Predicting the onset of acute coronary syndrome events and in-hospital mortality using machine learning approaches / Song Cheen

Song , Cheen (2023) Predicting the onset of acute coronary syndrome events and in-hospital mortality using machine learning approaches / Song Cheen. PhD thesis, Universiti Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (249Kb)
	PDF (Thesis PhD) Download (5Mb)

Abstract

Acute coronary syndrome (ACS) represents a significant health concern, and its risk increases with exposure to environmental factors, particularly air pollution. Understanding this association is crucial given the increasing prevalence of air pollution in many regions, particularly in Malaysia, which is affected by air pollution. This study used a comprehensive methodology to investigate the relationship between air pollution and ACS patient outcomes utilizing machine learning (ML) algorithms, including: 1) Linear Regression, 2) Logistic Regression, 3) Support Vector Machine (SVM), 4) Random Forest (RF), 5) XGBoost, 6) Naïve Bayes (NB), and 7) Stacked Ensemble ML utilizing data from the National Cardiovascular Disease Database (NCVD) Malaysia registry and air quality data from the Department of Environment (DOE) Malaysia. The ML models for regression and classification were developed and optimized; the regression models aimed to predict ACS patients’ hospitalization and mortality rates, while the classification models were designed to predict the mortality risk of ACS patients under the influence of air pollution. The regression models reported an RMSE of 1.701 (RF) for predicting hospitalization rate and 0.440 (XGBoost) for predicting cardiac mortality rate on daily basis. The classification models demonstrated an AUC of 0.843 (95% CI: 0.813 – 0.873) (RF) with the in-hospital dataset and 0.840 (95% CI: 0.828 – 0.862) (XGBoost) using the emergency dataset, outperforming the conventional TIMI risk score, and the features importance is visualized using SHAP summary plots, whereby Nitrogen Oxides (NOx) and Ozone (O3) were identified as significant features impacting the ACS patient’s outcome for hospitalization, mortality rate and mortality risk. The best-performing ML models were then integrated into the 'My Heart ACS Air' web system (https://myheartacsair.uitm.edu.my/home.php), ensuring predictions are visualized and made accessible for healthcare professionals. This web system was developed using a prototype-driven approach, emphasizing user feedback, and evaluated using the System Usability Scale (SUS). The models not only provide accurate predictions but also outperform established risk scores in the presence of air pollution. The study's findings hold relevance for Malaysia, illustrating the importance of adopting such models in regions with significant air pollution. By visualizing these predictions via a web system, healthcare professionals can gain actionable insights, potentially leading to improved patient outcomes.

Actions (For repository staff only : Login required)