Exploring Artificial Intelligence in Emergency Medicine for Predicting Disposition, Death, and Diagnosis
Tolestam Heyman, Ellen LU
(2024)
In Lund University, Faculty of Medicine Doctoral Dissertation Series
- Abstract
- In the emergency department (ED), physicians and nurses make rapid, critical decisions. With the advancement of artificial intelligence (AI), we aimed to explore whether AI decision support could offer
clinical benefits in the ED.
Paper I: This study found no correlation between ED hospital admission likelihood and in-hospital bed occupancy levels, suggesting that previous findings may vary across hospitals, possibly due to differences in average hospital bed occupancy. Traditional statistical methods, not AI, were used in this paper.
Paper II: All-cause mortality is often used to predict end-of-life (EOL) and palliative care needs. We aimed to improve this AI prediction by means of an adjudicating committee,... (More) - In the emergency department (ED), physicians and nurses make rapid, critical decisions. With the advancement of artificial intelligence (AI), we aimed to explore whether AI decision support could offer
clinical benefits in the ED.
Paper I: This study found no correlation between ED hospital admission likelihood and in-hospital bed occupancy levels, suggesting that previous findings may vary across hospitals, possibly due to differences in average hospital bed occupancy. Traditional statistical methods, not AI, were used in this paper.
Paper II: All-cause mortality is often used to predict end-of-life (EOL) and palliative care needs. We aimed to improve this AI prediction by means of an adjudicating committee, excluding patients dying unexpectedly, such as in accidents, since these patients most likely lack needs for EOL care. When only unsurprising deaths were included, AI mortality prediction improved significantly.
Paper III: We developed an AI model to identify patients with acute heart failure (AHF), exacerbated chronic obstructive pulmonary disease (eCOPD), and pneumonia among dyspnoeic adults at the beginning of their ED visit. After manually reviewing a subsample of diagnoses, we analysed unselected data from a complete regional healthcare system. The model performed well, and each patient received their own unique set of variables, displayed interpretably.
Paper IV: Building on Paper III, we added electrocardiograms (ECGs) and socioeconomic data. Each patient’s unique set of variables was added to create an AI generated list of thousands of diagnostic variables. The list’s top 20 variables were included in a simpler model, which demonstrated high diagnostic ability, both with and without added medical expertise. (Less) - Abstract (Swedish)
- In the emergency department (ED), physicians and nurses make rapid, critical decisions. With the advancement of artificial intelligence (AI), we aimed to explore whether AI decision support could offer
clinical benefits in the ED.
Paper I: This study found no correlation between ED hospital admission likelihood and in-hospital bed occupancy levels, suggesting that previous findings may vary across hospitals, possibly due to differences in average hospital bed occupancy. Traditional statistical methods, not AI, were used in this paper.
Paper II: All-cause mortality is often used to predict end-of-life (EOL) and palliative care needs. We aimed to improve this AI prediction by means of an adjudicating committee,... (More) - In the emergency department (ED), physicians and nurses make rapid, critical decisions. With the advancement of artificial intelligence (AI), we aimed to explore whether AI decision support could offer
clinical benefits in the ED.
Paper I: This study found no correlation between ED hospital admission likelihood and in-hospital bed occupancy levels, suggesting that previous findings may vary across hospitals, possibly due to differences in average hospital bed occupancy. Traditional statistical methods, not AI, were used in this paper.
Paper II: All-cause mortality is often used to predict end-of-life (EOL) and palliative care needs. We aimed to improve this AI prediction by means of an adjudicating committee, excluding patients dying unexpectedly, such as in accidents, since these patients most likely lack needs for EOL care. When
only unsurprising deaths were included, AI mortality prediction improved significantly.
Paper III: We developed an AI model to identify patients with acute heart failure (AHF), exacerbated chronic obstructive pulmonary disease (eCOPD), and pneumonia among dyspnoeic adults at the beginning of their ED visit. After manually reviewing a subsample of diagnoses, we analysed
unselected data from a complete regional healthcare system. The model performed well, and each patient received their own unique set of variables, displayed interpretably.
Paper IV: Building on Paper III, we added electrocardiograms (ECGs) and socioeconomic data. Each patient’s unique set of variables was added to create an AI generated list of thousands of diagnostic variables. The list’s top 20 variables were included in a simpler model, which demonstrated high diagnostic ability, both with and without added medical expertise. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/d008abae-4b4c-4822-a38f-b1b1614a4b1b
- author
- Tolestam Heyman, Ellen
LU
- supervisor
- opponent
-
- Associate Professor Polesie, Sam, Sahlgrenska University Hospital
- organization
- alternative title
- Prediktion av död, diagnos och inläggningsbeslut på akutmottagningen med hjälp av artificiell intelligens
- publishing date
- 2024
- type
- Thesis
- publication status
- published
- subject
- keywords
- akutsjukvård, Akutmottagning, diagnostik, artificiell intelligens, maskininlärning, Dyspné, mortalitet, Emergency Medicine, Emergency department, Artifical Intelligence, machine learning (ML), Diagnostics, dyspnoea, mortality, disposition decision
- in
- Lund University, Faculty of Medicine Doctoral Dissertation Series
- issue
- 2024:128
- pages
- 73 pages
- publisher
- Lund University, Faculty of Medicine
- defense location
- Belfragesalen, BMC D15, Klinikgatan 32 i Lund. Join by Zoom: https://lu-se.zoom.us/j/61949700061?pwd=HwX2yaDTBfyMKzz0b5PPUWTnXeT0w6.1
- defense date
- 2024-11-21 13:00:00
- ISSN
- 1652-8220
- ISBN
- 978-91-8021-626-5
- project
- AIR Lund - Artificially Intelligent use of Registers
- language
- English
- LU publication?
- yes
- id
- d008abae-4b4c-4822-a38f-b1b1614a4b1b
- date added to LUP
- 2024-10-24 17:35:45
- date last changed
- 2024-11-05 14:41:57
@phdthesis{d008abae-4b4c-4822-a38f-b1b1614a4b1b,
abstract = {{In the emergency department (ED), physicians and nurses make rapid, critical decisions. With the advancement of artificial intelligence (AI), we aimed to explore whether AI decision support could offer <br/>clinical benefits in the ED. <br/><br/>Paper I: This study found no correlation between ED hospital admission likelihood and in-hospital bed occupancy levels, suggesting that previous findings may vary across hospitals, possibly due to differences in average hospital bed occupancy. Traditional statistical methods, not AI, were used in this paper. <br/><br/>Paper II: All-cause mortality is often used to predict end-of-life (EOL) and palliative care needs. We aimed to improve this AI prediction by means of an adjudicating committee, excluding patients dying unexpectedly, such as in accidents, since these patients most likely lack needs for EOL care. When only unsurprising deaths were included, AI mortality prediction improved significantly. <br/><br/>Paper III: We developed an AI model to identify patients with acute heart failure (AHF), exacerbated chronic obstructive pulmonary disease (eCOPD), and pneumonia among dyspnoeic adults at the beginning of their ED visit. After manually reviewing a subsample of diagnoses, we analysed unselected data from a complete regional healthcare system. The model performed well, and each patient received their own unique set of variables, displayed interpretably. <br/><br/>Paper IV: Building on Paper III, we added electrocardiograms (ECGs) and socioeconomic data. Each patient’s unique set of variables was added to create an AI generated list of thousands of diagnostic variables. The list’s top 20 variables were included in a simpler model, which demonstrated high diagnostic ability, both with and without added medical expertise.}},
author = {{Tolestam Heyman, Ellen}},
isbn = {{978-91-8021-626-5}},
issn = {{1652-8220}},
keywords = {{akutsjukvård; Akutmottagning; diagnostik; artificiell intelligens; maskininlärning; Dyspné; mortalitet; Emergency Medicine; Emergency department; Artifical Intelligence; machine learning (ML); Diagnostics; dyspnoea; mortality; disposition decision}},
language = {{eng}},
number = {{2024:128}},
publisher = {{Lund University, Faculty of Medicine}},
school = {{Lund University}},
series = {{Lund University, Faculty of Medicine Doctoral Dissertation Series}},
title = {{Exploring Artificial Intelligence in Emergency Medicine for Predicting Disposition, Death, and Diagnosis}},
url = {{https://lup.lub.lu.se/search/files/198535505/ETH_LUCRIS.pdf}},
year = {{2024}},
}