Friedman, Paul A.

Research ObjectiveAn artificial intelligence (AI) algorithm based on ECG was integrated into clinical workflows to identify patients with a high likelihood of low ejection (EF), a condition that is underdiagnosed but treatable. The trial aimed to assess whether the AI‐powered clinical decision support enabled early diagnosis of low EF.Study DesignA pragmatic, cluster randomized controlled trial (RCT) was conducted in 45 primary care practices, including an academic medical center as well as community and rural clinics across Minnesota and Wisconsin. A total of 120 primary care teams, including 358 clinicians, were randomized to either intervention (access to AI results with positive results indicating a high likelihood of low EF; 181 clinicians) or control (usual care; 177 clinicians). The primary endpoint was a new diagnosis of EF≤50% within 90 days. A positive deviance qualitative study and a survey were used to assess implementation strategies.Population StudiedIn total, 22,641 adult patients (N = 11,573 intervention; N = 11,068 control) with ECG performed as part of routine care between 8/5/2019 and 3/31/2020 were included. Patients were excluded if they had known EF≤50% or a history of documented heart failure (HF) identified by clinical notes abstracted by natural language processing.Principal FindingsThe mean age of patients was 60.5 ± 17.6 years; 53.9% were women. Positive AI results were found in 6.0% of patients in both arms. Clinicians in the intervention group obtained more echocardiograms for patients with positive results (38.1% control vs. 49.6% intervention, p < 0.001), but in the overall cohort, the echocardiogram utilization was similar (18.2% vs. 19.2%, p = 0.17). The intervention increased the diagnosis of low EF in the overall cohort (1.6% vs. 2.1%, odds ratio [OR] 1.32 [1.01–1.61], p = 0.007) and among patients with positive results (14.5% vs. 19.5% OR 1.43 [1.08–1.91], p = 0.01). Results using different thresholds for low EF were consistent with the primary result. Among patients with positive results, the intervention increased echocardiography more significantly in women vs. men (p for interaction 0.02), in patients without prior echocardiogram vs. those with prior echocardiogram (p for interaction 0.04), and in patients whose ECGs were ordered in outpatient clinics (p for interaction 0.02).ConclusionsAn AI algorithm run on existing ECGs enabled the early diagnosis of low EF in a large cohort of patients managed in routine primary care practices. Because ECG is a low‐cost test frequently performed for a variety of purposes, the algorithm could improve early diagnosis and treatment of a condition that is often asymptomatic but has effective treatments, and thus, reduce the disease burden in broad populations.Implications for Policy or PracticeThe current study is one of the first RCTs that demonstrated the value of AI in routine practice. Low EF confers an increased risk of HF progression and mortality. Although there are effective therapies, low EF is undiagnosed, leading to missed opportunities for treatment. EF can be readily determined by echocardiography, but the imagining test is too resource‐intensive to screen unselected populations. The AI applied to the existing ECGs could identify high‐risk patients for further assessment.Primary Funding SourceMayo Clinic.

Abstract Artificial intelligence (AI) has given the electrocardiogram (ECG) and clinicians reading them super-human diagnostic abilities. Trained without hard-coded rules by finding often subclinical patterns in huge datasets, AI transforms the ECG, a ubiquitous, non-invasive cardiac test that is integrated into practice workflows, into a screening tool and predictor of cardiac and non-cardiac diseases, often in asymptomatic individuals. This review describes the mathematical background behind supervised AI algorithms, and discusses selected AI ECG cardiac screening algorithms including those for the detection of left ventricular dysfunction, episodic atrial fibrillation from a tracing recorded during normal sinus rhythm, and other structural and valvular diseases. The ability to learn from big data sets, without the need to understand the biological mechanism, has created opportunities for detecting non-cardiac diseases as COVID-19 and introduced challenges with regards to data privacy. Like all medical tests, the AI ECG must be carefully vetted and validated in real-world clinical environments. Finally, with mobile form factors that allow acquisition of medical-grade ECGs from smartphones and wearables, the use of AI may enable massive scalability to democratize healthcare.

Abstract Aims Cardiovascular disease is a major threat to maternal health, with cardiomyopathy being among the most common acquired cardiovascular diseases during pregnancy and the postpartum period. The aim of our study was to evaluate the effectiveness of an electrocardiogram (ECG)-based deep learning model in identifying cardiomyopathy during pregnancy and the postpartum period. Methods and results We used an ECG-based deep learning model to detect cardiomyopathy in a cohort of women who were pregnant or in the postpartum period seen at Mayo Clinic. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. We compared the diagnostic probabilities of the deep learning model with natriuretic peptides and a multivariable model consisting of demographic and clinical parameters. The study cohort included 1807 women; 7%, 10%, and 13% had left ventricular ejection fraction (LVEF) of 35% or less, &lt;45%, and &lt;50%, respectively. The ECG-based deep learning model identified cardiomyopathy with AUCs of 0.92 (LVEF ≤ 35%), 0.89 (LVEF &lt; 45%), and 0.87 (LVEF &lt; 50%). For LVEF of 35% or less, AUC was higher in Black (0.95) and Hispanic (0.98) women compared to White (0.91). Natriuretic peptides and the multivariable model had AUCs of 0.85 to 0.86 and 0.72, respectively. Conclusions An ECG-based deep learning model effectively identifies cardiomyopathy during pregnancy and the postpartum period and outperforms natriuretic peptides and traditional clinical parameters with the potential to become a powerful initial screening tool for cardiomyopathy in the obstetric care setting.

AbstractBackgroundTransvenous implantable cardioverter defibrillators (TV ICD) provide life‐saving therapy for millions of patients worldwide. However, they are susceptible to several potential short‐ and long‐ term complications including cardiac perforation and pneumothorax, lead dislodgement, venous obstruction, and infection. The extravascular ICD system's novel design and substernal implant approach avoids the risks associated with TV ICDs while still providing pacing features and similar generator size to TV ICDs.Study DesignThe EV ICD pivotal study is a prospective, multicenter, single‐arm, nonrandomized, premarket clinical study designed to examine the safety and acute efficacy of the system. This study will enroll up to 400 patients with a Class I or IIa indication for implantation of an ICD. Implanted subjects will be followed up to approximately 3.5 years, depending on when the patient is enrolled.ObjectiveThe clinical trial is designed to demonstrate safety and effectiveness of the EV ICD system in human use. The safety endpoint is freedom from major complications, while the efficacy endpoint is defibrillation success. Both endpoints will be assessed against prespecified criteria. Additionally, this study will evaluate antitachycardia pacing performance, electrical performance, extracardiac pacing sensation, asystole pacing, appropriate and inappropriate shocks, as well as a summary of adverse events.ConclusionThe EV ICD pivotal study is designed to provide clear evidence addressing the safety and efficacy performance of the EV ICD System.

Abstract Aims We sought to investigate whether artificial intelligence (AI) and specifically deep neural networks (NNs) for electrocardiogram (ECG) signal analysis can be explained using human-selected features. We also sought to quantify such explainability and test if the AI model learns features that are similar to a human expert. Methods and results We used a set of 100 000 ECGs that were annotated by human explainable features. We applied both linear and non-linear models to predict published ECG AI models output for the detection of patients’ age and sex. We further used canonical correlation analysis to quantify the amount of shared information between the NN features and human-selected features. We reconstructed single human-selected ECG features from the unexplained NN features using a simple linear model. We noticed a strong correlation between the simple models and the AI output (R2 of 0.49–0.57 for the linear models and R2 of 0.69–0.70 for the non-linear models). We found that the correlation of the human explainable features with either 13 of the strongest age AI features or 15 of the strongest sex AI features was above 0.85 (for comparison, the first 14 principal components explain 90% of the human feature variance). We linearly reconstructed single human-selected ECG features from the AI features with R2 up to 0.86. Conclusion This work shows that NNs for ECG signals extract features in a similar manner to human experts and that they also generate additional novel features that help achieve superior performance.

Abstract Aims Spectrum bias can arise when a diagnostic test is derived from study populations with different disease spectra than the target population, resulting in poor generalizability. We used a real-world artificial intelligence (AI)-derived algorithm to detect severe aortic stenosis (AS) to experimentally assess the effect of spectrum bias on test performance. Methods and results All adult patients at the Mayo Clinic between 1 January 1989 and 30 September 2019 with transthoracic echocardiograms within 180 days after electrocardiogram (ECG) were identified. Two models were developed from two distinct patient cohorts: a whole-spectrum cohort comparing severe AS to any non-severe AS and an extreme-spectrum cohort comparing severe AS to no AS at all. Model performance was assessed. Overall, 258 607 patients had valid ECG and echocardiograms pairs. The area under the receiver operator curve was 0.87 and 0.91 for the whole-spectrum and extreme-spectrum models, respectively. Sensitivity and specificity for the whole-spectrum model was 80% and 81%, respectively, while for the extreme-spectrum model it was 84% and 84%, respectively. When applying the AI-ECG derived from the extreme-spectrum cohort to patients in the whole-spectrum cohort, the sensitivity, specificity, and area under the curve dropped to 83%, 73%, and 0.86, respectively. Conclusion While the algorithm performed robustly in identifying severe AS, this study shows that limiting datasets to clearly positive or negative labels leads to overestimation of test performance when testing an AI algorithm in the setting of classifying severe AS using ECG data. While the effect of the bias may be modest in this example, clinicians should be aware of the existence of such a bias in AI-derived algorithms.

AbstractIntroductionThe efficacy of cardiac resynchronization therapy (CRT) has been widely studied in the medical literature; however, about 30% of candidates fail to respond to this treatment strategy. Smart computational approaches based on clinical data can help expose hidden patterns useful for identifying CRT responders.MethodsWe retrospectively analyzed the electronic health records of 1664 patients who underwent CRT procedures from January 1, 2002 to December 31, 2017. An ensemble of ensemble (EoE) machine learning (ML) system composed of a supervised and an unsupervised ML layers was developed to generate a prediction model for CRT response.ResultsWe compared the performance of EoE against traditional ML methods and the state‐of‐the‐art convolutional neural network (CNN) model trained on raw electrocardiographic (ECG) waveforms. We observed that the models exhibited improvement in performance as more features were incrementally used for training. Using the most comprehensive set of predictors, the performance of the EoE model in terms of the area under the receiver operating characteristic curve and F1‐score were 0.76 and 0.73, respectively. Direct application of the CNN model on the raw ECG waveforms did not generate promising results.ConclusionThe proposed CRT risk calculator effectively discriminates which heart failure (HF) patient is likely to respond to CRT significantly better than using clinical guidelines and traditional ML methods, thus suggesting that the tool can enhanced care management of HF patients by helping to identify high‐risk patients.