Hostname: page-component-848d4c4894-wg55d Total loading time: 0 Render date: 2024-06-01T13:44:15.780Z Has data issue: false hasContentIssue false

Applications and Performance of Machine Learning Algorithms in Emergency Medical Services: A Scoping Review

Published online by Cambridge University Press:  17 May 2024

Ahmad Alrawashdeh*
Affiliation:
Department of Allied Medical Sciences, Jordan University of Science and Technology, Irbid, Jordan
Saeed Alqahtani
Affiliation:
Department of Emergency Medical Services, Prince Sultan Military College for Health Sciences, Dhahran, Saudi Arabia
Zaid I. Alkhatib
Affiliation:
Department of Allied Medical Sciences, Jordan University of Science and Technology, Irbid, Jordan
Khalid Kheirallah
Affiliation:
Department of Public Health and Family Medicine, Faculty of Medicine, Jordan University of Science and Technology, Irbid, Jordan
Nebras Y. Melhem
Affiliation:
Department of Anatomy, Physiology and Biochemistry, Faculty of Medicine, The Hashemite University, Zarqa, Jordan
Mahmoud Alwidyan
Affiliation:
Department of Allied Medical Sciences, Jordan University of Science and Technology, Irbid, Jordan
Arwa M. Al-Dekah
Affiliation:
Kernel research and data analytics center, Irbid, Jordan
Talal Alshammari
Affiliation:
Department of Emergency Medical Care, College of Applied Medical Sciences, Imam Abdulrahman Bin Faisal University, Dammam, Saudi Arabia
Ziad Nehme
Affiliation:
Ambulance Victoria, Doncaster, Victoria, Australia School of Public Health and Preventive Medicine, Monash University, Melbourne, Victoria, Australia
*
Correspondence: Ahmad Alrawashdeh Department of Allied Medical Science Jordan University of Science and Technology Irbid 22110, Jordan E-mail: aaalrawashdeh@just.edu.jo
Rights & Permissions [Opens in a new window]

Abstract

Objective:

The aim of this study was to summarize the literature on the applications of machine learning (ML) and their performance in Emergency Medical Services (EMS).

Methods:

Four relevant electronic databases were searched (from inception through January 2024) for all original studies that employed EMS-guided ML algorithms to enhance the clinical and operational performance of EMS. Two reviewers screened the retrieved studies and extracted relevant data from the included studies. The characteristics of included studies, employed ML algorithms, and their performance were quantitively described across primary domains and subdomains.

Results:

This review included a total of 164 studies published from 2005 through 2024. Of those, 125 were clinical domain focused and 39 were operational. The characteristics of ML algorithms such as sample size, number and type of input features, and performance varied between and within domains and subdomains of applications. Clinical applications of ML algorithms involved triage or diagnosis classification (n = 62), treatment prediction (n = 12), or clinical outcome prediction (n = 50), mainly for out-of-hospital cardiac arrest/OHCA (n = 62), cardiovascular diseases/CVDs (n = 19), and trauma (n = 24). The performance of these ML algorithms varied, with a median area under the receiver operating characteristic curve (AUC) of 85.6%, accuracy of 88.1%, sensitivity of 86.05%, and specificity of 86.5%. Within the operational studies, the operational task of most ML algorithms was ambulance allocation (n = 21), followed by ambulance detection (n = 5), ambulance deployment (n = 5), route optimization (n = 5), and quality assurance (n = 3). The performance of all operational ML algorithms varied and had a median AUC of 96.1%, accuracy of 90.0%, sensitivity of 94.4%, and specificity of 87.7%. Generally, neural network and ensemble algorithms, to some degree, out-performed other ML algorithms.

Conclusion:

Triaging and managing different prehospital medical conditions and augmenting ambulance performance can be improved by ML algorithms. Future reports should focus on a specific clinical condition or operational task to improve the precision of the performance metrics of ML models.

Type
Research Report
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2024. Published by Cambridge University Press on behalf of World Association for Disaster and Emergency Medicine

Introduction

Machine learning (ML) is a valuable and increasingly common tool for modern health care systems. Across many medical areas, ML algorithms have aided health care professionals to provide an accurate diagnosis for a disease, make an appropriate clinical decision for a case, and predict an outcome for a treatment intervention. Reference May1,Reference Obermeyer and Emanuel2 However, many of the applications of ML algorithms have focused on improving the safety and quality of care for patients within the in-hospital setting. Reference Obermeyer and Emanuel2Reference Chang, Yu and Yoon4 The potential benefits of ML applications in the prehospital setting are not yet well-established. In-depth knowledge regarding ML applications on prehospital patient care is important for identifying effective technologies and providing guidance for ML-driven research in this specific domain.

Personnel from Emergency Medical Services (EMS) often respond to prehospital cases requiring a rapid response time, early recognition of a life-threatening condition, and timely treatments. Across these time-sensitive tasks, there is evidence that ML algorithms have the potential to augment EMS performance. For example, the response time of a dynamic ambulance re-deployment strategy has been reduced by 1.5 minutes using ML algorithms in reference to conventional methods that are manually designed. Reference Ji, Zheng and Wang5 In addition, dispatchers of EMS who used a clinical decision support algorithm to recognize out-of-hospital cardiac arrest (OHCA) were able to capture more cases within a shorter time interval than those who used the conventional “No-No-Go” approach. Reference Blomberg, Christensen and Lippert6 Despite this, according to the 2015-2020 report for approving ML-based medical products in the United States and Europe, none of the EMS-guided ML applications have been translated into a real product. Reference Muehlematter, Daniore and Vokinger7

To date, few reviews have summarized the application of ML in emergency medicine, but none is exclusive to prehospital emergency medical care or ambulance service. Reference Kirubarajan, Taher and Khan8Reference Kareemi, Vaillancourt and Rosenberg10 While many EMS-guided ML applications have been developed, it is difficult to determine the value of each application in the absence of consolidated evidence. This review, therefore, aimed to systematically review and describe the literature on EMS-guided ML applications in both clinical and operational purposes, and further to describe the characteristics of ML algorithms and their performance.

Methods

The study approach involved conducting a scoping review of ML applications in EMS, complemented by a quantitative synthesis of their performance metrics. This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Scoping Reviews (PRISMA-ScR) guidelines. Reference Tricco, Lillie and Zarin11 The protocol of this review is registered with PROSPERO (ID: CRD42021271256).

Search Strategy

The literature was systematically reviewed from inception until January 9, 2024 across four databases, including: Medline (US National Library of Medicine, National Institutes of Health; Bethesda, Maryland USA); Scopus (Elsevier; Amsterdam, Netherlands); CINAHL (EBSCO Information Services; Ipswich, Massachusetts USA); and Computers and Applied Sciences (EBSCO Information Services; Ipswich, Massachusetts USA). A comprehensive search strategy was created using relevant keywords, phrases, text terms, and subject headings (Supplementary Material S1 Table; available online only). The search strategy was designed to capture all reports related to EMS, ambulance, or paramedic and ML algorithms, deep learning, or neural networks. Search results were imported to the EndNote reference software (Version X9; Clarivate Analytics; Philadelphia, Pennsylvania USA). After removing duplicates, search results were then imported to Covidence software (Covidence; Melbourne, VIC, Australia) for managing and streamlining the review. 12

Study Selection

Two independent reviewers (AA and ZA) initially screened the titles and abstracts for eligibility. At this stage, a study was deemed pertinent if it was written in English; original research; peer-reviewed trials, observational, or experimental studies; and utilized a data-driven methodology involving at least one ML algorithm aimed to classify or predict an EMS-specific outcome related to either clinical or operational applications specific to EMS use.

For a study to be classified under clinical application, the implemented ML algorithm must have been engineered with the aim of augmenting the predictive or classifying capabilities concerning a clinical condition, a clinical intervention, or a potential clinical outcome. For operational applications, the ML algorithm must have been geared towards enhancing the operational efficiency of an EMS system’s resources such as optimizing, streamlining, or improving the use of EMS resources.

Conference proceedings, news, case reports, editorial letters, commentaries, reviews, and non-English reports were excluded. Studies that have used EMS data to aid emergency department and in-hospital outcomes were also excluded. After screening the titles and abstracts, a full-text screening was carried out by the two reviewers (AA and ZA) for final inclusion. Any disagreement was resolved through consensus.

Data Collection

A standardized extraction form was developed to collect relevant data from the included studies. Baseline characteristics such as the first author, year of publication, study objective and design, location, duration, data source, and the number of cases were extracted. Information related to ML algorithms such as the number and types of ML algorithms used, input features, output outcomes, and model performance were also extracted. Additionally, the values of the performance metrics were extracted. This includes the values of the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and positive predictive value. One reviewer extracted the relevant data (ZA), and this was cross-checked by a second reviewer (AA). Disagreements were resolved through consensus.

Data Synthesis

The applications of ML algorithms across EMS studies were dichotomized into clinical and operational domains. Study characteristics such as the year of publication, geographic region, study design, data source, and number of ML algorithms employed across these domains were summarized using descriptive statistics. Numerical variables were, as appropriate, reported as medians and ranges and categorical variables were reported as counts and percentages.

The studies were then classified within the clinical domain into subdomains based on the medical condition, including OHCA, cardiovascular diseases (CVDs), trauma, and others (ie, including unspecified medical conditions). Similarly, the studies were classified within the operational domain into subdomains, and this was based on the operational task, including ambulance allocation (ie, ambulance demand forecasting and relocation), deployment (ie, ambulance dispatching), detection (ie, ambulance recognition by image or siren capturing), route optimization (ie, route selection), and quality assurance (ie, quality indicators or documentation assessment). All subdomains were assessed, using descriptive statistics as described above, against the following variables: number of studies, clinical tasks, number of cases, the number and type of input features, methods of ML algorithm, and performance metrics.

The values of performance metrics were evaluated by estimating the unweighted medians and ranges. The performance metrics estimated included AUC, accuracy, sensitivity, and specificity for the best ML algorithm as declared by individual studies. STATA statistical software, version 16.0 (Stata Corp; College Station, Texas USA) was used to carry out all statistical analyses.

Results

The search strategy identified 6,661 records, of which 1,478 were duplicates and 5,183 were excluded in the title and abstract screening. A total of 647 full-text articles were retrieved and assessed for eligibility. Finally, a total of 164 studies were included in this review (Figure 1).

Figure 1. PRISMA Flow Chart.

Abbreviations: ML, machine learning; EMS, Emergency Medical Services; CAS, Computer & Applied Sciences.

Characteristics of Included Studies

Table 1 summarizes the characteristics of the included studies by domain. Overall (n = 164), the median year of publication was 2020 ranging from 2005 through 2024; the number of publications exponentially increased after 2017. Figure 2 shows the distribution of the included studies over the years for different characteristics of the included studies. More EMS-guided ML applications were reported in the Asia-Pacific region (37.2%) than those that were reported in Europe (29.3%) and the United States (28.7%); most studies were retrospective in design (94.5%) and the data were mainly derived from the EMS records (45.1%) and registries (40.9%); and 51.2% of the included studies used one ML algorithm, whereas 48.8% of them used more than one. More than two-thirds of the included studies deployed ML for a clinical application (125 studies; 76.2%), while 39 (23.8%) studies applied ML for an operational application. All included studies are listed and described in Table S2 and Table S3 in the Supplementary Material (available online only). Operational studies were reported mostly in the Asia-Pacific (51.3%), whereas the United States and Europe had a similar number of studies, with a predominant focus on clinical research.

Table 1. Baseline Characteristics of the Included Studies across the Two Domains

Abbreviations: EMS, Emergency Medical Services; ML, machine learning.

a Reported in 125 studies (n = 104 for clinical domain and n = 21 for operational).

Figure 2. Distribution of the Included Studies Over the Years.

Abbreviations: ML, machine learning; EMS, Emergency Medical Services.

Characteristics of ML Algorithms in the Clinical Domain

Table 2 presents the characteristics of EMS-guided ML algorithms applied in the clinical domain studies, categorized by medical conditions, including OHCA (n = 62 studies; 49.6%), CVDs (n = 19 studies; 15.2%), trauma (n = 24 studies; 19.2%), and others (n = 20 studies; 16.0%). The purpose of most studies in this domain was to primarily triage or diagnose a medical condition (50.0%), followed by predicting a clinical outcome (40.3%) and a clinical intervention (9.7%).

Table 2. Characteristics of EMS-Guided ML Applications across Studies in Clinical Conditions

Abbreviations: OHCA, out-of-hospital cardiac arrest; CVD, cardiovascular diseases; ECG, electrocardiography; AUC, area under the receiver operating characteristic curve; ML, machine learning.

a Number of cases/instances is missing in 23 studies.

b Number of input features is missing in 54 studies.

Among OHCA studies, the most frequent purposes were predicting a clinical outcome (n = 38; 61.3%) and detecting or triaging OHCA (n = 22; 35.5%). Almost all CVD studies employed ML for triage and diagnosis (n = 18; 94.7%). The purpose of most studies concentrating on trauma was to triage traumatic cases (n = 11; 45.8%) or to predict a clinical outcome (n = 9; 37.5%).

The median number of cases/instances for most studies within the clinical domain was 3,405 (n = 102 studies; range: 37-8,981,181). The sample size was larger among studies focusing on trauma and other medical conditions than studies focusing on OHCA and CVDs. When reported, the median number of input features for studies within the clinical domain was 19 (n = 71 studies; range: 1-886). The number of input features was larger for CVD studies compared to other clinical conditions. Among all clinical domain studies, the most frequent type of input features was clinical (ie, demographics, vitals, and findings of physical examination) obtained from EMS medical records (53.7%), followed by electrocardiograph (ECG) strips (35.0%), text scripts (11.0%), and audio recordings (5.1%). The majority of the studies focused on OHCA involved ECG input features (n = 36; 59.0%), while most studies related to all other conditions involved clinical features.

The most frequently employed ML algorithms were neural network (n = 60 studies; 50.0%), followed by random forest (n = 57 studies; 48.3%), logistic regression (n = 47 studies; 39.8%), and support vector machine (n = 43 studies; 36.4%). This distribution was similar within all clinical conditions. To evaluate the performance of the ML algorithms, the most frequently used performance metric was AUC (n = 76; 60.8%), followed by sensitivity (n = 73; 58.4%) and specificity (n = 67; 53.6%).

Performance of ML Algorithms in the Clinical Domain

In the clinical domain, the median AUC for the best ML algorithms was 85.6% (n = 76 studies; range: 64.0%-99.9%). The median accuracy was 88.1% (n = 40 studies; range: 57.8%-99.2%), sensitivity was 86.0% (n = 73 studies; range: 38.8%-99.7%), and specificity was 86.5% (n = 67 studies; range: 11.0%-99.9%).

Figure 3 depicts the performance of the best ML algorithms employed to triage and diagnose clinical conditions and predict clinical outcomes across the clinical conditions and ML algorithms. Overall, the medians of accuracy, sensitivity, and specificity of the ML algorithms designed to triage and diagnose a clinical condition (≥90.0%) were higher than the algorithms designed to predict a clinical outcome (<90.0%).

Figure 3. Median and Range of the Best Machine Learning (ML) Algorithm amid to Triage or Diagnose the Clinical Condition and Predict their Clinical Outcomes.

Abbreviations: OHCA, out-of-hospital cardiac arrest; CVD, cardiovascular diseases; AUC, area under the receiver operating characteristic curve.

The median values of most performance metrics for ML algorithms deployed to triage or diagnose clinical conditions were highest in the context of OHCA and lowest in traumatic emergencies. However, compared to algorithms predicting a clinical outcome after OHCA, algorithms predicting a clinical outcome after traumatic emergencies had a higher median AUC (87.3% versus 90.5%, respectively) and sensitivity (81.7% versus 91.0%), but lower accuracy (86.3% versus 87.6%) and specificity (80.3% versus 61.0%). Neural network, as well as ensemble algorithms, appeared to outperform other ML algorithms across clinical purposes and conditions.

Characteristics of ML Algorithms in the Operational Domain

Table 3 presents the characteristics of EMS-guided ML algorithms applied in the operational domain studies, categorized by the operational task, including ambulance allocation (n = 21 studies), ambulance detection (n = 5 studies), ambulance deployment (n = 5 studies), route optimization (n = 5 studies), and quality assurance (n = 3 studies).

Table 3. Characteristics of EMS-Guided ML Applications across Studies in Operational Subdomains

Abbreviations: EMS, Emergency Medical Service; ML, machine learning; AUC, area under the receiver operating characteristic curve.

a Number of cases is missing in 19 studies.

b Number of input features is missing in 29 studies.

The overall median sample size was 11,899 (n = 20 studies; range: 77-7,364,275) and the largest sample sizes were reported in studies focusing on ambulance allocation and deployment. The overall median number of input features was 10 (n = 10 studies; range: 3-9,224) and the largest median was found in ambulance deployment studies (18.0; range: 8-9,224). Among studies within the operational domain, the most frequent types of input features were EMS-related data (n = 17 studies; 43.6%), spatiotemporal (n = 13 studies; 33.3%), clinical findings (n = 10 studies; 25.6%), audio recordings (n = 3 studies; 7.7%), and image (n = 3 studies; 7.7%).

The neural network was the most frequently employed ML algorithm (n = 24 studies; 61.5%) across all operational purposes. This was followed by ensemble algorithms (n = 7 studies; 17.9%), support vector machine (n = 6 studies; 15.4%), random forest (n = 6 studies; 15.4%), and decision tree (n = 4 studies; 10.3%). To evaluate the performance of the ML algorithms, accuracy (n = 17 studies; 43.6%) was the most frequently used performance metric, followed by sensitivity (n = 10 studies; 25.6%) and specificity (n = 6 studies; 15.4%). Notably, a total of 18 (46.2%) studies did not report the performance of their ML models.

Performance of ML Algorithms in the Operational Domain

For the operational domain, the median AUC of the best ML algorithms was 96.1% (n = 6 studies; range: 89.8%-99.0%), accuracy was 90.0% (n = 17; range: 24.5%-98.7%), sensitivity was 94.4% (n = 10; range: 83.0%-99.5%), and specificity was 87.7% (n = 6; range: 66.8%-99.4%).

Figure 4 illustrates the performance of the best ML models categorized by operational tasks and types of ML algorithms. The medians of accuracy metrics across studies related to ambulance detection and deployment were higher than those related to ambulance allocation, route optimization, and quality assurance. The frequently applied algorithm was a neural network which, along with the ensemble, out-performed other algorithms to a certain level.

Figure 4. Median and Range of the Best Machine Learning (ML) Algorithm Categorized by Operational Tasks and Algorithm Type.

Abbreviation: AUC, area under the receiver operating characteristic curve.

Discussion

The journey of prehospital emergency care for a patient, including call receipt, dispatch, travel, triage, diagnosis, treatment, and transport to appropriate definitive care, is efficiently managed by multiple operational and medical components of EMS. This review showed that ML algorithms were implemented in almost all EMS components, including dispatch and communication system, first responders, ambulances and paramedics, medical oversight, and quality improvement. In addition, ML has the potential to augment the decision-making capabilities of EMS personnel in both clinical and operational domains or even supplement their judgment in specific aspects of EMS components. However, the integration of ML in EMS is still evolving, and further research, development, and validation may be needed to fully leverage its capabilities in clinical and operational applications. Considerable heterogeneity was observed in the included studies in terms of population characteristics, the ML algorithms employed, input features, and performance metrics which limited a more comprehensive review. The implications and performance of ML should be in-depth evaluated for each clinical condition, outcome, or operational task.

Several studies showed the potential of ML algorithms in improving emergency medical response to OHCA. More than 15 studies concentrated on implementing ML for classifying cardiac arrest ECG rhythms or detecting the return of spontaneous circulation aiming to improve resuscitation performance and provide timely guidance for delivering electrical defibrillation. Multiple ML algorithms exhibit outstanding performance (AUC > 0.90), demonstrating remarkable accuracy and practical applicability in the real-world context. For example, Jaureguibeitia, et al proposed novel deep learning architectures (AUC = 98.6%) for shock decisions using ECG segments as brief as one second, which likely shorten interruptions in resuscitation for rhythm analysis. Reference Jaureguibeitia, Zubia and Irusta13 Elola, et al also implemented deep neural networks for pulse detection using only the ECG (AUC = 86.2%), Reference Elola, Aramendi and Irusta14 while Alonso, et al (AUC = 92.6%) combined ECG and thoracic impedance signals for superior pulse detection. Reference Alonso, Irusta and Aramendi15 Further, deep neural networks have demonstrated the ability to detect OHCA at the time of calling for help, arguably with greater accuracy and speed than the EMS telecommunicators. Reference Blomberg, Christensen and Lippert6,Reference Blomberg, Folke and Ersbøll16,Reference Byrsell, Claesson and Ringh17

Predictive ML models were also developed for three clinical outcomes after OHCA, including the return of spontaneous circulation, survival, and neurological recovery. Among many ML models, deep neural networks achieved the highest performance (AUC = 0.84 to 0.90) in predicting the success of defibrillation using ECG features. Reference He, Lu and Zhang18Reference Shandilya, Kurz and Ward20 Acknowledging the limited number of cases in most of these studies (n <500 cases), larger and more diverse samples of cases are likely essential to limit the risk of bias and to enhance the performance, reliability, and applicability of ML algorithms in predicting defibrillation success. Reference Yang21

Deep neural networks and ensemble ML algorithms performed exceptionally (AUC = 0.90 to 0.96) among others to predict survival and neurological recovery using prehospital clinical and EMS data. Reference Cheng, Chiu and Zeng22Reference Kawai, Okuda and Kinoshita24 However, these predictive models exhibit considerable heterogeneity in quality, type of population, input and output features, and algorithm selection, potentially affecting generalizability. Additionally, the complexity and lack of transparency in certain ML algorithms might obstruct their integration into clinical practice. Reference Yang21 Future ML studies should discuss features attribution or explain the ML models output by using several methods such as Shapley Additive Explanation, Local Interpretable Model-agnostic Explanations, and integrated gradients.

Most ML models have also been shown to improve the detection of CVDs in the prehospital setting. A deep learning model achieved an impressive AUC score of 0.997 in detecting ST-elevation myocardial infarction (STEMI). Reference Chen, Wang and Liu25 This model has been deployed in a prehospital 12-lead ECG device and resulted in a shorter median time between first medical contact and hospital arrival (18.5 minutes), compared to the weighted mean estimated by a previous meta-analysis (41 minutes). Reference Alrawashdeh, Nehme and Williams26 Further, Bouzid, et al have developed a fusion model to detect non-STEMI using prehospital 12-lead ECG which significantly outperformed the performance of expert clinicians and commercial software. Reference Bouzid, Faramand and Gregg27 In addition, prehospital stroke diagnostic ML algorithms have been shown to be accurate in identifying and differentiating stroke using prehospital clinical presentation variables. Reference Chen, Zhang and Xu28Reference Uchida, Kouno and Yoshimura31 Two studies have shown that neural network algorithms out-performed other available prehospital stroke prediction scales, such as the Prehospital Acute Stroke Severity scale and the Cincinnati Prehospital Stroke Severity scale. Reference Chen, Zhang and Xu28,Reference Zhang, Zhou and Zhang29 However, a larger number of parameters (ie, 18 variables) were included in these neural network models, which demand more effort and time to be collected by ambulance clinicians.

For traumatic conditions, ML models have been increasingly applied to primarily improve the triage decision by dispatchers, first responders, and ambulance clinicians in various out-of-hospital contexts (ie, general trauma, road accidents, and military). These models were developed either to classify traumatic patients by their level of severity (ie, Injury Severity Score) Reference Candefjord, Sheikh Muhammad and Bangalore32 or to predict the level of treatment needed (ie, advanced life-saving interventions, trauma team activation, and critical care) Reference Morris, Karam and Zolfaghari33,Reference Kang, Cho and Kwon34 or the clinical outcomes (ie, survival). Reference Kim, You and So35 The input features in most models were relatively limited and mainly consisted of demographics, vital signs, other clinical manifestations, or accident-related variables. Neural networks and ensemble techniques (random forest, XGBoosting) were frequently used and had the best performance compared to other models. Two studies found that binary logistic regression performed better than ML algorithms when classifying the severity of injuries. Reference Candefjord, Sheikh Muhammad and Bangalore32,Reference Larsson, Berg and Gellerfors36 This is likely attributed to the limited number of input features in these studies. Overall, the contribution of ML seems to be promising in this context, but further validation is required to better understand the applicability of ML in supporting triage decisions and optimizing resource allocation.

Several ML models have been implemented to solve operational problems and augment the decision making of EMS personnel in operational tasks. The over-arching aims of the operational studies were to reduce ambulance travel time and deploy the appropriate level of prehospital care to the right patients. A variety of supervised, unsupervised, and reinforcement models have been used to improve EMS demand forecasting in different settings to optimize ambulance allocation strategies, and eventually reduce response times. The characteristics of ML models forecasting EMS demand varied in terms of time horizons (hourly, daily, weekly, and monthly demand) and input features (geospatial, temporal, metrological, and EMS data). Several studies aimed to facilitate ambulance passage through traffic congestion by either audio-based and/or visual-based detection. The audio-based models using random forest, support vector machine, and neural networks achieved AUC scores between 0.97-1.00. Reference Dhakal, Damacharla and Javaid37Reference Tran and Tsai40 Visual-based models mainly used deep learning with variable performance (AUC = 0.78-1.00) Reference Tran and Tsai40Reference Roy and Rahman42 while hybrid models incorporated both audio and image inputs and typically relied on deep learning yield strong results (0.86-0.98). Reference Tran and Tsai40,Reference Raman, Kaushik and Rao43 The performance of these models indicates the viability of employing ML to detect ambulances and expedite their passage through intersections and traffic congestion.

Finally, the implementation of the developed ML models in real time was rarely discussed in the included studies. Several challenges such as the generalizability and explainability of these ML models may complicate their implication in clinical and real time. Several studies reported perfect discrimination, especially for operational purposes, which warrant a critical examination of the underlying data and methodologies employed. While perfect discrimination can be attributed to factors like over-fitting, reliance on internal validation, data leakage, or imbalanced datasets, it is imperative to consider the broader context of its applicability, generalizability, and potential limitations in real-world EMS scenarios. For example, the generalization of ML algorithms for ambulance detection using audio or image data may pose significant challenges due to the wide variation in ambulance sirens and morphologies across different EMS systems.

It is also important to acknowledge the complications and limitations associated with the retrospective nature of most included studies. Rich prehospital datasets are still necessary; however, they are likely challenging to obtain, owing to limitations in data quality and availability, patient privacy, and the heterogeneity of EMS system data frameworks. Reference Aggarwal, Sounderajah and Martin44 This necessitates collaborative efforts among researchers, EMS systems, and other health care institutions, as well as policymakers to establish data-sharing frameworks, best practices, and governance models that facilitate the responsible and impactful application of these technologies in the challenging and dynamic landscape of EMS. For better explainability, future ML studies should also discuss features attribution or explain the output of ML models by using available methods such as Shapley additive explanation, local interpretable model-agnostic explanations, and integrated gradients. Reference Petch, Di and Nelson45

Limitations

This review has several limitations. Of note, this review may have some selection bias. It only included English peer-reviewed articles identified by the search strategy. The variation in the terminology of ML and EMS in the literature may also contribute to selection bias. The wide scope of this review may hinder proper summarization of the applications and performance of the ML in specific clinical conditions and operational tasks. For example, the approaches for features selection, model validation, and optimization used in the included studies were not detailed. Future systematic reviews could be done to quantify the use of ML in specific clinical conditions such as detecting OHCA or predicting the clinical outcomes after OHCA.

Also not examined was the risk of bias and quality of the included studies. This is attributed to the large number and the wide variation in the objectives and characteristics of the included studies. Although the performance of the ML algorithms performed well in most applications, the issue of risk of bias should be considered in most ML models, particularly within the analysis domain. Reference Andaur Navarro, Damen and Takada46,Reference Dhiman, Ma and Andaur Navarro47 Several studies had not reported all relevant details such as sampling, input features selection, validation procedures, and model performance, particularly for studies within the operational domain. These missing data also limit the overall analysis and the generalizability of the findings.

Conclusion

This systematic review highlights the diverse applications and potential benefits of ML in EMS. Most studies focused on the use of ML algorithms to enhance the clinical domain, particularly in detecting and predicting the severity, treatment, and clinical outcome of patients with OHCA, CVDs, and trauma. The performance of ML models varied widely, indicating the need for further research and standardization of ML algorithms in EMS. Additionally, the review found that ML can improve the operational management of ambulance services, including optimizing ambulance allocations, reducing travel time, and enhancing resource utilization. This review is expected to provide useful benchmarks and implications for possible future research. Further research is needed to investigate the implementation and quality of ML in specific clinical conditions or operational tasks within the EMS.

Conflicts of interest/funding

This work was supported by the Jordan University of Science and Technology under Grant number 20210307. Ziad Nehme is funded by a National Heart Foundation fellowship. The authors report there are no competing interests to declare.

Supplementary Materials

To view supplementary material for this article, please visit https://doi.org/10.1017/S1049023X24000414

Author Contribution

AA and SA conceived the idea; AA, ZA, and NM conducted study selection and data extraction; AA, SA, and ZA synthesized the findings and wrote the first draft. All authors reviewed and confirmed the final manuscript.

References

May, M. Eight ways machine learning is assisting medicine. Nat Med. 2021;27(1):23.CrossRefGoogle ScholarPubMed
Obermeyer, Z, Emanuel, EJ. Predicting the future - big data, machine learning, and clinical medicine. N Engl J Med. 2016;375(13):12161219.CrossRefGoogle ScholarPubMed
Buchlak, QD, Esmaili, N, Leveque, JC, et al. Machine learning applications to clinical decision support in neurosurgery: an artificial intelligence augmented systematic review. Neurosurg Rev. 2020;43(5):12351253.CrossRefGoogle ScholarPubMed
Chang, H, Yu, JY, Yoon, S, et al. Machine learning-based suggestion for critical interventions in the management of potentially severe conditioned patients in emergency department triage. Sci Rep. 2022;12(1):10537.CrossRefGoogle ScholarPubMed
Ji, S, Zheng, Y, Wang, Z, et al. A deep reinforcement learning-enabled dynamic redeployment system for mobile ambulances. Proceedings of the ACM on Interactive, Mobile, Wearable, and Ubiquitous Technologies. 2019;3:120.CrossRefGoogle Scholar
Blomberg, SN, Christensen, HC, Lippert, F, et al. Effect of machine learning on dispatcher recognition of out-of-hospital cardiac arrest during calls to Emergency Medical Services: a randomized clinical trial. JAMA Netw Open. 2021;4(1):e2032320.CrossRefGoogle ScholarPubMed
Muehlematter, UJ, Daniore, P, Vokinger, KN. Approval of artificial intelligence and machine learning-based medical devices in the USA and Europe (2015–20): a comparative analysis. Lancet Digit Health. 2021;3(3):e195e203.CrossRefGoogle ScholarPubMed
Kirubarajan, A, Taher, A, Khan, S, et al. Artificial intelligence in emergency medicine: a scoping review. J Am Coll Emerg Physicians Open. 2020;1(6):16911702.CrossRefGoogle Scholar
Tang, KJW, Ang, CKE, Constantinides, T, et al. Artificial intelligence and machine learning in emergency medicine. Biocybernetics and Biomedical Engineering. 2021;41:156172.CrossRefGoogle Scholar
Kareemi, H, Vaillancourt, C, Rosenberg, H, et al. Machine learning versus usual care for diagnostic and prognostic prediction in the emergency department: a systematic review. Acad Emerg Med. 2021;28(2):184196.CrossRefGoogle ScholarPubMed
Tricco, AC, Lillie, E, Zarin, W, et al. PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation. Ann Intern Med. 2018;169(7):467473.CrossRefGoogle ScholarPubMed
Covidence Systematic Review Software. VHI, Melbourne, Australia. www.covidence.org. Accessed January 2024.Google Scholar
Jaureguibeitia, X, Zubia, G, Irusta, U, et al. Shock decision algorithms for automated external defibrillators based on convolutional networks. IEEE Access. 2020;8:154746154758.CrossRefGoogle Scholar
Elola, A, Aramendi, E, Irusta, U, et al. Deep neural networks for ECG-based pulse detection during out-of-hospital cardiac arrest. Entropy. 2019;21(3):305.CrossRefGoogle ScholarPubMed
Alonso, E, Irusta, U, Aramendi, E, et al. A machine learning framework for pulse detection during out-of-hospital cardiac arrest. IEEE Access. 2020;8:161031161041.CrossRefGoogle Scholar
Blomberg, SN, Folke, F, Ersbøll, AK, et al. Machine learning as a supportive tool to recognize cardiac arrest in emergency calls. Resuscitation. 2019;138:322329.CrossRefGoogle ScholarPubMed
Byrsell, F, Claesson, A, Ringh, M, et al. Machine learning can support dispatchers to better and faster recognize out-of-hospital cardiac arrest during emergency calls: a retrospective study. Resuscitation. 2021;162:218226.CrossRefGoogle ScholarPubMed
He, M, Lu, Y, Zhang, L, et al. Combining amplitude spectrum area with previous shock information using neural networks improves prediction performance of defibrillation outcome for subsequent shocks in out-of-hospital cardiac arrest patients. PLoS One. 2016;11(2):e0149115.CrossRefGoogle ScholarPubMed
He, M, Gong, Y, Li, Y, et al. Combining multiple ECG features does not improve prediction of defibrillation outcome compared to single features in a large population of out-of-hospital cardiac arrests. Critical Care. 2015;19:425.CrossRefGoogle ScholarPubMed
Shandilya, S, Kurz, MC, Ward, KR, et al. Integration of attributes from non-linear characterization of cardiovascular time-series for prediction of defibrillation outcomes. PLoS One. 2016;11(1):e0141313.CrossRefGoogle ScholarPubMed
Yang, CC. Explainable artificial intelligence for predictive modeling in healthcare. J Healthc Inform Res. 2022;6(2):228239.CrossRefGoogle ScholarPubMed
Cheng, C-Y, Chiu, IM, Zeng, W-H, et al. Machine learning models for survival and neurological outcome prediction of out-of-hospital cardiac arrest patients. BioMed Res Int. 2021;2021:9590131.CrossRefGoogle ScholarPubMed
Kwon, J-m, Jeon, K-H, Kim, HM, et al. Deep-learning-based out-of-hospital cardiac arrest prognostic system to predict clinical outcomes. Resuscitation. 2019;139:8491.CrossRefGoogle ScholarPubMed
Kawai, Y, Okuda, H, Kinoshita, A, et al. Visual assessment of interactions among resuscitation activity factors in out-of-hospital cardiopulmonary arrest using a machine learning model. PLoS One. 2022;17:e0273787.CrossRefGoogle ScholarPubMed
Chen, K-W, Wang, Y-C, Liu, M-H, et al. Artificial intelligence-assisted remote detection of ST-elevation myocardial infarction using a mini-12-lead electrocardiogram device in prehospital ambulance care. Front Cardiovas Med. 2022;9:1001982.CrossRefGoogle ScholarPubMed
Alrawashdeh, A, Nehme, Z, Williams, B, et al. Emergency medical service delays in ST-elevation myocardial infarction: a meta-analysis. Heart. 2020;106(5):365373.CrossRefGoogle ScholarPubMed
Bouzid, Z, Faramand, Z, Gregg, RE, et al. Novel ECG features and machine learning to optimize culprit lesion detection in patients with suspected acute coronary syndrome. J Electrocardiol. 2021;69s:3137.CrossRefGoogle ScholarPubMed
Chen, Z, Zhang, R, Xu, F, et al. Novel prehospital prediction model of large vessel occlusion using artificial neural network. Front Aging Neurosci. 2018;10:181.CrossRefGoogle ScholarPubMed
Zhang, Z, Zhou, D, Zhang, J, et al. Multilayer perceptron-based prediction of stroke mimics in prehospital triage. Sci Rep. 2022;12(1):17994.CrossRefGoogle ScholarPubMed
Hayashi, Y, Shimada, T, Hattori, N, et al. A prehospital diagnostic algorithm for strokes using machine learning: a prospective observational study. Sci Rep. 2021;11(1):20519.CrossRefGoogle ScholarPubMed
Uchida, K, Kouno, J, Yoshimura, S, et al. Development of machine learning models to predict probabilities and types of stroke at prehospital stage: the Japan Urgent Stroke Triage Score Using Machine Learning (JUST-ML). Transl Stroke Res. 2022;13(3):370381.CrossRefGoogle ScholarPubMed
Candefjord, S, Sheikh Muhammad, A, Bangalore, P, et al. On-Scene Injury Severity Prediction (OSISP) machine learning algorithms for motor vehicle crash occupants in US. J Transport Health. 2021;22:101124.CrossRefGoogle Scholar
Morris, R, Karam, BS, Zolfaghari, EJ, et al. Need for emergent intervention within 6 hours: a novel prediction model for hospital trauma triage. Prehosp Emerg Care. 2022;26(4):556565.CrossRefGoogle Scholar
Kang, D-Y, Cho, K-J, Kwon, O, et al. Artificial intelligence algorithm to predict the need for critical care in prehospital emergency medical services. Scand J Trauma Resusc Emerg Med. 2020;28(1):17.CrossRefGoogle ScholarPubMed
Kim, D, You, S, So, S, et al. A data-driven artificial intelligence model for remote triage in the prehospital environment. PLoS One. 2018;13(10):e0206006.CrossRefGoogle ScholarPubMed
Larsson, A, Berg, J, Gellerfors, M, et al. The advanced machine learner XGBoost did not reduce prehospital trauma mis-triage compared with logistic regression: a simulation study. BMC Med Inform Decis Mak. 2021;21(1):192.CrossRefGoogle ScholarPubMed
Dhakal, P, Damacharla, P, Javaid, AY, et al. Detection and identification of background sounds to improvise voice interface in critical environments. 2018 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT). 2018:078083.CrossRefGoogle Scholar
Suhaimy, MA, Halim, ISA, Hassan, SLM, et al. Classification of ambulance siren sound with MFCC-SVM. AIP Conference Proceedings. 2020;2306.CrossRefGoogle Scholar
Chinvar, DC, Rajat, M, Bellubbi, RL, et al. Ambulance siren detection using an MFCC based support vector machine. 2021 IEEE International Conference on Mobile Networks and Wireless Communications (ICMNWC). 2021:1-5.CrossRefGoogle Scholar
Tran, VT, Tsai, WH. Acoustic-based emergency vehicle detection using convolutional neural networks. IEEE Access. 2020;8:7570275713.CrossRefGoogle Scholar
Agrawal, K, Nigam, M, Bhattacharya, S, et al. Ambulance detection using image processing and neural networks. Journal of Physics: Conference Series. 2021;2115:012036.Google Scholar
Roy, S, Rahman, MS. Emergency vehicle detection on heavy traffic road from CCTV footage using deep convolutional neural network. 2019 International Conference on Electrical, Computer, and Communication Engineering (ECCE). 2019:1-6.Google Scholar
Raman, A, Kaushik, S, Rao, KR, et al. A hybrid framework for expediting emergency vehicle movement on Indian roads. 2020 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA). 2020:459-464.CrossRefGoogle Scholar
Aggarwal, R, Sounderajah, V, Martin, G, et al. Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis. NPJ Digit Med. 2021;4(1):65.CrossRefGoogle ScholarPubMed
Petch, J, Di, S, Nelson, W. Opening the black box: the promise and limitations of explainable machine learning in cardiology. Can J Cardiol. 2022;38(2):204213.CrossRefGoogle ScholarPubMed
Andaur Navarro, CL, Damen, JAA, Takada, T, et al. Risk of bias in studies on prediction models developed using supervised machine learning techniques: systematic review. BMJ. 2021;375:n2281.CrossRefGoogle ScholarPubMed
Dhiman, P, Ma, J, Andaur Navarro, CL, et al. Risk of bias of prognostic models developed using machine learning: a systematic review in oncology. Diagn Progn Res. 2022;6(1):13.CrossRefGoogle ScholarPubMed
Figure 0

Figure 1. PRISMA Flow Chart.Abbreviations: ML, machine learning; EMS, Emergency Medical Services; CAS, Computer & Applied Sciences.

Figure 1

Table 1. Baseline Characteristics of the Included Studies across the Two Domains

Figure 2

Figure 2. Distribution of the Included Studies Over the Years.Abbreviations: ML, machine learning; EMS, Emergency Medical Services.

Figure 3

Table 2. Characteristics of EMS-Guided ML Applications across Studies in Clinical Conditions

Figure 4

Figure 3. Median and Range of the Best Machine Learning (ML) Algorithm amid to Triage or Diagnose the Clinical Condition and Predict their Clinical Outcomes.Abbreviations: OHCA, out-of-hospital cardiac arrest; CVD, cardiovascular diseases; AUC, area under the receiver operating characteristic curve.

Figure 5

Table 3. Characteristics of EMS-Guided ML Applications across Studies in Operational Subdomains

Figure 6

Figure 4. Median and Range of the Best Machine Learning (ML) Algorithm Categorized by Operational Tasks and Algorithm Type.Abbreviation: AUC, area under the receiver operating characteristic curve.

Supplementary material: File

Alrawashdeh et al. supplementary material

Alrawashdeh et al. supplementary material
Download Alrawashdeh et al. supplementary material(File)
File 689.5 KB