Use of Artificial Intelligence in Improving Outcomes in Heart Disease: A Scientific Statement From the American Heart Association.

A major focus of academia, industry, and global governmental agencies is to develop and apply artificial intelligence and other advanced analytical tools to transform health care delivery. The American Heart Association supports the creation of tools and services that would further the science and practice of precision medicine by enabling more precise approaches to cardiovascular and stroke research, prevention, and care of individuals and populations. Nevertheless, several challenges exist, and few artificial intelligence tools have been shown to improve cardiovascular and stroke care sufficiently to be widely adopted. This scientific statement outlines the current state of the art on the use of artificial intelligence algorithms and data science in the diagnosis, classification, and treatment of cardiovascular disease. It also sets out to advance this mission, focusing on how digital tools and, in particular, artificial intelligence may provide clinical and mechanistic insights, address bias in clinical studies, and facilitate education and implementation science to improve cardiovascular and stroke outcomes. Last, a key objective of this scientific statement is to further the field by identifying best practices, gaps, and challenges for interested stakeholders.

Ventricular tachycardia and in-hospital mortality in the intensive care unit.

Background: Continuous electrocardiographic (ECG) monitoring is used to identify ventricular tachycardia (VT), but false alarms occur frequently. Objective: The purpose of this study was to assess the rate of 30-day in-hospital mortality associated with VT alerts generated from bedside ECG monitors to those from a new algorithm among intensive care unit (ICU) patients. Methods: We conducted a retrospective cohort study in consecutive adult ICU patients at an urban academic medical center and compared current bedside monitor VT alerts, VT alerts from a new-unannotated algorithm, and true-annotated VT. We used survival analysis to explore the association between VT alerts and mortality. Results: We included 5679 ICU admissions (mean age 58 ± 17 years; 48% women), 503 (8.9%) experienced 30-day in-hospital mortality. A total of 30.1% had at least 1 current bedside monitor VT alert, 14.3% had a new-unannotated algorithm VT alert, and 11.6% had true-annotated VT. Bedside monitor VT alert was not associated with increased rate of 30-day mortality (adjusted hazard ratio [aHR] 1.06; 95% confidence interval [CI] 0.88-1.27), but there was an association for VT alerts from our new-unannotated algorithm (aHR 1.38; 95% CI 1.12-1.69) and true-annotated VT(aHR 1.39; 95% CI 1.12-1.73). Conclusion: Unannotated and annotated-true VT were associated with increased rate of 30-day in-hospital mortality, whereas current bedside monitor VT was not. Our new algorithm may accurately identify high-risk VT; however, prospective validation is needed.

Engaging Multidisciplinary Clinical Users in the Design of an Artificial Intelligence-Powered Graphical User Interface for Intensive Care Unit Instability Decision Support.

BACKGROUND: Critical instability forecast and treatment can be optimized by artificial intelligence (AI)-enabled clinical decision support. It is important that the user-facing display of AI output facilitates clinical thinking and workflow for all disciplines involved in bedside care. OBJECTIVES: Our objective is to engage multidisciplinary users (physicians, nurse practitioners, physician assistants) in the development of a graphical user interface (GUI) to present an AI-derived risk score. METHODS: Intensive care unit (ICU) clinicians participated in focus groups seeking input on instability risk forecast presented in a prototype GUI. Two stratified rounds (three focus groups [only nurses, only providers, then combined]) were moderated by a focus group methodologist. After round 1, GUI design changes were made and presented in round 2. Focus groups were recorded, transcribed, and deidentified transcripts independently coded by three researchers. Codes were coalesced into emerging themes. RESULTS: Twenty-three ICU clinicians participated (11 nurses, 12 medical providers [3 mid-level and 9 physicians]). Six themes emerged: (1) analytics transparency, (2) graphical interpretability, (3) impact on practice, (4) value of trend synthesis of dynamic patient data, (5) decisional weight (weighing AI output during decision-making), and (6) display location (usability, concerns for patient/family GUI view). Nurses emphasized having GUI objective information to support communication and optimal GUI location. While providers emphasized need for recommendation interpretability and concern for impairing trainee critical thinking. All disciplines valued synthesized views of vital signs, interventions, and risk trends but were skeptical of placing decisional weight on AI output until proven trustworthy. CONCLUSION: Gaining input from all clinical users is important to consider when designing AI-derived GUIs. Results highlight that health care intelligent decisional support systems technologies need to be transparent on how they work, easy to read and interpret, cause little disruption to current workflow, as well as decisional support components need to be used as an adjunct to human decision-making.

Forecasting imminent atrial fibrillation in long-term electrocardiogram recordings.

BACKGROUND: Despite the morbidity associated with acute atrial fibrillation (AF), no models currently exist to forecast its imminent onset. We sought to evaluate the ability of deep learning to forecast the imminent onset of AF with sufficient lead time, which has important implications for inpatient care. METHODS: We utilized the Physiobank Long-Term AF Database, which contains 24-h, labeled ECG recordings from patients with a history of AF. AF episodes were defined as ≥5 min of sustained AF. Three deep learning models incorporating convolutional and transformer layers were created for forecasting, with two models focusing on the predictive nature of sinus rhythm segments and AF epochs separately preceding an AF episode, and one model utilizing all preceding waveform as input. Cross-validated performance was evaluated using area under time-dependent receiver operating characteristic curves (AUC(t)) at 7.5-, 15-, 30-, and 60-min lead times, precision-recall curves, and imminent AF risk trajectories. RESULTS: There were 367 AF episodes from 84 ECG recordings. All models showed average risk trajectory divergence of those with an AF episode from those without ∼15 min before the episode. Highest AUC was associated with the sinus rhythm model [AUC = 0.74; 7.5-min lead time], though the model using all preceding waveform data had similar performance and higher AUCs at longer lead times. CONCLUSIONS: In this proof-of-concept study, we demonstrated the potential utility of neural networks to forecast the onset of AF in long-term ECG recordings with a clinically relevant lead time. External validation in larger cohorts is required before deploying these models clinically.

Machine learning for ECG diagnosis and risk stratification of occlusion myocardial infarction.

Patients with occlusion myocardial infarction (OMI) and no ST-elevation on presenting electrocardiogram (ECG) are increasing in numbers. These patients have a poor prognosis and would benefit from immediate reperfusion therapy, but, currently, there are no accurate tools to identify them during initial triage. Here we report, to our knowledge, the first observational cohort study to develop machine learning models for the ECG diagnosis of OMI. Using 7,313 consecutive patients from multiple clinical sites, we derived and externally validated an intelligent model that outperformed practicing clinicians and other widely used commercial interpretation systems, substantially boosting both precision and sensitivity. Our derived OMI risk score provided enhanced rule-in and rule-out accuracy relevant to routine care, and, when combined with the clinical judgment of trained emergency personnel, it helped correctly reclassify one in three patients with chest pain. ECG features driving our models were validated by clinical experts, providing plausible mechanistic links to myocardial injury.

Unsupervised machine learning identifies symptoms of indigestion as a predictor of acute decompensation and adverse cardiac events in patients with heart failure presenting to the emergency department.

BACKGROUND: Patients with known heart failure (HF) present to emergency departments (ED) with a plethora of symptoms. Although symptom clusters have been suggested as prognostic features, accurately triaging HF patients is a longstanding challenge. OBJECTIVES: We sought to use machine learning to identify subtle phenotypes of patient symptoms and evaluate their diagnostic and prognostic value among HF patients seeking emergency care. METHODS: This was a secondary analysis of a prospective cohort study of consecutive patients seen in the ED for chest pain or equivalent symptoms. Independent reviewers extracted clinical data from charts, including nine categories of subjective symptoms reported during initial evaluation. The diagnostic outcome was acute HF exacerbation and prognostic outcome was 30-day major adverse cardiac events (MACE). Outcomes were adjudicated by two independent reviewers. K-means clustering was used to derive latent patient symptom clusters, and their associations with outcomes were assessed using multivariate logistic regression. RESULTS: Sample included 438 patients (age 65±14 years; 45% female, 49% Black, 18% HF exacerbation, 32% MACE). K-means clustering identified three presentation phenotypes: patients with dyspnea only (Cluster A, 40%); patients with indigestion, with or without dyspnea (Cluster B, 23%); patients with neither dyspnea nor indigestion (Cluster C, 37%). Compared to Cluster C, indigestion was a significant predictor of acute HF exacerbation (OR=1.8, 95%CI=1.0-3.4) and 30-day MACE (OR=1.8, 95%CI=1.0-3.1), independent of age, sex, race, and other comorbidities. CONCLUSION: Indigestion symptoms in patients with known HF signify excess risk of adverse events, suggesting that these patients should be triaged as high-risk during initial ED evaluation.

Integrating multimodal information in machine learning for classifying acute myocardial infarction.

Objective. Prompt identification and recognization of myocardial ischemia/infarction (MI) is the most important goal in the management of acute coronary syndrome. The 12-lead electrocardiogram (ECG) is widely used as the initial screening tool for patients with chest pain but its diagnostic accuracy remains limited. There is early evidence that machine learning (ML) algorithms applied to ECG waveforms can improve performance. Most studies are designed to classify MI from healthy controls and thus are limited due to the lack of consideration of ECG abnormalities from other cardiac conditions, leading to false positives. Moreover, clinical information beyond ECG has not yet been well leveraged in existing ML models.Approach.The present study considered downstream clinical implementation scenarios in the initial model design by dichotomizing study recordings from a public large-scale ECG dataset into a MI class and a non-MI class with the inclusion of MI-confounding conditions. Two experiments were conducted to systematically investigate the impact of two important factors entrained in the modeling process, including the duration of ECG, and the value of multimodal information for model training. A novel multimodal deep learning architecture was proposed to learn joint features from both ECG and patient demographics.Main results.The multimodal model achieved better performance than the ECG-only model, with a mean area under the receiver operating characteristic curve of 92.1% and a mean accuracy of 87.4%, which is on par with existing studies despite the increased task difficulty due to the new class definition. By investigation of model explainability, it revealed the contribution of patient information in model performance and clinical concordance of the model's attention with existing clinical insights.Significance.The findings in this study help guide the development of ML solutions for prompt MI detection and move the models one step closer to real-world clinical applications.

An annotated ventricular tachycardia (VT) alarm database: Toward a uniform standard for optimizing automated VT identification in hospitalized patients.

BACKGROUND: False ventricular tachycardia (VT) alarms are common during in-hospital electrocardiographic (ECG) monitoring. Prior research shows that the majority of false VT can be attributed to algorithm deficiencies. PURPOSE: The purpose of this study was: (1) to describe the creation of a VT database annotated by ECG experts and (2) to determine true vs. false VT using a new VT algorithm created by our group. METHODS: The VT algorithm was processed in 5320 consecutive ICU patients with 572,574 h of ECG and physiologic monitoring. A search algorithm identified potential VT, defined as: heart rate >100 beats/min, QRSs > 120 ms, and change in QRS morphology in >6 consecutive beats compared to the preceding native rhythm. Seven ECG channels, SpO2 , and arterial blood pressure waveforms were processed and loaded into a web-based annotation software program. Five PhD-prepared nurse scientists performed the annotations. RESULTS: Of the 5320 ICU patients, 858 (16.13%) had 22,325 VTs. After three levels of iterative annotations, a total of 11,970 (53.62%) were adjudicated as true, 6485 (29.05%) as false, and 3870 (17.33%) were unresolved. The unresolved VTs were concentrated in 17 patients (1.98%). Of the 3870 unresolved VTs, 85.7% (n = 3281) were confounded by ventricular paced rhythm, 10.8% (n = 414) by underlying BBB, and 3.5% (n = 133) had a combination of both. CONCLUSIONS: The database described here represents the single largest human-annotated database to date. The database includes consecutive ICU patients, with true, false, and challenging VTs (unresolved) and could serve as a gold standard database to develop and test new VT algorithms.

Machine Learning for the ECG Diagnosis and Risk Stratification of Occlusion Myocardial Infarction at First Medical Contact.

Patients with occlusion myocardial infarction (OMI) and no ST-elevation on presenting ECG are increasing in numbers. These patients have a poor prognosis and would benefit from immediate reperfusion therapy, but we currently have no accurate tools to identify them during initial triage. Herein, we report the first observational cohort study to develop machine learning models for the ECG diagnosis of OMI. Using 7,313 consecutive patients from multiple clinical sites, we derived and externally validated an intelligent model that outperformed practicing clinicians and other widely used commercial interpretation systems, significantly boosting both precision and sensitivity. Our derived OMI risk score provided superior rule-in and rule-out accuracy compared to routine care, and when combined with the clinical judgment of trained emergency personnel, this score helped correctly reclassify one in three patients with chest pain. ECG features driving our models were validated by clinical experts, providing plausible mechanistic links to myocardial injury.

Incorporation of Serial 12-Lead Electrocardiogram With Machine Learning to Augment the Out-of-Hospital Diagnosis of Non-ST Elevation Acute Coronary Syndrome.

STUDY OBJECTIVE: Ischemic electrocardiogram (ECG) changes are subtle and transient in patients with suspected non-ST-segment elevation (NSTE)-acute coronary syndrome. However, the out-of-hospital ECG is not routinely used during subsequent evaluation at the emergency department. Therefore, we sought to compare the diagnostic performance of out-of-hospital and ED ECG and evaluate the incremental gain of artificial intelligence-augmented ECG analysis. METHODS: This prospective observational cohort study recruited patients with out-of-hospital chest pain. We retrieved out-of-hospital-ECG obtained by paramedics in the field and the first ED ECG obtained by nurses during inhospital evaluation. Two independent and blinded reviewers interpreted ECG dyads in mixed order per practice recommendations. Using 179 morphological ECG features, we trained, cross-validated, and tested a random forest classifier to augment non ST-elevation acute coronary syndrome (NSTE-ACS) diagnosis. RESULTS: Our sample included 2,122 patients (age 59 [16]; 53% women; 44% Black, 13.5% confirmed acute coronary syndrome). The rate of diagnostic ST elevation and ST depression were 5.9% and 16.2% on out-of-hospital-ECG and 6.1% and 12.4% on ED ECG, with ∼40% of changes seen on out-of-hospital-ECG persisting and ∼60% resolving. Using expert interpretation of out-of-hospital-ECG alone gave poor baseline performance with area under the receiver operating characteristic (AUC), sensitivity, and negative predictive values of 0.69, 0.50, and 0.92. Using expert interpretation of serial ECG changes enhanced this performance (AUC 0.80, sensitivity 0.61, and specificity 0.93). Interestingly, augmenting the out-of-hospital-ECG alone with artificial intelligence algorithms boosted its performance (AUC 0.83, sensitivity 0.75, and specificity 0.95), yielding a net reclassification improvement of 29.5% against expert ECG interpretation. CONCLUSION: In this study, 60% of diagnostic ST changes resolved prior to hospital arrival, making the ED ECG suboptimal for the inhospital evaluation of NSTE-ACS. Using serial ECG changes or incorporating artificial intelligence-augmented analyses would allow correctly reclassifying one in 4 patients with suspected NSTE-ACS.

Emerging ECG methods for acute coronary syndrome detection: Recommendations & future opportunities.

Despite being the mainstay for the initial noninvasive assessment of patients with symptomatic coronary artery disease, the 12­lead ECG remains a suboptimal diagnostic tool for myocardial ischemia detection with only acceptable sensitivity and specificity scores. Although myocardial ischemia affects the configuration of the QRS complex and the STT waveform, current guidelines primarily focus on ST segment amplitude, which constitutes a missed opportunity and may explain the suboptimal diagnostic performance of the ECG. This possible opportunity and the low cost and ease of use of the ECG provide compelling motivation to enhance the diagnostic accuracy of the ECG to ischemia detection. This paper describes numerous computational ECG methods and approaches that have been shown to dramatically increase ECG sensitivity to ischemia detection. Briefly, these emerging approaches can be conceptually grouped into one of the following four approaches: (1) leveraging novel ECG waveform features and signatures indicative of ischemic injury other than the classical ST-T amplitude measures; (2) applying body surface potentials mapping (BSPM)-based approaches to enhance the spatial coverage of the surface ECG to detecting ischemia; (3) developing an inverse ECG solution to reconstruct anatomical models of activation and recovery pathways to detect and localize injury currents; and (4) exploring artificial intelligence (AI)-based techniques to harvest ECG waveform signatures of ischemia. We present recent advances, shortcomings, and future opportunities for each of these emerging ECG methods. Future research should focus on the prospective clinical testing of these approaches to establish clinical utility and to expedite potential translation into clinical practice.

Exploring decision making 'noise' when interpreting the electrocardiogram in the context of cardiac cath lab activation.

In this commentary paper, we discuss the use of the electrocardiogram to help clinicians make diagnostic and patient referral decisions in acute care settings. The paper discusses the factors that are likely to contribute to the variability and noise in the clinical decision making process for catheterization lab activation. These factors include the variable competence in reading ECGs, the intra/inter rater reliability, the lack of standard ECG training, the various ECG machine and filter settings, cognitive biases (such as automation bias which is the tendency to agree with the computer-aided diagnosis or AI diagnosis), the order of the information being received, tiredness or decision fatigue as well as ECG artefacts such as the signal noise or lead misplacement. We also discuss potential research questions and tools that could be used to mitigate this 'noise' and improve the quality of ECG based decision making.

Remote and wearable ECG devices with diagnostic abilities in adults: A state-of-the-science scoping review.

The electrocardiogram (ECG) records the electrical activity in the heart in real time, providing an important opportunity to detecting various cardiac pathologies. The 12-lead ECG currently serves as the "standard" ECG acquisition technique for diagnostic purposes for many cardiac pathologies other than arrhythmias. However, the technical aspects of acquiring a 12-lead ECG are not easy. and its usage currently is restricted to trained medical personnel, which limits the scope of its usefulness. Remote and wearable ECG devices have attempted to bridge this gap by enabling patients to take their own ECG using a simplified method at the expense of a reduced number of leads, usually a single-lead ECG. In this review, we summarize the studies that investigated the use of remote ECG devices and their clinical utility in diagnosing cardiac pathologies. Eligible studies discussed Food and Drug Administration-cleared, commercially available devices that were validated in an adult population. We summarize technical logistics of signal quality and device reliability, dimensional and functional features, and diagnostic value. Our synthesis shows that reduced-set ECG wearables have huge potential for long-term monitoring, particularly if paired with real-time notification techniques. Such capabilities make them primarily useful for abnormal rhythm detection, and there is sufficient evidence that a remote ECG device can be superior to the traditional 12-lead ECG in diagnosing specific arrhythmias such as atrial fibrillation. However, this review identifies important challenges faced by this technology and highlights the limited availability of clinical research examining their usefulness.