This paper is in the following e-collection/theme issue:

Reviews in Medical Informatics (93) Cardiac Arrhythmia, Atrial Fibrillation (84) Cardiac Risk and Cardiac Risk Calculators (45) Myocardial Infarction (Heart Attack) (44) Artificial Intelligence (1371) Machine Learning (1465) Decision Support for Health Professionals (1185) Computer-Aided Diagnosis (128)

Published on in Vol 9, No 12 (2021): December

Preprints (earlier versions) of this paper are available at https://preprints.jmir.org/preprint/30798, first published .
Artificial Intelligence in Predicting Cardiac Arrest: Scoping Review

Artificial Intelligence in Predicting Cardiac Arrest: Scoping Review

Artificial Intelligence in Predicting Cardiac Arrest: Scoping Review

Authors of this article:

Asma Alamgir1 Author Orcid Image ;   Osama Mousa1 Author Orcid Image ;   Zubair Shah1, 2 Author Orcid Image
Article Authors Cited by (14) Tweetations (2) Metrics

    Review

    1College of Science and Engineering, Hamad Bin Khalifa University, Qatar Foundation, Doha, Qatar

    2Centre for Health Informatics, Australian Institute of Health Innovation, Macquarie University, Sydney, Australia

    *these authors contributed equally

    Corresponding Author:

    Zubair Shah, PhD

    College of Science and Engineering

    Hamad Bin Khalifa University

    Qatar Foundation

    Education City, PO BOX 34110

    Street 2731, Al Luqta St, Ar-Rayyan

    Doha

    Qatar

    Phone: 974 5074 4851

    Email: zshah@hbku.edu.qa


    Background: Cardiac arrest is a life-threatening cessation of activity in the heart. Early prediction of cardiac arrest is important, as it allows for the necessary measures to be taken to prevent or intervene during the onset. Artificial intelligence (AI) technologies and big data have been increasingly used to enhance the ability to predict and prepare for the patients at risk.

    Objective: This study aims to explore the use of AI technology in predicting cardiac arrest as reported in the literature.

    Methods: A scoping review was conducted in line with the guidelines of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extension for scoping reviews. Scopus, ScienceDirect, Embase, the Institute of Electrical and Electronics Engineers, and Google Scholar were searched to identify relevant studies. Backward reference list checks of the included studies were also conducted. Study selection and data extraction were independently conducted by 2 reviewers. Data extracted from the included studies were synthesized narratively.

    Results: Out of 697 citations retrieved, 41 studies were included in the review, and 6 were added after backward citation checking. The included studies reported the use of AI in the prediction of cardiac arrest. Of the 47 studies, we were able to classify the approaches taken by the studies into 3 different categories: 26 (55%) studies predicted cardiac arrest by analyzing specific parameters or variables of the patients, whereas 16 (34%) studies developed an AI-based warning system. The remaining 11% (5/47) of studies focused on distinguishing patients at high risk of cardiac arrest from patients who were not at risk. Two studies focused on the pediatric population, and the rest focused on adults (45/47, 96%). Most of the studies used data sets with a size of <10,000 samples (32/47, 68%). Machine learning models were the most prominent branch of AI used in the prediction of cardiac arrest in the studies (38/47, 81%), and the most used algorithm was the neural network (23/47, 49%). K-fold cross-validation was the most used algorithm evaluation tool reported in the studies (24/47, 51%).

    Conclusions: AI is extensively used to predict cardiac arrest in different patient settings. Technology is expected to play an integral role in improving cardiac medicine. There is a need for more reviews to learn the obstacles to the implementation of AI technologies in clinical settings. Moreover, research focusing on how to best provide clinicians with support to understand, adapt, and implement this technology in their practice is also necessary.

    JMIR Med Inform 2021;9(12):e30798

    doi:10.2196/30798

    Keywords

    artificial intelligence; machine learning; deep learning; cardiac arrest; predict 


    Background

    Cardiac arrest, also known as sudden cardiac death, is the cessation of the ability of the heart to pump blood. This acute cessation requires immediate intervention, as vital organs, such as the brain and the heart itself, are deprived of blood flow. A delay in intervention can lead to lifelong complications and even death. The global rate of mortality after cardiac arrest is significantly high—78% of out-of-hospital cardiac arrest (OHCA) cases die before they reach the hospital [1]. For those who do receive advanced care, the survival rate remains low. The survival rate for OHCA from the time of cardiac arrest to the time of discharge ranges from 2% to 11% worldwide [2]. The number of cardiac arrest deaths that occur within an in-hospital setting is also significant. In the United States alone, over 290,000 in-hospital cardiac arrests occur annually, with survival rates varying from as low as 0% to 36.2%, out of which a small percentage have favorable neurological prognoses [3].

    Artificial intelligence (AI) is reforming health care every day. AI technologies have the perfect platform to thrive and mature with the growing adoption of electronic health records, development in computational power, continuous monitoring systems, and availability of big data [4]. It has become an important clinical decision-making tool that allows for personalized diagnoses, solutions, prognoses, and predictions of future health outcomes, guiding clinicians and other stakeholders in doing what is best for their patients [4]. AI technology is also rapidly progressing in cardiology, like in any other field of medicine [5]. AI-guided diagnosis and therapy selection have allowed for advancement in research, clinical practice, and population health in cardiovascular medicine [6]. Machine learning (ML) models have also been shown to outperform traditional statistical models in detecting sex differences in cardiovascular disease, further enhancing individualized medicine [7]. AI also plays a major role in improving care for cardiac arrest. AI technologies are being used to prevent cardiac arrest through early identification of risk factors [8], early detection [9], improved management (eg, effective cardiopulmonary resuscitation) [10], and prognosis determination for patients post cardiac arrest [11]. A large part of cardiac arrest research is the prediction of cardiac arrest before its occurrence, as it gives clinicians time to prepare and achieve better patient outcomes.

    Thus, what are AI technologies and their counterparts in this context? AI refers to the field of science revolving around building computational systems and algorithms that facilitate the ability of a machine to mimic human behavior to learn and find solutions to tasks autonomously [4,12]. ML is a subset of AI. ML algorithms focus on building smart solutions after learning from patterns and experiences provided by a structured sample of training data [12]. Deep learning (DL) is a class of ML. It consists of a complex, interconnected, multilayered neural network, resembling a human brain. The aim of DL is to learn and understand patterns from a large amount of unstructured data [5]. In short, the more information it is fed, the more accurate the outcome.

    The ability of AI technologies to process and evaluate patient data to generate predictions is important to support clinicians in making critical decisions, provide effective management, and, ultimately, improve patient outcomes in cardiac arrest cases [13]. Therefore, we believe it is crucial to explore the use of AI technology in predicting cardiac arrest and report our findings to help clinicians and researchers.

    Research Problem and Aim

    Numerous studies have proposed the use of AI in cardiac care, especially the use of AI in the prediction of cardiac arrest. However, there is a lack of consolidating existing evidence that describes the features of AI technologies, data sets, and data sources currently being used. It is essential to summarize recent findings that allow health care providers and researchers to implement appropriate guidelines, as well as to identify research gaps in the current literature. We encountered one review that examined the use of AI in the prediction of cardiac arrest [14]. However, the review was conducted in 2018 and did not include a large influx of studies in the past 2 years. Therefore, it is necessary to conduct a scoping review that focuses on various types of AI technologies currently being used in different settings to predict cardiac arrest.

    This scoping review aims to explore the use and features of AI technologies applied to the prediction of cardiac arrest as reported in the literature. The results of our review will be a useful reference for health care professionals, researchers, and others involved in patient care to understand the application of AI and leverage it for the benefit of the community.


    The scoping review was conducted by AA and OM to address this objective. The guidelines of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) extension for scoping reviews [15] were followed to help conduct a transparent review.

    Search Strategy

    Five bibliographic databases were searched for this study: Scopus, ScienceDirect, Embase, the Institute of Electrical and Electronics Engineers, and Google Scholar. The databases were searched using search terms related to the target technology, population, and outcomes of interest. Search terms for our population included Cardiac Arrest OR Heart Arrest OR Sudden Cardiac Death OR asystole OR cardiopulmonary arrest and, for our intervention, Artificial Intelligence OR Deep Learning OR Machine Learning OR Natural Language Processing OR Neural network OR Supervised learning OR Unsupervised learning OR Data mining. Outcome- or purpose-related search terms included Detect* OR Predict* OR Anticipat* OR Diagnos*. The search query used for each database is presented in Multimedia Appendix 1.

    For ScienceDirect and Google Scholar, only the first 100 and 50 results, respectively, were considered. This is because the reviewers found that the results became less relevant to the topic of interest and applicability after the mentioned number of citations. In addition to searching the databases, a backward reference list screening of the included studies was also carried out to identify additional relevant studies. The search was conducted between March 15 and 20, 2021.

    Eligibility Criteria

    AI technologies implemented to predict cardiac arrest were included, with no restrictions on age, gender, geography, and type of AI technology used. Studies that focused primarily on predicting cardiac arrest were included. In contrast, studies dedicated to other aspects or contributing factors of cardiac arrest, such as arrhythmia and other cardiac diseases, were excluded. The review included peer-reviewed articles, preprints, articles in press, conference proceedings, theses, and dissertations written in English. Reviews, conference abstracts, study protocols, and proposals were excluded. No restrictions were imposed on the study design, study setting, country of publication, and publication year during the search query. However, only studies published between 2013 and 2021 were included in the review. The period between 2013 and 2016 constitutes a time when AI technologies saw a rapid increase of 175% in application [16]; therefore, the reviewers considered it to be a reasonable time period to include. The study eligibility criteria are summarized in Textbox 1.

    Inclusion and exclusion criteria.

    Inclusion criteria

    • Studies that focused on the use of artificial intelligence (AI) technologies in cardiac arrest prediction for the benefit of the human population
    • Studies published from 2013 to 2021
    • Peer-reviewed articles, articles in press, theses, dissertations, and conference proceedings
    • Primary studies

    Exclusion criteria

    • Articles that did not address the use of AI in cardiac arrest prediction
    • Reviews, conference abstracts or proposals, letters, news, books, and protocols
    • Published in a language other than English
    Textbox 1. Inclusion and exclusion criteria.

    Study Selection

    The studies retrieved from the databases were first imported to Rayyan (Rayyan System Inc) [17], a collaborative research tool, to undergo 3 phases of the filtering process. This ensured that the articles we included in the review were relevant to our study objective. The 3 phases of the filtering process were as follows: (1) identification phase, where citations were identified after applying the search terms to the databases and duplicates were removed; (2) screening phase, where titles and abstracts were screened to remove articles that did not match our inclusion criteria; and (3) eligibility phase, where the full texts of the articles were read to determine their applicability on the basis of the inclusion criteria. The 2 reviewers conducted all 3 phases independently, facilitated by the Rayyan application. In case of conflict, a discussion was held to reach a consensus.

    Data Extraction and Data Synthesis

    To conduct a reliable and consistent extraction of data from the included studies, a data extraction form was used (Multimedia Appendix 2). The 2 reviewers independently extracted data related to the characteristics of the included studies, AI technology, and data sets. The extracted information was recorded on a shared Microsoft Excel sheet for easy data management. Similar to the study selection, any conflict between the 2 reviewers was resolved through discussions to reach a consensus.

    A narrative synthesis of the extracted data was performed. The findings from the included studies were classified and described in terms of their purpose, AI branch, algorithm, and platform used to implement the algorithm. The data sets used for the development and validation of the technology were considered and described. The data sources, size of the data set. validation type, and proportion of training, validation, and test data sets were included when available. An sheet Excel (Multimedia Appendix 3) was used to record the extracted data to facilitate data synthesis.


    Search Findings

    As shown in Figure 1, 697 studies were retrieved from our search, of which 173 (24.8%) duplicates were removed. A total of 524 underwent title and abstract screening, of which 443 (84.5%) studies were excluded. The reasons for exclusion are shown in Figure 1. In total, 81 unique studies underwent full-text screening to evaluate eligibility, of which 41 (51%) studies met the inclusion criteria and were included in the review. Six additional studies were identified and added by checking the reference lists of those 41 studies. Overall, 47 studies were included in the review.

    ‎
    Figure 1. Flowchart of the study selection process.
    View this figure

    Characteristics of the Included Studies

    Of the 47 studies included, 46 (98%) were published in peer-reviewed journals, whereas 1 (2%) was still in press. Approximately 81% (38/47) of the studies were research articles, whereas the rest were conference proceedings (9/47, 19%). Only 2 studies from 2013 were included, whereas most of the studies were from 2020 (12/47, 26%). The other included studies were conducted in 2014 (4/47, 9%), 2015 (5/47, 11%), 2016 (3/47, 6%), 2017 (3/47, 6%), 2018 (5/47, 11%), 2019 (9/47, 19%), and 2021 (4/47, 9%). The included studies were conducted in 15 countries, and most of the studies were published in India and the United States (9/47, 19%). Table 1 shows the characteristics of the studies included in our review. Multimedia Appendix 3 demonstrates the attributes of each study.

    Table 1. Characteristics of the included studies (N=47).
    CharacteristicStudies, n (%)
    Paper status

    		Published46 (98)

    		In press1 (2)
    Publication type

    		Conference proceeding9 (19)

    		Research article38 (81)
    Country

    		Australia1 (2)

    		China3 (6)

    		Greece1 (2)

    		India9 (19)

    		Iran5 (11)

    		Japan1 (2)

    		Malaysia4 (9)

    		Poland1 (2)

    		Portugal1 (2)

    		Singapore1 (2)

    		South Korea7 (15)

    		Spain1 (2)

    		Taiwan2 (4)

    		United Kingdom1 (2)

    		United States9 (19)
    Year published

    		20132 (4)

    		20144 (9)

    		20155 (11)

    		20163 (6)

    		20173 (6)

    		20185 (11)

    		20199 (19)

    		202012 (26)

    		20214 (9)

    AI Characteristics in the Included Studies

    Use of AI in Predicting Cardiac Arrest

    The approaches taken by the included studies to predict cardiac arrest using AI technologies were divided into 3 categories: analysis of variables and parameters, development of an early warning system or prediction model, and stratification of patients at a high risk of cardiac arrest.

    Analysis of Variables and Parameters

    The studies in this category focused on analyzing one or more patient parameters to determine their impact on the efficiency of improving the prediction of cardiac arrest in combination with AI algorithms. We observed 26 studies that fit into this category [14,18-42]. Of these 26 studies, 11 (42%) used ML models [14,18,19,23,25,27,30,32,34-36] and 3 (12%) used DL algorithms [20,31,38]. We observed that 12 studies incorporated both ML and DL models to analyze and validate different parameters [21,22,24,26,28,29,33,37,39-42].

    Random forest (RF) [14,21,23,28-30,32,35-37,39-41] and support vector machine (SVM) [18,22,24,26,28,34,40-42] were the most used ML models observed in these studies, followed by decision tree (DT) [22,29,30,40-42], logistic regression (LR) [28-30,40], Naive Bayes [19,28,29,41], gradient boosting [27,28], extreme gradient boosting [27,29], LogitBoost [21], AdaBoost [29], TreeBagger [34], and sequential feature selection [24]. The most used DL-based algorithm in the studies was k-nearest neighbors (KNN) [20,22,26,29,33,41,42]. The probabilistic neural network [24,31,42], artificial neural network [29,40], multilayer perceptron [21,33], long short-term memory [39], convolutional neural network [37], and enhanced probabilistic neural network [31] were also algorithms used in DL model studies. Furthermore, 2 studies did not specify the algorithm used [25,38].

    The parameters analyzed and validated in the included studies were diverse. The majority of the studies focused on using various characteristics from patient electrocardiogram readings [14,18,20-22,24,26,27,30-34,36-38,40-42], especially heart rate variability (HRV) [14,21,22,26,30,32,34,36-38,42]. HRV is the variation in time between each heartbeat that can be tracked on an electrocardiogram [43]. This noninvasive assessment tool provides important information about the autonomic nervous system, allowing clinicians to determine current and impending cardiac disease [44]. Its usefulness in determining cardiac-related prognosis is also well-documented in the literature [45,46]. In the included studies, HRV appeared to improve prediction outcomes in the studies that integrated it into the data set. All studies using HRV reported higher performance in terms of accuracy and other outcome indicators. Other unique parameters, such as genetic data [20], smoking habit [29], nursing documentation [25], and dialysis status [23,28], were also used to evaluate their effect on the performance of the AI technology to predict cardiac arrest. Accuracy [18,19,21,22,29,31,33,34,36,37,40,41] and sensitivity [14,22,26-28,30,32-34,41,42] were the most used measures of outcome in this category.

    Development of an Early Warning System Using AI

    In 16 studies [47-62], the focus of AI technologies was to develop an early warning system alerting health care professionals when patients were at risk of going into cardiac arrest in the future. To develop a warning model, most studies used ML model algorithms [49-51,53,54,59,61], whereas 5 only used DL-based algorithms [47,48,56,60,62]. Four studies used both ML- and DL-based algorithms [52,55,57,58], comparing them with each other to observe which yielded the best outcome. The ML algorithms used in these studies included LR [50,52,55,58], SVM [50-52,58], DT [52,53,57,59], RF [55,57,58], Naive Bayes [57,58], gradient boosting [58], Bayesian networks [49], AdaBoost [57], transfer learning [54], and multichannel Hidden Markov Model [61]. The KNN [52,58], artificial neural network [48,58,61], long short-term memory [47,56,59], and recurrent neural network [47,55,56,62] algorithms were used in the studies to constitute a DL-based early warning system. A total of 10 of the studies compared their outcomes to existing or traditional early warning systems [47,48,50,52,53,56-58,60,62]. The studies compared their models to scoring systems such as the Modified Early Warning Score [48,50,52,53,56,58,60,62], Early Warning Score [57], National Early Warning Score [60], and Pediatric Early Warning Score [47]. Only 1 study showed similar outcomes when using an AI model compared with a traditional warning system [53], whereas, in other studies, the AI-based model outperformed the system it was compared with. For example, deep early warning systems detected 50%-78% more cardiac arrests compared with the Modified Early Warning Score [56,62]. Moreover, the prediction period of the algorithms was reported to range from 30 minutes to as early as 24 hours before the onset of cardiac arrest [50,53,57,58,62].

    Three of the most used outcome measures in this category included the area under the receiver operating characteristic curve [47,48,51,53,55-57,60,62], sensitivity [49,52,58,60,62], and accuracy [51,54,58-60].

    Stratification of High-risk Patients

    In 5 studies [63-67], AI technologies were used to distinguish patients who were at high risk of cardiac arrest from patients who were not at risk. Three studies highlighted HRV [63-65] as an important feature to distinguish high-risk patients.

    ML was used in the majority of the studies [63,64,67], and only 1 study used a DL algorithm [66]. One study used both ML and DL models to stratify patients [65]. The ML algorithms used were SVM [63,64], linear discriminant analysis [64], DT [63], LR [67], RF [67], extreme gradient boosting [67], and fuzzy classifier [65]. The DL algorithms included KNN [65,66] and multilayer perceptron [66]. The outcome measures in the studies included accuracy [63-66], sensitivity, specificity [63-65], area under the receiver operating characteristic curve, and the precision-recall curve [66].

    Features of AI Techniques in the Studies

    Most studies used traditional ML models and algorithms to predict cardiac arrest (38/47, 81%) whereas 55% (26/47) used DL techniques. We observed 15 types of AI classifiers used in the studies to predict cardiac arrest (Table 2). A notable observation is that 6 models were commonly used; neural network–based models, which are a DL model, and RF, which is a traditional ML model, were used 20 and 18 times, respectively, making them the top 2 most used models found in the studies, followed by SVM (15/47, 32%), DT (12/47, 26%), LR (11/47, 23%), and KNN (10/47, 21%). Less common models, such as transfer learning, linear discriminant analysis, fuzzy classifier, multichannel Hidden Markov Model, LogitBoost, AdaBoost, Bayesian networks, Naive Bayes, and extreme gradient boosting, were used between 1 and 6 times in the studies. Two studies used wearable devices as the platform for their AI techniques [24,59], whereas the remaining studies used computers. Multimedia Appendix 3 presents the features of the AI techniques.

    Table 2. Features of artificial intelligence (AI)–based techniques used for cardiac arrest prediction (N=47).
    FeatureStudy IDaStudies, n (%)b
    AI modelc

    		Neural network1, 3, 4, 6, 11, 13, 14, 15, 16, 19, 21, 25, 26, 28, 32, 34, 26, 38, 45, 4620 (43)

    		Random forest3, 6, 7, 8, 9, 10, 13, 14, 15, 17, 18, 19, 28, 20, 35, 37, 41, 4518 (38)

    		Support vector machine2, 5, 19, 20, 27, 30, 31, 32, 34, 38, 41, 42, 43, 45, 4615 (32)

    		Decision tree3, 5, 15, 16, 17, 18, 19, 20, 32, 34, 40, 4212 (26)

    		Logistic regression3, 6, 10, 15, 16, 18, 19, 30, 32, 45, 4711 (23)

    		K-nearest neighbors3, 20, 24, 32, 33, 34, 36, 42, 43, 4610 (21)

    		Extreme gradient boosting3, 10, 15, 16, 44, 456 (13)

    		Naive Bayes16, 20, 22, 454 (9)

    		AdaBoost151 (2)

    		Bayesian networks291 (2)

    		LogitBoost281 (2)

    		Multichannel Hidden Markov Model231 (2)

    		Fuzzy classifier331 (2)

    		Linear discriminant analysis271 (2)

    		Transfer learning471 (2)
    Platform

    		Computer1-16, 18-37, 39-4745 (96)

    		Wearable17, 382 (4)

    aThe order of the reviewed studies in this table follows the order shown in Multimedia Appendix 3.

    bTwo studies did not specify the artificial intelligence model used.

    cThe numbers do not add up as some studies used more than one artificial intelligence model or algorithm.

    Features of Data Sets Used for Development and Validation of AI Models

    Clinical setting sources (such as hospital databases and medical centers) were the most commonly used data sources for the development and validation of AI models [14,25,27,28,31,32,34-36,38,39,47-53,55-57,60,62,67]. Public resources (eg, the MIT-BIH Arrhythmia and Normal Sinus Rhythm databases) [18-24,26,29,30,33,37,41,42,54,58,61,63-66] were the other sources of data for AI models.

    Several types of data were retrieved from these sources. We grouped the types of data into 5 categories: clinical data, demographic data (eg, age, gender, and ethnicity), laboratory data (eg, blood samples), radiology data (eg, x-rays), and biological data (eg, genetic information). As shown in Table 3, 58% (34/47) of the studies used clinical data as the data type. Different variables fall under this category; Table 4 breaks down the type of clinical data observed in the studies. Demographic data were the second most used data type in predicting cardiac arrest (15/47, 26%), followed by laboratory data (8/47, 14%) and biological data (1/47, 2%).

    Table 3. Data types.
    Data typeStudies, n (%)
    Clinical data34 (72)
    Demographic data15 (32)
    Laboratory data8 (17)
    Biological data1 (2)
    Table 4. Clinical data breakdowna.
    Clinical data typesStudies, n (%)
    Vital signs23 (49)
    ECGb variables18 (38)
    Medical history10 (21)
    Chief complaint3 (6)
    Medication3 (6)
    Cardiopulmonary exercise testing2 (4)
    Diagnosis2 (4)
    Risk score2 (4)
    Renal status2 (4)
    Cardiopulmonary resuscitation information1 (2)
    Lifestyle1 (2)
    Nursing notes1 (2)

    aSeveral studies collected more than one clinical data type.

    bECG: echocardiogram.

    For data set sizes, 42 (89%) out of 47 studies mentioned the size of the training data set used for the ML model. Of the 47 studies, 23 (49%) used data sets of less than 1000 samples, whereas 14 (30%) used data sets of between 1000 and 9999 samples. Moreover, 11% (5/47) of studies used more than 10,000 data samples. Various validation types for the AI models were reported in 41 studies. These validation methods were divided into 3 main categories: k-fold cross-validation, which was the most common validation technique used (24/47, 51%), followed by train-test split (11/47, 23%) and external validation (6/47, 13%). Table 5 provides a breakdown of the features of data used in the included studies.

    Table 5. Features of the data used (N=47).
    FeatureStudies, n (%)
    Data sources

    		Public database21 (45)

    		Clinical setting24 (51)

    		Other2 (4)
    Data set sizea

    		<100023 (49)

    		1000-999914 (28)

    		≥10,0005 (11)
    Type of validationb

    		K-fold cross-validation24 (51)

    		Train-test split11 (23)

    		External validation6 (13)

    aData set size mentioned in 42 studies.

    bTypes of validation mentioned in only 41 studies.


    Principal Findings

    In this review, we explored the use of AI in predicting cardiac arrest. From a total of 617 retrieved studies, 47 (7.6%) were included in this review. We found that the number of studies increased in the past 2 years (9 in 2019 and 11 in 2020), which is not surprising given that the use of AI technology in health care has been increasing. India and the United States (9/47, 19%) represent the countries that published the most studies related to AI in predicting cardiac arrest, with a total of 18. To explore the use of AI technology in predicting cardiac arrest, we divided our findings into 3 categories, each representing a classification of the reviewed studies from a different perspective. The first category focuses on the way AI technologies are used in predicting cardiac arrest and comprises 3 main subcategories: (1) stratification between patients with cardiac arrest and non-at-risk patients, in which the AI technology was trained using the history of patients who had cardiac arrest and classified patients with a high risk of cardiac arrest; (2) development of an early warning system using AI, in which AI technology was used to alert physicians 1 to 16 hours before cardiac arrest and its accuracy was compared with other existing traditional warning systems; and (3) analysis of different variables and parameters to observe the efficiency of prediction.

    The second category identifies the features of the AI techniques as observed in the literature. Two AI branches were used, ML and DL, where ML was the most used branch in a total of 38 studies, and the most used model in this branch was RF (18/47, 38%). In contrast, DL was used 16 times, and the most used model were neural network–based models (20/47, 43%). Finally, the third category classifies the data and validation method used for the AI, where we expanded on the data sources, data types, and validation processes found in the literature for the AI techniques. A total of 42 out of the 47 studies mentioned the data set size used, the majority of the studies using data sets of less than 1000 samples (23/47, 49%). Most studies used k-fold cross-validation to test the AI models (24/47, 51%).

    The Implications for Practice and Research

    This review highlighted the most common AI models used in predicting cardiac arrest and the different approaches used in predicting it. On the basis of our findings AI models can predict cardiac arrest using a variety of data types. In our review, ML techniques were used much more than DL techniques. One explanation for this is that the data used to train the AI model were mostly structured (eg, vital signs are recorded, and the threshold for the measurements of a normal human being is known and then compared with the vital signs of a patient who had cardiac arrest). Therefore, it is understandable that most researchers used ML techniques, because they were dealing with structured data. In contrast, DL works best with unstructured data, which was less commonly used in the articles reviewed. Another explanation is the size of the data sets used, as most studies used relatively small data sets to train DL models (eg, only 5 studies out of 47 used data sets of more than 10,000 samples). Finally, many studies explained the use of ML techniques such as DTs, LR, and RF, which consist of many DTs given that the main outcome is binary (at risk of cardiac arrest or not at risk of cardiac arrest). This explains the rapid use of these techniques in the reviewed studies.

    Future research should explore ways to attain higher prediction accuracy in terms of the time before cardiac arrest may occur to the patient and the percentage of true positive and true negative (accurately predicting that the patient will experience cardiac arrest). Moreover, more research is required to address and investigate hyperparameter optimization, as it could lead to different performance results of ML models across the studies selected and influence which parameters are important for the prediction of cardiac arrest. Early prediction of cardiac arrest could be achieved through the correlation between the clinical data obtained and the demographic data of the patient. ML seems to be the best technique to be used because the data used is structured (eg, age, vital signs, and electrocardiogram variables). The earlier the prediction time, the higher the likelihood that the physicians can save the patients from sudden cardiac death. Furthermore, the potential to evaluate the effectiveness of less frequently used data types, such as laboratory and biological data, in predicting cardiac arrest should also be explored.

    Only 5 studies reviewed used data sets of more than 10,000 samples, whereas most of the studies used data sets of less than 1000 samples. Future studies need to evaluate AI models using larger data sets to improve their effectiveness. In addition, comparing the prediction accuracy of AI techniques with each other is a good method of evaluation. However, AI techniques need to be compared with other techniques used to predict cardiac arrest.

    Studies that did research in clinical settings limited the population to a specific hospital or country, which produced biased results that do not apply everywhere. Future studies should consider public databases that contain cases from different hospitals and countries.

    Many studies explored the potential of AI in the prediction of arrhythmia and irregular heartbeat, and future studies should investigate the potential of the proposed models in the prediction of cardiac arrest. Finally, future research should explore the potential of physiological and psychological data in the prediction of cardiac arrest.

    Strengths

    The review addressed the use of all types of AI technologies to predict cardiac arrest in all populations with no restrictions on paper status, study settings, and geographic location in a comprehensive manner. Moreover, an in-depth exploration was conducted on the features of AI technology and the data sets that were used to develop and validate these technologies.

    Other reviews have explored the use of ML and DL in detecting arrhythmia [53,68] or the use of AI in cardiology in general [69-71] but have not gone into detail on how this technology can be used to predict cardiac arrest. A previous systematic review explored the use of ML in predicting cardiac arrest [72]; however, to the best of our knowledge, this is the first review to explore the different approaches to predicting cardiac arrest to fill the research gap with a better understanding of the prediction techniques rather than focusing on whether the model was able to predict only cardiac arrest. Moreover, this study did not focus on a specific AI branch (ML, DL, or natural language); rather, it focused on categorizing the AI techniques into branches to provide insight into the most common AI technique in every branch.

    The studies included in the review comprised the latest publications, reducing the selection bias date. In addition to published research articles, conference proceedings were also included to maximize the extent of inclusion. This was also done by conducting a backward reference list check of the included studies. Furthermore, study selection and data extraction involved 2 reviewers independently overseeing the process, which ensured minimal selection bias.

    Limitations

    This review did not include databases such as ACM and JSTOR, which limited our access to gray literature and other potentially relevant studies. This was because of the lack of access to some of the databases and others specialized in physiological or engineering studies rather than medical studies. Moreover, owing to practical constraints, only English-language studies were included in the review, excluding studies in other languages. Furthermore, our search query did not include MeSH (Medical Subject Headings) terms or algorithm-specific search terms, which might have hidden studies that would otherwise have been appropriate for our review.

    Conclusions

    Our scoping review included 47 studies that focused on the use of AI technologies to predict cardiac arrest in all settings. With the big data available from patient monitoring systems and electronic health records, it is possible to delve deeper into making our approach to cardiac arrest reliable and more effective, increasing the rate of survival over time. Moreover, with the increasing adoption of wearable devices with sensors tracking various aspects of health and activity, there are opportunities for research to develop techniques to predict and alert patients at risk of OHCAs. Furthermore, clinicians need to be on board with the rapidly growing technology as, without them, we cannot move forward. Therefore, more research on AI paired with education initiatives within health care professionals needs to be considered.

    Conflicts of Interest

    None declared.

    ‎
    Multimedia Appendix 1

    Search strategy.

    DOCX File , 14 KB
    ‎
    Multimedia Appendix 2

    Data extraction form.

    DOCX File , 14 KB
    ‎
    Multimedia Appendix 3

    Characteristics of the included studies and features of artificial intelligence techniques.

    DOCX File , 31 KB
    1. Yan S, Gan Y, Jiang N, Wang R, Chen Y, Luo Z, et al. The global survival rate among adult out-of-hospital cardiac arrest patients who received cardiopulmonary resuscitation: a systematic review and meta-analysis. Crit Care 2020 Feb 22;24(1):61 [FREE Full text] [CrossRef] [Medline]
    2. Berdowski J, Berg RA, Tijssen JG, Koster RW. Global incidences of out-of-hospital cardiac arrest and survival rates: systematic review of 67 prospective studies. Resuscitation 2010 Nov;81(11):1479-1487. [CrossRef] [Medline]
    3. Chan PS, Krein SL, Tang F, Iwashyna TJ, Harrod M, Kennedy M, American Heart Association's Get With the Guidelines–Resuscitation Investigators. Resuscitation practices associated with survival after in-hospital cardiac arrest: a nationwide survey. JAMA Cardiol 2016 May 01;1(2):189-197 [FREE Full text] [CrossRef] [Medline]
    4. Bohr A, Memarzadeh K, editors. Artificial Intelligence in Healthcare. Amsterdam: Elsevier; 2020.
    5. Lopez-Jimenez F, Attia Z, Arruda-Olson AM, Carter R, Chareonthaitawee P, Jouni H, et al. Artificial intelligence in cardiology: present and future. Mayo Clin Proc 2020 May;95(5):1015-1039. [CrossRef] [Medline]
    6. Johnson KW, Soto JT, Glicksberg BS, Shameer K, Miotto R, Ali M, et al. Artificial intelligence in cardiology. J Am Coll Cardiol 2018 Jun 12;71(23):2668-2679 [FREE Full text] [CrossRef] [Medline]
    7. Sarajlic P, Plunde O, Franco-Cereceda A, Bäck M. Artificial intelligence models reveal sex-specific gene expression in aortic valve calcification. JACC Basic Transl Sci 2021 May;6(5):403-412 [FREE Full text] [CrossRef] [Medline]
    8. JayaSree M, Koteswara Rao L. Survey on - identification of coronary artery disease using deep learning. Mater Today Proc 2020 Oct:526. [CrossRef]
    9. Kumari CU, Murthy AS, Prasanna LW, Reddy MP, Panigrahy AK. An automated detection of heart arrhythmias using machine learning technique: SVM. Mater Today Proc 2021 Aug;45:1393-1398. [CrossRef]
    10. Didon J, Ménétré S, Jekova I, Stoyanov T, Krasteva V. Analyze Whilst Compressing Algorithm for detection of ventricular fibrillation during CPR: a comparative performance evaluation for automated external defibrillators. Resuscitation 2021 Mar;160:94-102. [CrossRef] [Medline]
    11. Cronberg T, Greer DM, Lilja G, Moulaert V, Swindell P, Rossetti AO. Brain injury after cardiac arrest: from prognostication of comatose patients to rehabilitation. Lancet Neurol 2020 Jul;19(7):611-622. [CrossRef]
    12. de Marvao A, Dawes TJ, Howard JP, O'Regan DP. Artificial intelligence and the cardiologist: what you need to know for 2020. Heart 2020 Mar 23;106(5):399-400 [FREE Full text] [CrossRef] [Medline]
    13. Miyazawa AA. Artificial intelligence: the future for cardiology. Heart 2019 Aug 12;105(15):1214. [CrossRef] [Medline]
    14. Elola A, Aramendi E, Rueda E, Irusta U, Wang H, Idris A. Towards the prediction of rearrest during out-of-hospital cardiac arrest. Entropy (Basel) 2020 Jul 09;22(7):758 [FREE Full text] [CrossRef] [Medline]
    15. Tricco AC, Lillie E, Zarin W, O'Brien KK, Colquhoun H, Levac D, et al. PRISMA extension for Scoping Reviews (PRISMA-ScR): checklist and explanation. Ann Intern Med 2018 Oct 02;169(7):467-473 [FREE Full text] [CrossRef] [Medline]
    16. Boom in artificial intelligence patents, points to ‘quantum leap’ in tech: UN report. UN News. 2019.   URL: https://news.un.org/en/story/2019/01/1031702 [accessed 2021-04-08]
    17. Rayyan QCRI.   URL: https://www.rayyan.ai [accessed 2021-11-11]
    18. Murugappan M, Murugesan L, Jerritta S, Adeli H. Sudden Cardiac Arrest (SCA) prediction using ECG morphological features. Arab J Sci Eng 2020 Jul 26;46(2):947-961. [CrossRef]
    19. Karankar N, Shukla P, Agrawal N. Comparative study of various machine learning classifiers on medical data. In: Proceedings of the 7th International Conference on Communication Systems and Network Technologies (CSNT). 2017 Presented at: 7th International Conference on Communication Systems and Network Technologies (CSNT); Nov. 11-13, 2017; Nagpur, India p. 267-271. [CrossRef]
    20. Alfarhan KA, Mashor MY, Zakaria A, Omar MI. Automated electrocardiogram signals based risk marker for early sudden cardiac death prediction. J Med Imaging Heal Informatics 2018 Dec 01;8(9):1769-1775. [CrossRef]
    21. Tapas N, Lone T, Reddy D, Kuppili V. Prediction of cardiac arrest recurrence using ensemble classifiers. Sādhanā 2017 Jun 17;42(7):1135-1141. [CrossRef]
    22. Fujita H, Acharya UR, Sudarshan VK, Ghista DN, Sree SV, Eugene LW, et al. Sudden cardiac death (SCD) prediction based on nonlinear heart rate variability features and SCD index. Appl Soft Comput 2016 Jun;43:510-519. [CrossRef]
    23. Goldstein BA, Chang TI, Mitani AA, Assimes TL, Winkelmayer WC. Near-term prediction of sudden cardiac death in older hemodialysis patients using electronic health records. Clin J Am Soc Nephrol 2013 Oct 31;9(1):82-91. [CrossRef]
    24. Murukesan L, Murugappan M, Iqbal M, Saravanan K. Machine Learning Approach for Sudden Cardiac Arrest Prediction Based on Optimal Heart Rate Variability Features. J Med Imaging Hlth Inform 2014 Aug 01;4(4):521-532. [CrossRef]
    25. Collins SA, Cato K, Albers D, Scott K, Stetson PD, Bakken S, et al. Relationship between nursing documentation and patients' mortality. Am J Crit Care 2013 Jul 01;22(4):306-313 [FREE Full text] [CrossRef] [Medline]
    26. Houshyarifar V, Chehel Amirani M. An approach to predict Sudden Cardiac Death (SCD) using time domain and bispectrum features from HRV signal. Biomed Mater Eng 2016 Aug 12;27(2-3):275-285. [CrossRef]
    27. Wu TT, Lin XQ, Mu Y, Li H, Guo YS. Machine learning for early prediction of in-hospital cardiac arrest in patients with acute coronary syndromes. Clin Cardiol 2021 Mar 14;44(3):349-356 [FREE Full text] [CrossRef] [Medline]
    28. Nakajima K, Nakata T, Doi T, Tada H, Maruyama K. Machine learning-based risk model using I-metaiodobenzylguanidine to differentially predict modes of cardiac death in heart failure. J Nucl Cardiol 2020 May 14;11(10):897-904 [FREE Full text] [CrossRef] [Medline]
    29. L PR, Jinny SV, Mate YV. Early prediction model for coronary heart disease using genetic algorithms, hyper-parameter optimization and machine learning techniques. Health Technol 2020 Nov 13;11(1):63-73. [CrossRef]
    30. Shashikant R, Chetankumar P. Predictive model of cardiac arrest in smokers using machine learning technique based on Heart Rate Variability parameter. Appl Comput Informatics 2020 Jul 28:Ahead of Print. [CrossRef]
    31. Amezquita-Sanchez JP, Valtierra-Rodriguez M, Adeli H, Perez-Ramirez CA. A novel wavelet transform-homogeneity model for sudden cardiac death prediction using ECG signals. J Med Syst 2018 Aug 16;42(10):176. [CrossRef] [Medline]
    32. Liu N, Koh ZX, Goh J, Lin Z, Haaland B, Ting BP, et al. Prediction of adverse cardiac events in emergency department patients with chest pain using machine learning for variable selection. BMC Med Inform Decis Mak 2014 Aug 23;14:75 [FREE Full text] [CrossRef] [Medline]
    33. Ebrahimzadeh E, Foroutan A, Shams M, Baradaran R, Rajabion L, Joulani M, et al. An optimal strategy for prediction of sudden cardiac death through a pioneering feature-selection approach from HRV signal. Comput Methods Programs Biomed 2019 Feb;169(3):19-36. [CrossRef] [Medline]
    34. Mirhoseini SR, Jahedmotlagh MR, Pooyan M. Improve accuracy of early detection Sudden Cardiac Deaths (SCD) using decision forest and SVM. In: Proceedings of the International Conference on Robotics and Artificial Intelligence. 2016 Presented at: International Conference on Robotics and Artificial Intelligence; April 20-22, 2016; Los Angeles   URL: https:/​/www.​researchgate.net/​publication/​296701627_Improve_Accuracy_of_Early_Detection_Sudden_Cardiac_Deaths_SCD_Using_Decision_Forest_and_SVM
    35. Ueno R, Xu L, Uegami W, Matsui H, Okui J, Hayashi H, et al. Value of laboratory results in addition to vital signs in a machine learning algorithm to predict in-hospital cardiac arrest: a single-center retrospective cohort study. PLoS One 2020;15(7):e0235835 [FREE Full text] [CrossRef] [Medline]
    36. Balachander T, Pradeep K, Balaji JS. An integrated approach for early risk detection of sudden cardiac death using machine learning approach. Int J Adv Sci Technol 2020;29(6):2500-2509 [FREE Full text]
    37. RamKumar RP, Polepaka S. Performance comparison of random forest classifier and convolution neural network in predicting heart diseases. In: Advances in Intelligent Systems and Computing. Singapore: Springer; 2020:683-691.
    38. Kwon JM, Kim KH, Jeon KH, Lee SY, Park JJ, Oh BH. Artificial intelligence algorithm for predicting cardiac arrest using electrocardiography. Scand J Trauma Resusc Emerg Med 2020 Oct 06;28(1):98 [FREE Full text] [CrossRef] [Medline]
    39. Chang HK, Wu CT, Liu JH, Lim WS, Wang HC, Chiu SI, et al. Early detecting in-hospital cardiac arrest based on machine learning on imbalanced data. In: Proceedings of the IEEE International Conference on Healthcare Informatics (ICHI). 2019 Presented at: IEEE International Conference on Healthcare Informatics (ICHI); June 10-13, 2019; Xi'an, China. [CrossRef]
    40. Chauhan U, Kumar V, Chauhan V, Tiwary S, Kumar A. Cardiac Arrest Prediction using Machine Learning Algorithms. In: Proceedings of the 2nd International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT). 2019 Presented at: 2nd International Conference on Intelligent Computing, Instrumentation and Control Technologies (ICICICT); 5-6 July 2019; Kannur, India. [CrossRef]
    41. Lai D, Zhang Y, Zhang X, Su Y, Bin Heyat MB. An automated strategy for early risk identification of sudden cardiac death by using machine learning approach on measurable arrhythmic risk markers. IEEE Access 2019;7(2):94701-94716. [CrossRef]
    42. Acharya UR, Fujita H, Sudarshan V, Ghista D, Eugene L, Koh J. Automated prediction of sudden cardiac death risk using kolmogorov complexity and recurrence quantification analysis features extracted from HRV signals. In: Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics. 2019 Presented at: IEEE International Conference on Systems, Man, and Cybernetics; Oct. 9-12, 2015; Hong Kong, China p. 94701-94716. [CrossRef]
    43. Acharya UR, Joseph KP, Kannathal N, Lim CM, Suri JS. Heart rate variability: a review. Med Biol Eng Comput 2006 Dec;44(12):1031-1051. [CrossRef] [Medline]
    44. ChuDuc H, NguyenPhan K, NguyenViet D. A review of heart rate variability and its applications. APCBEE Procedia 2013;7:80-85. [CrossRef]
    45. Vuoti AO, Tulppo MP, Ukkola OH, Junttila MJ, Huikuri HV, Kiviniemi AM, et al. Prognostic value of heart rate variability in patients with coronary artery disease in the current treatment era. PLoS One 2021;16(7):e0254107 [FREE Full text] [CrossRef] [Medline]
    46. Devi R, Tyagi HK, Kumar D. Heart rate variability analysis for early stage prediction of sudden cardiac death. World Academy of Science, Engineering and Technology. 2007.   URL: https:/​/publications.​waset.org/​10004326/​heart-rate-variability-analysis-for-early-stage-prediction-of-sudden-cardiac-death [accessed 2021-05-14]
    47. Park SJ, Cho K, Kwon O, Park H, Lee Y, Shim WH, et al. Development and validation of a deep-learning-based pediatric early warning system: a single-center study. Biomed J 2021 Jan:In Press. [CrossRef]
    48. Jang D, Kim J, Jo YH, Lee JH, Hwang JE, Park SM, et al. Developing neural network models for early detection of cardiac arrest in emergency department. Am J Emerg Med 2020 Jan;38(1):43-49. [CrossRef] [Medline]
    49. Tylman W, Waszyrowski T, Napieralski A, Kamiński M, Trafidło T, Kulesza Z, et al. Real-time prediction of acute cardiovascular events using hardware-implemented Bayesian networks. Comput Biol Med 2016 Feb 01;69:245-253. [CrossRef] [Medline]
    50. Somanchi S, Adhikari S, Lin A, Eneva E, Ghani R. Early prediction of cardiac arrest (Code Blue) using electronic medical records. In: Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. 2015 Presented at: KDD '15: The 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining; August 10 - 13, 2015; Sydney NSW Australia p. 2119-2126. [CrossRef]
    51. Kennedy CE, Aoki N, Mariscalco M, Turley JP. Using time series analysis to predict cardiac arrest in a PICU. Pediatr Crit Care Med 2015;16(9):332-339. [CrossRef]
    52. Churpek MM, Yuen TC, Winslow C, Meltzer DO, Kattan MW, Edelson DP. Multicenter comparison of machine learning methods and conventional regression for predicting clinical deterioration on the ward. Crit Care Med 2016 Feb;44(2):368-374 [FREE Full text] [CrossRef] [Medline]
    53. Badriyah T, Briggs JS, Meredith P, Jarvis SW, Schmidt PE, Featherstone PI, et al. Decision-tree early warning score (DTEWS) validates the design of the National Early Warning Score (NEWS). Resuscitation 2014 Mar;85(3):418-423. [CrossRef] [Medline]
    54. Ho JC, Park Y. Learning from different perspectives: robust cardiac arrest prediction via temporal transfer learning. In: Proceedings of the 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC). 2017 Presented at: 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC); July 11-15, 2017; Jeju, Korea (South). [CrossRef]
    55. Hong S, Lee S, Lee J, Cha WC, Kim K. Prediction of cardiac arrest in the emergency department based on machine learning and sequential characteristics: model development and retrospective clinical validation study. JMIR Med Inform 2020 Aug 04;8(8):e15932 [FREE Full text] [CrossRef] [Medline]
    56. Cho K, Kwon O, Kwon J, Lee Y, Park H, Jeon K, et al. Detecting patient deterioration using artificial intelligence in a rapid response system. Crit Care Med 2020;48(4):285-289. [CrossRef]
    57. Liu JH, Chang HK, Wu CT, Lim WS, Wang HC, Jang JS. Machine learning based early detection system of cardiac arrest. In: Proceedings of the International Conference on Technologies and Applications of Artificial Intelligence (TAAI). 2019 Presented at: International Conference on Technologies and Applications of Artificial Intelligence (TAAI); Nov. 21-23, 2019; Kaohsiung, Taiwan. [CrossRef]
    58. Javan SL, Sepehri MM, Javan ML, Khatibi T. An intelligent warning model for early prediction of cardiac arrest in sepsis patients. Comput Methods Programs Biomed 2019 Sep;178:47-58. [CrossRef] [Medline]
    59. Majumder AJ, ElSaadany YA, Young R, Ucci DR. An energy efficient wearable smart IoT system to predict cardiac arrest. Adv Hum Comput Interact 2019 Feb 12;2019:1-21. [CrossRef]
    60. Kim J, Chae M, Chang H, Kim Y, Park E. Predicting cardiac arrest and respiratory failure using feasible artificial intelligence with simple trajectories of patient data. J Clin Med 2019 Aug 29;8(9):1336 [FREE Full text] [CrossRef] [Medline]
    61. Akrivos E, Papaioannou V, Maglaveras N, Chouvarda I. Prediction of cardiac arrest in intensive care patients through machine learning. In: Maglaveras N, Chouvarda I, de Carvalho P, editors. Precision Medicine Powered by pHealth and Connected Health. Singapore: Springer; 2018:25-29.
    62. Kwon J, Lee Y, Lee Y, Lee S, Park J. An algorithm based on deep learning for predicting in‐hospital cardiac arrest. J Am Heart Assoc 2018 Jul 03;7(13):e008678. [CrossRef]
    63. Rohila A, Sharma A. Detection of sudden cardiac death by a comparative study of heart rate variability in normal and abnormal heart conditions. Biocybern Biomed Eng 2020 Jul 03;40(3):1140-1154. [CrossRef]
    64. Raka A, Naik G, Chai R. Computational algorithms underlying the time-based detection of sudden cardiac arrest via electrocardiographic markers. Appl Sci 2017 Sep 16;7(9):954. [CrossRef]
    65. Murugappan M, Murukesan L, Omar I, Khatun S, Murugappan S. Time domain features based sudden cardiac arrest prediction using machine learning algorithms. J Med Imaging Hlth Inform 2015 Nov 01;5(6):1267-1271. [CrossRef]
    66. Ebrahimzadeh E, Pooyan M, Bijar A. A novel approach to predict sudden cardiac death (SCD) using nonlinear and time-frequency analyses from HRV signals. PLoS One 2014 Feb 4;9(2):e81896 [FREE Full text] [CrossRef] [Medline]
    67. Fernandes M, Mendes R, Vieira SM, Leite F, Palos C, Johnson A, et al. Risk of mortality and cardiopulmonary arrest in critical patients presenting to the emergency department using machine learning and natural language processing. PLoS One 2020 Apr 2;15(4):e0230876 [FREE Full text] [CrossRef] [Medline]
    68. Ebrahimi Z, Loni M, Daneshtalab M, Gharehbaghi A. A review on deep learning methods for ECG arrhythmia classification. Expert Syst Appl X 2020 Sep;7:100033. [CrossRef]
    69. Sahoo S, Dash M, Behera S, Sabut S. Machine learning approach to detect cardiac arrhythmias in ECG signals: a survey. IRBM 2020 Aug;41(4):185-194. [CrossRef]
    70. Lonsdale H, Jalali A, Ahumada L, Matava C. Machine learning and artificial intelligence in pediatric research: current state, future prospects, and examples in perioperative and critical care. J Pediatr 2020 Jun;221S:3-10. [CrossRef] [Medline]
    71. Hong S, Zhou Y, Shang J, Xiao C, Sun J. Opportunities and challenges of deep learning methods for electrocardiogram data: a systematic review. Comput Biol Med 2020 Jul;122:103801. [CrossRef] [Medline]
    72. Itchhaporia D. Artificial intelligence in cardiology. Trends Cardiovasc Med 2020 Nov 23:A. [CrossRef] [Medline]

    ‎
    AI: artificial intelligence
    DL: deep learning
    DT: decision tree
    HRV: heart rate variability
    KNN: k-nearest neighbors
    LR: logistic regression
    MeSH: Medical Subject Headings
    ML: machine learning
    OHCA: out-of-hospital cardiac arrest
    PRISMA: Preferred Reporting Items for Systematic Reviews and Meta-Analyses
    RF: random forest
    SVM: support vector machine

    Edited by C Lovis; submitted 28.05.21; peer-reviewed by J Walsh, P Sarajlic; comments to author 23.09.21; revised version received 07.10.21; accepted 10.10.21; published 17.12.21

    Copyright

    ©Asma Alamgir, Osama Mousa, Zubair Shah. Originally published in JMIR Medical Informatics (https://medinform.jmir.org), 17.12.2021.

    This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on https://medinform.jmir.org/, as well as this copyright and license information must be included.