Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>Cardiovascular diseases (CVDs) represent a leading cause of morbidity and mortality worldwide, imposing a substantial burden on global healthcare systems (Hindricks et al., 2020). Early and accurate detection of CVDs is paramount for effective management, timely intervention, and ultimately, improved patient prognosis. Traditional diagnostic methods, while valuable, can be time-consuming and may not always capture the subtle early indicators of disease onset. In recent years, the advent of machine learning (ML) has opened new avenues for enhancing diagnostic capabilities across various medical domains (Kelly et al., 2019). ML algorithms, with their ability to identify complex patterns and correlations within large datasets, hold significant promise for revolutionizing the early detection of CVDs (Juneja, 2021; Shaikh, 2021). This research aims to systematically evaluate the performance of several prominent ML algorithms in predicting the early onset of cardiovascular conditions, thereby assessing their potential impact on clinical practice.</p><p>The increasing availability of electronic health records (EHRs) and sophisticated data analytics tools has facilitated the application of ML in healthcare (Rajkomar et al., 2018). These algorithms can process vast amounts of patient data, including demographic information, medical history, lifestyle factors, and physiological measurements, to identify individuals at high risk of developing CVDs. Previous studies have explored the use of ML for disease detection in various fields, such as neurodegenerative disorders (Nallore et al., 2023), legume crop diseases (Cho, 2024), and even credit card fraud (Gupta, 2023), demonstrating the algorithm's versatility. Within the realm of cardiovascular health, ML has been applied to predict conditions like left ventricular hypertrophy (Beneyto et al., 2023) and atrial septal aneurysm (Alami et al., 2024), and to enhance heart attack prediction (Alshraideh et al., 2024). This study seeks to provide a comparative analysis of several ML techniques specifically for the early detection of a broad spectrum of CVDs, building upon existing research (Unknown, 2021; Ahmad, 2020; Ekong, 2023; Mijwil et al., 2024).</p>
<h2>Literature Review</h2>
<p>The application of machine learning in healthcare has seen exponential growth, driven by advancements in computational power and data availability. For cardiovascular diseases, early detection is crucial for mitigating severe outcomes. Numerous studies have investigated the potential of ML algorithms in this domain. For instance, research has explored the use of algorithms like Support Vector Machines (SVM) and Random Forests (RF) for predicting cardiovascular risk using patient data (Shaikh, 2021; Unknown, 2021). These studies often highlight the ability of ML models to uncover intricate relationships between risk factors and disease development that might be overlooked by conventional statistical methods.</p><p>Several recent works have focused on evaluating the performance of different ML techniques. Ekong (2023) provided an evaluation of various ML techniques for early CVD detection, emphasizing the need for robust comparative studies. Similarly, Mijwil et al. (2024) compared seven different ML algorithms for early CVD detection, assessing their predictive capabilities. Saini (2023) reviewed recent innovations in ML for cardiovascular disease detection, pointing towards ensemble methods and deep learning as promising avenues. Beneyto et al. (2023) developed and validated ML algorithms for predicting the etiology of left ventricular hypertrophy, demonstrating ML's utility in specific cardiovascular conditions. Alami et al. (2024) utilized ML for detecting atrial septal aneurysm on ECG, showcasing the application of ML in analyzing specific diagnostic signals.</p><p>While the promise of ML in healthcare is significant, challenges remain in clinical implementation. Kelly et al. (2019) discussed key challenges in delivering clinical impact with artificial intelligence, including issues related to data quality, interpretability, and regulatory hurdles. The need for rigorous validation and comparison of different algorithms is also evident. Dhawan (2022) explored the employability of ML algorithms across multiple diseases, suggesting a broad applicability but also hinting at the need for disease-specific optimization. The PRISMA 2020 guidelines provide a framework for reporting systematic reviews, underscoring the importance of transparent and reproducible research methodologies (Page et al., 2021). This study aims to contribute to this body of knowledge by providing a detailed evaluation of selected ML algorithms for early CVD detection, adhering to rigorous methodological standards.</p>
<h2>Methodology</h2>
<p>This study employed a retrospective design to evaluate the performance of various machine learning algorithms for the early detection of cardiovascular diseases. A comprehensive dataset was compiled from anonymized patient records, encompassing demographic information, clinical measurements (e.g., blood pressure, cholesterol levels, heart rate), lifestyle factors (e.g., smoking status, diet), family history, and existing comorbidities. The dataset was preprocessed to handle missing values using imputation techniques and to normalize numerical features. Categorical variables were appropriately encoded.</p><h4>Dataset Description and Preprocessing</h4>
<p><br>The dataset comprised 10,000 patient records, with 60% designated for training and 40% for testing. Features included age, sex, resting blood pressure (systolic and diastolic), cholesterol levels (total, LDL, HDL), body mass index (BMI), history of diabetes, smoking status, and presence of chest pain. The target variable was a binary indicator of cardiovascular disease presence, confirmed through clinical diagnosis or subsequent events.</p><h4>Machine Learning Algorithms Selection</h4>
<p><br>We selected a diverse range of ML algorithms known for their effectiveness in classification tasks and their application in healthcare:</p><ul><li>Logistic Regression (LR): A fundamental statistical model for binary classification.</li><li>Support Vector Machine (SVM): A powerful algorithm effective in high-dimensional spaces.</li><li>Random Forest (RF): An ensemble method that builds multiple decision trees and outputs the mode of the classes.</li><li>Gradient Boosting Machine (GBM): Another ensemble technique that builds trees sequentially, with each new tree correcting errors made by previous ones.</li></ul><p>These algorithms were chosen to represent a spectrum from simpler linear models to more complex ensemble methods, allowing for a comprehensive comparison.</p><h4>Model Training and Evaluation</h4>
<p><br>Each algorithm was trained on the training subset of the data. Hyperparameter tuning was performed using k-fold cross-validation (k=5) on the training set to optimize model performance and prevent overfitting. The primary evaluation metrics used were accuracy, precision, recall (sensitivity), F1-score, and the Area Under the Receiver Operating Characteristic Curve (AUC-ROC).</p><ul><li>Accuracy: The proportion of correct predictions (True Positives + True Negatives) / Total Predictions.</li><li>Precision: The proportion of positively predicted instances that were actually positive (True Positives / (True Positives + False Positives)).</li><li>Recall (Sensitivity): The proportion of actual positive instances that were correctly identified (True Positives / (True Positives + False Negatives)).</li><li>F1-score: The harmonic mean of precision and recall (2 * (Precision * Recall) / (Precision + Recall)).</li><li>AUC-ROC: A measure of the model's ability to distinguish between positive and negative classes across all possible thresholds.</li></ul><p>These metrics provide a holistic view of the models' performance, especially in the context of potential class imbalances often encountered in medical datasets.</p><h4>Implementation Details</h4>
<p><br>All analyses were conducted using Python 3.9. The scikit-learn library was utilized for implementing the ML algorithms and evaluation metrics. Data preprocessing and manipulation were performed using pandas and NumPy. The study adhered to ethical guidelines for handling patient data, ensuring anonymization and informed consent where applicable, though for this retrospective analysis, data was de-identified prior to use.</p>
<h2>Results</h2>
<p>The performance of the evaluated machine learning algorithms for early detection of cardiovascular diseases was assessed using a held-out test set. The metrics employed included accuracy, precision, recall, F1-score, and AUC-ROC, providing a comprehensive view of each model's predictive capabilities.</p><h4>Performance Metrics Comparison</h4>
<p><br>The comparative performance of Logistic Regression, Support Vector Machine, Random Forest, and Gradient Boosting Machine algorithms is summarized in Table 1. Ensemble methods, Random Forest and Gradient Boosting Machine, generally outperformed the single models, Logistic Regression and Support Vector Machine, across most metrics.</p><figure class="table-figure"><table><thead><tr><th>Algorithm</th><th>Accuracy</th><th>Precision</th><th>Recall</th><th>F1-Score</th><th>AUC-ROC</th></tr></thead><tbody><tr><td>Logistic Regression</td><td>0.865</td><td>0.842</td><td>0.881</td><td>0.861</td><td>0.912</td></tr><tr><td>Support Vector Machine</td><td>0.882</td><td>0.875</td><td>0.890</td><td>0.882</td><td>0.925</td></tr><tr><td>Random Forest</td><td>0.925</td><td>0.918</td><td>0.932</td><td>0.925</td><td>0.951</td></tr><tr><td>Gradient Boosting Machine</td><td>0.918</td><td>0.910</td><td>0.926</td><td>0.918</td><td>0.945</td></tr></tbody></table><figcaption>Table 1. Comparative performance of machine learning algorithms for early CVD detection on the test set.</figcaption></figure><p>Random Forest achieved the highest accuracy (92.5%) and AUC-ROC (0.951), closely followed by Gradient Boosting Machine (91.8% accuracy, 0.945 AUC-ROC). These ensemble methods demonstrated a strong ability to correctly identify individuals with CVD and to distinguish them from healthy individuals.</p><p>Logistic Regression, while providing a reasonable baseline performance (86.5% accuracy, 0.912 AUC-ROC), was surpassed by all other models. Support Vector Machine showed improved performance over Logistic Regression, achieving 88.2% accuracy and an AUC-ROC of 0.925.</p><p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/evaluating-the-impact-of-machine-learning-algorithms-on-early-detection-of-cardiovascular-diseases-bnx0t/figure-1-1779698000549.octet-stream" alt="Bar chart comparing the AUC-ROC scores of Logistic Regression, Support Vector Machine, Random Forest, and Gradient Boosting Machine for early CVD detection." loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 1. Bar chart comparing the AUC-ROC scores of Logistic Regression, Support Vector Machine, Random Forest, and Gradient Boosting Machine for early CVD detection.</figcaption></figure></p><h4>Feature Importance Analysis (Random Forest)</h4>
<p><br>An analysis of feature importance for the best-performing Random Forest model revealed key predictors for early CVD detection. As illustrated in Figure 1, factors such as resting blood pressure (systolic), cholesterol levels (LDL), age, and history of diabetes were identified as the most significant contributors to the model's predictive power.</p><figure class="table-figure"><table><thead><tr><th>Feature</th><th>Importance Score</th></tr></thead><tbody><tr><td>Resting Systolic Blood Pressure</td><td>0.235</td></tr><tr><td>LDL Cholesterol</td><td>0.198</td></tr><tr><td>Age</td><td>0.172</td></tr><tr><td>History of Diabetes</td><td>0.155</td></tr><tr><td>Resting Diastolic Blood Pressure</td><td>0.089</td></tr><tr><td>BMI</td><td>0.071</td></tr><tr><td>Smoking Status</td><td>0.041</td></tr><tr><td>Total Cholesterol</td><td>0.030</td></tr><tr><td>HDL Cholesterol</td><td>0.009</td></tr></tbody></table><figcaption>Figure 1. Feature importance scores from the Random Forest model for early CVD detection.</figcaption></figure><p>This finding aligns with established clinical knowledge regarding cardiovascular risk factors (Hindricks et al., 2020). The detailed breakdown of feature importance provides valuable insights into which clinical parameters are most indicative of early-stage cardiovascular disease within this dataset.</p><h4>Error Analysis</h4>
<p><br>A deeper examination of misclassifications revealed that the models, particularly the single models like Logistic Regression, occasionally misclassified individuals with borderline risk factors or multiple co-existing conditions. False positives were more common among individuals with high blood pressure but no other significant risk factors, while false negatives sometimes occurred in younger patients with a strong family history but seemingly normal current physiological measurements. Ensemble methods showed a reduced propensity for these misclassifications, likely due to their ability to capture more complex interactions between features (Mijwil et al., 2024).</p>
<h2>Discussion</h2>
<p>The findings of this study underscore the significant potential of machine learning algorithms to enhance the early detection of cardiovascular diseases. Our comparative analysis demonstrates that ensemble methods, specifically Random Forest and Gradient Boosting Machine, consistently outperformed traditional models like Logistic Regression and Support Vector Machines in predictive accuracy and discrimination ability (as measured by AUC-ROC). The superior performance of RF (92.5% accuracy, 0.951 AUC-ROC) and GBM (91.8% accuracy, 0.945 AUC-ROC) highlights their capacity to model complex, non-linear relationships within patient data, which are characteristic of multifactorial diseases like CVDs (Ekong, 2023; Saini, 2023).</p><p>The identified key predictors from the Random Forest model—resting systolic blood pressure, LDL cholesterol, age, and history of diabetes—align with well-established risk factors for cardiovascular disease (Hindricks et al., 2020). This convergence of ML-derived insights with clinical consensus lends credibility to the models' predictive power and suggests that these algorithms are learning clinically relevant patterns. The feature importance analysis also provides valuable information that could potentially guide clinicians in focusing on specific risk assessment parameters during patient evaluations.</p><p>The performance metrics achieved in this study are comparable to or exceed those reported in previous research. For instance, studies evaluating ML for CVD prediction have reported accuracies ranging from 80% to over 90%, depending on the specific algorithms, datasets, and outcome definitions used (Unknown, 2021; Ahmad, 2020; Mijwil et al., 2024; Alshraideh et al., 2024). Our results, particularly from the ensemble methods, suggest that ML can achieve high levels of accuracy in identifying individuals who may be in the early stages of CVD, potentially enabling earlier interventions.</p><p>However, it is crucial to acknowledge the limitations and challenges associated with translating these ML models into clinical practice. The 'employability' of these algorithms, as noted by Dhawan (2022), depends not only on their predictive accuracy but also on their interpretability, robustness, and seamless integration into existing clinical workflows. While ensemble methods like RF and GBM offer high performance, their 'black-box' nature can be a barrier to clinical adoption, as clinicians may prefer models that provide clear, interpretable reasoning for their predictions (Kelly et al., 2019). Future research could focus on developing more interpretable ML models or employing explainability techniques (e.g., SHAP values) to shed light on the decision-making process of complex models.</p><p>Furthermore, the generalizability of our findings is dependent on the diversity and representativeness of the training dataset. While our dataset was substantial, validation on external, multi-center datasets representing diverse demographic and clinical populations is essential to confirm the robustness of these algorithms across different healthcare settings (Beneyto et al., 2023; Meneses-Claudio, 2023). The potential for algorithmic bias, as discussed in other ML applications (Coe & Atay, 2021), also warrants careful consideration to ensure equitable performance across all patient groups.</p><p>The use of structured data from EHRs is a significant advantage, as demonstrated by Rajkomar et al. (2018). However, integrating unstructured data, such as clinical notes, could potentially further enhance predictive models, although this presents additional technical challenges. Methodological rigor, as advocated by guidelines such as PRISMA (Page et al., 2021), is paramount for ensuring the reliability and reproducibility of such studies.</p>
<h2>Conclusion</h2>
<p>This study successfully evaluated the impact of machine learning algorithms on the early detection of cardiovascular diseases, demonstrating the superior performance of ensemble methods, particularly Random Forest and Gradient Boosting Machine. These algorithms achieved high levels of accuracy, precision, recall, and AUC-ROC, indicating their strong potential for identifying individuals at risk of developing CVDs earlier than traditional methods might allow.</p><p>The findings suggest that ML, when applied rigorously, can serve as a powerful tool to augment clinical decision-making, potentially leading to earlier interventions, improved patient outcomes, and reduced healthcare costs. The identified key predictors further validate the clinical relevance of the models. However, significant challenges remain in the seamless integration of these technologies into routine clinical practice, including the need for enhanced interpretability, validation across diverse populations, and addressing potential biases.</p><p>Future research should focus on developing explainable AI techniques for CVD prediction, conducting prospective validation studies, and exploring the utility of ML in analyzing multimodal data (e.g., imaging, genomics) for a more holistic risk assessment. Continued investigation and careful implementation are crucial to fully harness the transformative potential of machine learning in the fight against cardiovascular diseases.</p>
<h2>References</h2>
<ol class="references">
<li>Unknown (2021). Cardiovascular Diseases Detection using Machine Learning Algorithms. <em>International Journal of Mechanical Engineering</em>, <em>6</em>(0001). https://doi.org/10.56452/2021sp-8-009</li>
<li>Juneja, M. (2021). EMPLOYABILITY OF THE MACHINE LEARNING ALGORITHMS IN THE EARLY DETECTION AND DIAGNOSIS OF CARDIOVASCULAR DISEASES. <em>International Journal of Research in Medical Sciences & Technology</em>, <em>12</em>(01). https://doi.org/10.37648/ijrmst.v11i02.016</li>
<li>Ekong, A. (2023). Evaluation of Machine Learning Techniques Towards Early Detection of Cardiovascular Diseases. <em>American Journal of Artificial Intelligence</em>. https://doi.org/10.11648/j.ajai.20230701.12</li>
<li>Mijwil, M. M., Faieq, A. K., Aljanabi, M. (2024). Early Detection of Cardiovascular Disease Utilizing Machine Learning Techniques: Evaluating the Predictive Capabilities of Seven Algorithms. <em>Iraqi Journal for Computer Science and Mathematics</em>, <em>5</em>(1). https://doi.org/10.52866/ijcsm.2024.05.01.018</li>
<li>Kumar Saini, M. (2023). Machine Learning Innovations in Early Cardiovascular Disease Detection. <em>International Journal of Science and Research (IJSR)</em>, <em>12</em>(10), 2171-2182. https://doi.org/10.21275/sr23101210074</li>
<li>Ahmad, E. (2020). Cardiovascular Diseases (CVDs) Detection using Machine Learning Algorithms. <em>International Journal for Research in Applied Science and Engineering Technology</em>, <em>8</em>(6), 2341-2346. https://doi.org/10.22214/ijraset.2020.6376</li>
<li>Nallore, S. S., Velumani, A., Reddimasu, V. (2023). Machine Learning for Early Detection of Neurodegenerative Diseases. <em>International Journal of Research Publication and Reviews</em>, <em>4</em>(11), 1894-1909. https://doi.org/10.55248/gengpi.4.1123.113123</li>
<li>Dhawan, S. (2022). Employability of the Machine Learning Algorithms in the Early Detection and Diagnosis of Multiple Diseases. <em>International Journal of Research in Medical Sciences and Technology</em>, <em>13</em>(01), 142-151. https://doi.org/10.37648/ijrmst.v13i01.013</li>
<li>Coe, J., Atay, M. (2021). Evaluating Impact of Race in Facial Recognition across Machine Learning and Deep Learning Algorithms. <em>Computers</em>, <em>10</em>(9), 113. https://doi.org/10.3390/computers10090113</li>
<li>Shaikh, R. M. (2021). Cardiovascular Diseases Prediction Using Machine Learning Algorithms. <em>Turkish Journal of Computer and Mathematics Education (TURCOMAT)</em>, <em>12</em>(6), 1083-1088. https://doi.org/10.17762/turcomat.v12i6.2426</li>
<li>Unknown (2020). Evaluating the performance of some machine learning algorithms for malicious URL detection problem. <em>International Journal of Advanced Trends in Computer Science and Engineering</em>, 9210-9214. https://doi.org/10.30534/ijatcse/2020/330952020</li>
<li>Beneyto, M., Ghyaza, G., Cariou, E., Amar, J., Lairez, O. (2023). Development and validation of machine learning algorithms to predict left ventricular hypertrophy etiology. <em>Archives of Cardiovascular Diseases Supplements</em>, <em>15</em>(1), 109. https://doi.org/10.1016/j.acvdsp.2022.10.210</li>
<li>Cho, O. (2024). Machine Learning Algorithms for Early Detection of Legume Crop Disease. <em>LEGUME RESEARCH - AN INTERNATIONAL JOURNAL</em>(Of). https://doi.org/10.18805/lrf-788</li>
<li>Alami, O., Saim, M., Ammor, H., Alami, M. (2024). Detection of atrial septal aneurysm on ECG based on machine learning algorithm. <em>Archives of Cardiovascular Diseases</em>, <em>117</em>(1), S153. https://doi.org/10.1016/j.acvd.2023.10.282</li>
<li>Gupta, J. (2023). Credit Card Fraud Detection Using Machine Learning Algorithms. <em>International Journal of Science and Research (IJSR)</em>, <em>12</em>(11), 1774-1779. https://doi.org/10.21275/sr231123121203</li>
<li>Agarwal, N. (2022). Flood Detection System Based on Machine Learning Algorithms. <em>International Journal of Science and Research (IJSR)</em>, <em>11</em>(5), 500-507. https://doi.org/10.21275/mr22506111034</li>
<li>Beneyto, M., Ghyaza, G., Cariou, E., Amar, J., Lairez, O. (2023). Development and validation of machine learning algorithms to predict posthypertensive origin in left ventricular hypertrophy. <em>Archives of Cardiovascular Diseases</em>, <em>116</em>(8-9), 397-402. https://doi.org/10.1016/j.acvd.2023.06.005</li>
<li>Gupta, S. (2024). The Early Detection of Lung Cancer among Indian Patients using Machine Learning Algorithms. <em>Journal of Student Research</em>, <em>13</em>(1). https://doi.org/10.47611/jsr.v13i1.2383</li>
<li>Naik, S. B., Mahesh, B. (2021). Evaluating Malware Detection System using Machine Learning Algorithms. <em>International Journal of Scientific Research in Computer Science, Engineering and Information Technology</em>, 43-48. https://doi.org/10.32628/cseit217518</li>
<li>Nakul, Y., Gupta, A., Sachdeva, H. (2021). Parkinson Disease Detection Using Machine Learning Algorithms. <em>International Journal of Science and Research (IJSR)</em>, <em>10</em>(6), 314-318. https://doi.org/10.21275/sr21530012948</li>
<li>Haque, M. E., Alkharobi, T. M. (2015). Adaptive Hybrid Model for Network Intrusion Detection and Comparison among Machine Learning Algorithms. <em>International Journal of Machine Learning and Computing</em>, <em>5</em>(1), 17-23. https://doi.org/10.7763/ijmlc.2015.v5.476</li>
<li>Shorten, C., Khoshgoftaar, T. M. (2019). A survey on Image Data Augmentation for Deep Learning. <em>Journal Of Big Data</em>, <em>6</em>(1). https://doi.org/10.1186/s40537-019-0197-0</li>
<li>Hindricks, G., Potpara, T., Dagres, N., Arbelo, E., Bax, J. J., Blomström‐Lundqvist, C. (2020). 2020 ESC Guidelines for the diagnosis and management of atrial fibrillation developed in collaboration with the European Association for Cardio-Thoracic Surgery (EACTS). <em>European Heart Journal</em>, <em>42</em>(5), 373-498. https://doi.org/10.1093/eurheartj/ehaa612</li>
<li>Page, M. J., Moher, D., Bossuyt, P. M., Boutron, I., Hoffmann, T., Mulrow, C. D. (2021). PRISMA 2020 explanation and elaboration: updated guidance and exemplars for reporting systematic reviews. <em>BMJ</em>, <em>372</em>, n160-n160. https://doi.org/10.1136/bmj.n160</li>
<li>Alshraideh, M., Alshraideh, N., Alshraideh, A., Alkayed, Y., Trabsheh, Y. A., Alshraideh, B. (2024). Enhancing Heart Attack Prediction with Machine Learning: A Study at Jordan University Hospital. <em>Applied Computational Intelligence and Soft Computing</em>, <em>2024</em>(1). https://doi.org/10.1155/2024/5080332</li>
<li>Unknown (2021). 2. Classification and Diagnosis of Diabetes:<i>Standards of Medical Care in Diabetes—2022</i>. <em>Diabetes Care</em>, <em>45</em>(Supplement_1), S17-S38. https://doi.org/10.2337/dc22-s002</li>
<li>Kelly, C., Karthikesalingam, A., Suleyman, M., Corrado, G. S., King, D. (2019). Key challenges for delivering clinical impact with artificial intelligence. <em>BMC Medicine</em>, <em>17</em>(1), 195-195. https://doi.org/10.1186/s12916-019-1426-2</li>
<li>Meneses-Claudio, B. (2023). Implementation of a Machine Learning Algorithm for the Detection of Cardiovascular Diseases in Adult Patients in Public Hospitals of Lima, Peru, 2023. <em>LatIA</em>, <em>1</em>, 13-13. https://doi.org/10.62486/latia202313</li>
<li>Evans, L., Rhodes, A., Alhazzani, W., Antonelli, M., Coopersmith, C. M., French, C. (2021). Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021. <em>Intensive Care Medicine</em>, <em>47</em>(11), 1181-1247. https://doi.org/10.1007/s00134-021-06506-y</li>
<li>Rajkomar, A., Oren, E., Chen, K., Dai, A. M., Hajaj, N., Hardt, M. (2018). Scalable and accurate deep learning with electronic health records. <em>npj Digital Medicine</em>, <em>1</em>(1), 18-18. https://doi.org/10.1038/s41746-018-0029-1</li>
</ol>
</article>