A Systematic Review of Machine Learning Prediction and Cardiovascular Risk Assessment Público

Trimble, Shawn (Spring 2022)

Permanent URL: https://etd.library.emory.edu/concern/etds/6395w8447?locale=pt-BR
Published

Abstract

Artificial intelligence (AI) techniques such as machine learning (ML) and deep learning (DL) promise considerable improvements in cardiovascular disease (CVD) prediction. This systematic review and meta-analysis aim to assess and compare the predictive ability of ML algorithms to conventional risk assessment methods in CVDs. Cochrane, Embase, Scopus and Web of Science databases were searched for studies published between January 1, 2012, and January 1, 2022. Studies including the predictive performance of ML models and conventional risk assessment methods were included. Studies without sufficient evaluation data and model validation were excluded. Diagnostic accuracy data was extracted and used to create contingency tables to derive performance metrics of interest: sensitivity, specificity, threshold limits, and areas under the curve (AUC). Studies were included in a meta-analysis, using a bivariate random effects model. The search identified 1688 studies of which 25 studies were included. For the prediction of CVD, ML models had a pooled AUC of 0.88 and conventional risk assessment had a pooled AUC of 0.74. For the prediction of coronary artery disease, ML models had a pooled AUC of 0.83. For the prediction of heart failure, ML models had a pooled AUC of 0.90. For the prediction of stroke, ML models had a pooled AUC of 0.83. Of the 25 studies, 9 studies provided conventional risk assessment comparators. ML models had a pooled AUC of 0.82 compared to conventional risk assessment who had a pooled AUC of 0.78. Insufficient samples sizes for ML models prevented within ML comparisons and analysis of cardiac arrythmias. The predictive ability of ML models is comparable to conventional risk assessment methods. However, heterogeneity among ML models and insufficient data reporting methods within the literature calls into question the clinical applicability of these findings. This review may assist clinicians in assessing the current state of AI and provide insights to how ML models can better translate into the clinical setting.  

Table of Contents

Chapter 1: Introduction 1

Section 1.1: Background 1

Section 1.2: Statement of the Problem 3

Section 1.3: Statement of the Purpose 3

Section 1.4: Research Aims 3

Section 1.5: Significance of the Study 3

Chapter 2: Literature Review 4

Section 2.1: Cardiovascular Disease 4

Section 2.1.1: Current Burden 4

Section 2.1.2: Epidemiological, Economic, and Social Impact 5

Section 2.1.3: Risk Assessment 8

Section 2.2: Artificial Intelligence, Machine Learning, Neural Networks, and Deep Learning 9

Section 2.2.1: Artificial Intelligence 9

Section 2.2.2: Machine Learning 10

Section 2.2.3: Deep Learning 12

Section 2.3: Artificial Intelligence Clinical Applications 13

Section 2.3.1: Precision Medicine 13

Section 2.3.2: Clinical Prediction 15

Section 2.3.3: Diagnostic Imaging 16

Section 2.4: AI Concerns and Limitations 17

Section 2.4.1: Data Consent 17

Section 2.4.2: Data Privacy 18

Section 2.4.3: Data Transparency 20

Chapter 3: Methods 21

Section 3.1: Search Strategy 21

Section 3.2: Study Selection 21

Section 3.3: Data Extraction 22

Section 3.4: Statistical Analysis 23

Chapter 4: Results 24

Section 4.1: Study Search 24

Section 4.2: Study Characteristics 25

Section 4.3: Study Results 25

Section 4.3.1: Machine Learning Models and Prediction of Cardiovascular Disease 25

Section 4.3.2: Machine Learning Models and Prediction of Cardiac Arrythmias 27

Section 4.3.3: Machine Learning Models and Prediction of Coronary Artery Disease 27

Section 4.3.4: Machine Learning Algorithms and Prediction of Heart Failure 28

Section 4.3.5: Machine Learning Algorithms and Prediction of Stroke 29

Section 4.3.6: Matched Prediction of Cardiovascular Disease 30

Chapter 5: Discussion 32

Section 5.1: Discussion of Key Findings 32

Section 5.2: Strengths and Weaknesses 34

Section 5.3: Recommendations 35

Section 5.3.1: Quality Assessment Tools 35

Section 5.3.2: Meta-Analysis of Test Performance Gold Standards 36

Section 5.3.3: Clinical Level Contextualization 37

Section 5.4: Conclusion 39

References 41

Appendix 58

Section A-1: PRISMA Checklist 58

Section A-2: Search Strategy 59

Section A-3: Study Characteristics 62

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Palavra-chave
Committee Chair / Thesis Advisor
Última modificação

Primary PDF

Supplemental Files