measurements, whereas the RHIPH dataset contained vital parameters.
The machine learning models were developed iteratively through hyperparameter tuning and SHAP-based feature elimination. Each model was trained using a binary label representing whether a patient was time-critical. The models were designed to estimate the probability that a patient was time-critical, after which the predicted probabilities were transformed into triage priority groups from 1 to 5. Triage Effectiveness was then calculated to quantify the reduction in waiting time for time-critical patients compared with a first-come, first-served queueing system.
The results indicate that machine learning models have the potential to improve Triage Effectiveness compared to traditional nurse-based triage. Furthermore, the findings suggest that machine learning models may be suitable as emergency triage decision-support tools, although they are not currently suitable for fully replacing nurse-led triage assessment. (Less) - Popular Abstract
- Reducing Waiting Times in Emergency Care with Artificial Intelligence - Can artificial intelligence help emergency departments identify the sickest patients faster?
Long waiting times are a problem in emergency departments, where staff must decide which patients need immediate care and which can safely wait. In this master’s thesis, machine learning models were developed and evaluated to investigate whether artificial intelligence (AI) could support nurses in emergency triage and help reduce waiting times for patients with life-threatening conditions.
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9230105
- author
- Laurell, Niklas LU and Lundström, Em LU
- supervisor
- organization
- alternative title
- Reducering av Väntetid för Tidskritiska Patienter på Akutmottagningen genom Maskininlärningsbaserad Triagering
- course
- BMEM01 20261
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Emergency Triage, Machine Learning, Decision Support, Emergency Department, Triage Effectiveness, Time-Critical Patients, Predictive Modeling, Queue-Theory-Based Evaluation
- language
- English
- additional info
- 2026-07
- id
- 9230105
- date added to LUP
- 2026-06-02 13:16:06
- date last changed
- 2026-06-29 08:11:41
@misc{9230105,
abstract = {{Emergency department (ED) wait times can be long, and not all patients can safely wait for treatment. Therefore, an effective emergency triage system is required to prioritize patients according to the urgency of their condition upon arrival at the ED. Assessing a patient’s level of urgency without access to complete physiological measurements is a complex task that is currently performed by nurses.
This study investigated the possibility and suitability of implementing machine learning models for emergency triage, as well as identifying the important features for model performance. Previous work by André Johansson developed a Triage Effectiveness metric, enabling quantitative comparison between machine learning-based triage and nurse-assigned priority levels. The models in this study were trained using data from the Skåne Emergency Medicine (SEM) cohort and the Regional Healthcare Information Platform in Halland (RHIPH). While the datasets contained partially different features, the primary distinction was the available physiological measurements: the SEM cohort included blood gas measurements, whereas the RHIPH dataset contained vital parameters.
The machine learning models were developed iteratively through hyperparameter tuning and SHAP-based feature elimination. Each model was trained using a binary label representing whether a patient was time-critical. The models were designed to estimate the probability that a patient was time-critical, after which the predicted probabilities were transformed into triage priority groups from 1 to 5. Triage Effectiveness was then calculated to quantify the reduction in waiting time for time-critical patients compared with a first-come, first-served queueing system.
The results indicate that machine learning models have the potential to improve Triage Effectiveness compared to traditional nurse-based triage. Furthermore, the findings suggest that machine learning models may be suitable as emergency triage decision-support tools, although they are not currently suitable for fully replacing nurse-led triage assessment.}},
author = {{Laurell, Niklas and Lundström, Em}},
language = {{eng}},
note = {{Student Paper}},
title = {{Reducing Waiting Time for Time-Critical Patients in the Emergency Department Through Machine Learning-Based Triage}},
year = {{2026}},
}