Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Risk assessment in suspected acute coronary syndrome with decision support tools and artificial intelligence

Olsson de Capretz, Pontus LU (2025) In Lund University, Faculty of Medicine Doctoral Dissertation Series
Abstract (Swedish)
Acute coronary syndrome (ACS), including acute myocardial infarction (AMI), are a leading cause of
morbidity and mortality worldwide. Accurate risk evaluation in patients presenting with chest pain is
essential for optimizing emergency department (ED) resource allocation and patient outcomes. Current
clinical decision support tools rely on structured risk scores and biomarker measurements, such as highsensitivity
cardiac troponin T (hs-cTnT), but they may not fully capture the complexity of ACS
presentation.
This thesis explores the potential of machine learning (ML) models as decision support tools to improve
the early identification of AMI and reduce unnecessary diagnostic procedures.
The research consists... (More)
Acute coronary syndrome (ACS), including acute myocardial infarction (AMI), are a leading cause of
morbidity and mortality worldwide. Accurate risk evaluation in patients presenting with chest pain is
essential for optimizing emergency department (ED) resource allocation and patient outcomes. Current
clinical decision support tools rely on structured risk scores and biomarker measurements, such as highsensitivity
cardiac troponin T (hs-cTnT), but they may not fully capture the complexity of ACS
presentation.
This thesis explores the potential of machine learning (ML) models as decision support tools to improve
the early identification of AMI and reduce unnecessary diagnostic procedures.
The research consists of four studies. Study I evaluates whether combining glucose measurements with
hs-cTnT improves the 0/1h hs-cTnT protocol in ED patients. Study II investigates theuse of prior
electrocardiograms (ECGs) as inputs to ML-models. Study III compares ML models combining ECG and
laboratory data against the 0-hour hs-cTnT rule-out protocol. Study IV examines the performance of a
sequential ML approach that includes stepwise increasing patient information.
Results indicate that ML models can improve early risk stratification, with convolutional neural networks
(CNNs) outperforming traditional logistic regression and rule-based protocols in predicting AMI or death
within 30 days. However, prior ECG data provided limited additional value to ML models, and the
sequential use of models resulted in a decline in sensitivity. Feature importance analysis showed that
troponin and ECG features remain the dominant predictors used by the ML models.
These findings suggest that ML-based decision support could improve ED efficiency by safely reducing
unnecessary testing and admissions, but further external validation and implementation research are
required before clinical deployment. (Less)
Please use this url to cite or link to this publication:
author
supervisor
opponent
  • Professor Sundström, Johan, Department of Medical Sciences, Clinical Epidemiology, Uppsala
organization
publishing date
type
Thesis
publication status
published
subject
keywords
Chest pain, Emergency Department, Machine learning, Deep learning, ECG, Neural Network, Troponin
in
Lund University, Faculty of Medicine Doctoral Dissertation Series
issue
2025-43
pages
84 pages
publisher
Lund University, Faculty of Medicine
defense location
Föreläsningssal 5, Centralblocket, Entrégatan 7, Skånes Universitetssjukhus i Lund. Join via Zoom: https://lu-se.zoom.us/j/66389555174
defense date
2025-05-02 13:00:00
ISSN
1652-8220
ISBN
978-91-8021-696-8
language
English
LU publication?
yes
id
88e739ff-e3b2-42d3-804b-87c715e77b74
date added to LUP
2025-04-07 09:29:52
date last changed
2025-05-05 12:06:17
@phdthesis{88e739ff-e3b2-42d3-804b-87c715e77b74,
  abstract     = {{Acute coronary syndrome (ACS), including acute myocardial infarction (AMI), are a leading cause of<br/>morbidity and mortality worldwide. Accurate risk evaluation in patients presenting with chest pain is<br/>essential for optimizing emergency department (ED) resource allocation and patient outcomes. Current<br/>clinical decision support tools rely on structured risk scores and biomarker measurements, such as highsensitivity<br/>cardiac troponin T (hs-cTnT), but they may not fully capture the complexity of ACS<br/>presentation.<br/>This thesis explores the potential of machine learning (ML) models as decision support tools to improve<br/>the early identification of AMI and reduce unnecessary diagnostic procedures.<br/>The research consists of four studies. Study I evaluates whether combining glucose measurements with<br/>hs-cTnT improves the 0/1h hs-cTnT protocol in ED patients. Study II investigates theuse of prior<br/>electrocardiograms (ECGs) as inputs to ML-models. Study III compares ML models combining ECG and<br/>laboratory data against the 0-hour hs-cTnT rule-out protocol. Study IV examines the performance of a<br/>sequential ML approach that includes stepwise increasing patient information.<br/>Results indicate that ML models can improve early risk stratification, with convolutional neural networks<br/>(CNNs) outperforming traditional logistic regression and rule-based protocols in predicting AMI or death<br/>within 30 days. However, prior ECG data provided limited additional value to ML models, and the<br/>sequential use of models resulted in a decline in sensitivity. Feature importance analysis showed that<br/>troponin and ECG features remain the dominant predictors used by the ML models.<br/>These findings suggest that ML-based decision support could improve ED efficiency by safely reducing<br/>unnecessary testing and admissions, but further external validation and implementation research are<br/>required before clinical deployment.}},
  author       = {{Olsson de Capretz, Pontus}},
  isbn         = {{978-91-8021-696-8}},
  issn         = {{1652-8220}},
  keywords     = {{Chest pain; Emergency Department; Machine learning; Deep learning; ECG; Neural Network; Troponin}},
  language     = {{eng}},
  number       = {{2025-43}},
  publisher    = {{Lund University, Faculty of Medicine}},
  school       = {{Lund University}},
  series       = {{Lund University, Faculty of Medicine Doctoral Dissertation Series}},
  title        = {{Risk assessment in suspected acute coronary syndrome with decision support tools and artificial intelligence}},
  url          = {{https://lup.lub.lu.se/search/files/216485839/Pontus_Olsson_de_Capretz_Lucris.pdf}},
  year         = {{2025}},
}