Machine Learning of ECG Waveforms to Improve Selection for Testing for Asymptomatic Left Ventricular Dysfunction Prompt

Elizabeth L. Potter, MBBS, BSC, Carlos H.M. Rodrigues, BSC, David B. Ascher, PHD,
Walter P. Abhayaratna, MBBS, PHD, Partho P. Sengupta, MD, Thomas H. Marwick, MBBS, PHD, MPH

J Am Coll Cardiol Img. Jun 16, 2021.  Epublished DOI: 10.1016/j.jcmg.2021.04.020


OBJECTIVES To identify whether machine learning from processing of continuous wave transforms (CWTs) to provide an “energy waveform” electrocardiogram (ewECG) could be integrated with echocardiographic assessment of subclinical systolic and diastolic left ventricular dysfunction (LVD).

BACKGROUND Asymptomatic LVD has management implications, but routine echocardiography is not undertaken in subjects at risk of heart failure. Signal processing of the surface ECG with the use of CWT can identify abnormal myocardial relaxation.

METHODS EwECG and echocardiography were undertaken in 398 participants at risk of heart failure (HF). Reduced global longitudinal strain (GLS #16%)), diastolic abnormalities (E/e0 >15, left atrial enlargement with E/e0 >10 or impaired relaxation) or LV hypertrophy defined LVD. EwECG feature selection and supervised machine-learning by random forest (RF) classifier was undertaken with 643 CWT-derived features and the Atherosclerosis Risk in Communities (ARIC) heart failure risk score.

RESULTS The ARIC score and 18 CWT features were selected to build a RF predictive model for LVD in a training dataset (n ¼ 287; 60% female, median age 71 [interquartile range: 68 to 74] years). Model performance was tested in an in-dependent group (n ¼ 111; 49% female, median age 61 years [59 to 66 years]), demonstrating 85% sensitivity and 72%specificity (area under the receiver-operating characteristic curve [AUC]: 0.83; 95% confidence interval [CI]: 0.74 to 0.92). With ARIC score removed, sensitivity was 88% and specificity, 70% (AUC: 0.78; 95% CI: 0.70 to 0.86). RF models for reduced GLS and diastolic abnormalities including similar features had sensitivities that were unsuitable for screening. Conventional candidates for LVD screening (ARIC score, N-terminal pro–B-type natriuretic peptide, and standard auto-mated ECG analysis) had inferior discriminative ability. Integration of ewECG in screening of people at risk of HF would reduce need for echocardiography by 45% while missing 12% of LVD cases.

CONCLUSIONS Machine learning applied to ewECG is a sensitive screening test for LVD, and its integration into screening of patients at risk for HF would reduce the number of echocardiograms by almost one-half.

© 2021 by the American College of Cardiology Foundation.

Suicide prevention needs to go beyond the blue

From the Chair of Rural & Remote Mental Health – Genevieve Fraser

“Sunflowers” by Vincent Van Gogh

Visitors to Canberra during budget week were able to immerse themselves in more than the budget thanks to the National Gallery’s well timed “Masterpieces from the National Gallery, London”, featuring Vincent Van Gogh’s famous Sunflowers. $5 million dollars’ worth of refurbishments were undertaken last year to eliminate all dark corners of the Canberra exhibition space to shed equal light on each masterpiece. Despite these efforts the darkness of Van Gogh’s tragic story remains. In 1880, after the paint had dried on the petals of his celebrated flowers, Van Gogh walked into a nearby field in Auvers, France and shot himself in the stomach with a revolver.

Politicians who were able to take in Sunflowers over the past few months might spare a thought for those who equally lose hope in fields and homes across rural Australia 2021. Rural Australians suicide at two to three times the national average – a tragedy the 2020 Mental Health Productivity Commission asked our leaders to focus on in future funding decisions.

The report explained unambiguously who was dying, where, and why. The central recommendation for rural Australia was that half of those who suicide in Australia are NOT engaged with the medical system so future mental health spending needed to focus on community-based prevention.

Leaders were also told our suicide prevention model needs to change because suicide rates escalate the further someone lives from a capital city.  And this pattern is consistent across every state and Territory.  

Many of those who suicide in the country have no mental health history but are dealing with a trauma. It could be the loss of a child, a marriage, or a job, and they simply don’t have the tools to cope. They live a long way from mental health specialists, are not talking to their GP’s about their problems, and most of their friends and family don’t recognise the signs.

The families of Dolly Everett and Dr Andrew Bryant have both generously shared their experiences in recent times and both called for more community awareness because it is friends and families that need to know the signs to look for.

Rural Australians considering suicide are most likely to be saved by someone who already knows them or their community. Friends, family and work mates are the main line of defence.

Rural mental health charities like Rural & Remote Mental Health talk face to face with rural people and train locals in suicide prevention because it is the only way to save those not using our mental health services.

We don’t run million-dollar advertising campaigns on council buses and highways because the people we need to reach don’t have council buses or highways. Our work runs on cups of tea, biscuits, sandwiches, town halls and thousands of kilometres on country roads. Our trained presenters live and work across all states and territories. Most have other community-based roles – rural adversity nurses, indigenous elders, mine employees.

If people in these communities put their rural postcode into an online mental health directory it would often deliver the message  “there are no services in your area”. Clinics in neighbouring rural centres look promising until the fine print reveals it is visited once a month by a specialist from somewhere else.

The best way for us to save people is to visit more communities and more often. Without dedicated federal funding we can’t go where the risk is highest, only where there are funds ‘left over’ in local health network budget’s for prevention work. Communities support what we do and want us there. Since our funding stopped in June 2020 one community used the proceeds of a local pig race to fund us to come back. As grateful as we are, why are at-risk rural communities paying for a service that is funded via large charities in our more affluent cities? Rural Australians pay taxes too but are not receiving equitable investment in mental health.

That is why in 2020, charities like ours asked to be funded properly as part of the new mental health strategy.  Without the benefit of lobbyists or sophisticated marketing we provided honest hopeful submissions and genuinely expected that at least 5% of the billions being spent would be focused on the 29% of Australians who live in rural communities.

Our modest request for $2 million a year to reach at risk people across Australia was ignored and letters from politicians reminded us of the millions being spent on hospitals and resources in our cities and regional centres. Our charity is not alone. Many other rural mental health organisations lost some or all their funding and some folded or stopped going where they needed to go.

The responses we received from federal MP’s and ministers included:
“Your funding will not be renewed as the drought is over.”
“Apply to the rural Primary Health Network service – they might have funds left over”
“Doesn’t Beyond Blue/Lifeline/R U OK look after that?”

Many rural charities watched in despair during budget week. We had already received our “no’s” and we could only watch as hundreds of millions were allocated to the city based ‘national’ charities for core funding, project funding and special project funding – hundreds of millions of dollars allocated to a small number of very large organisations. Some of these organisations report over $50 million in cash reserves in their 2020 annual reports. The work they do is important and effective in the communities where they concentrate their efforts. But they don’t go where we go. We know it takes some digging to find those of us that work on the ground but we are there at the bottom of the pile of submissions on desks throughout Canberra.

Rather than give up, we are asking our leaders to allocate a fair percentage of our mental health prevention budget to charities serving the 29% of Australians who live in rural communities.

We need to reach places far from Canberra like the Kimberly where the suicide rate is four times the national average. Places like the Tablelands of Queensland where 55 people lost hope in 2020. Or Wangaratta (VIC) , Maryborough (QLD), West Tamar (TAS), Outback  North and the Outback Southern regions (QLD).

Places where the light of this mental health budget is not yet shining any light.

We need our politicians to see beyond the blue coastlines and cities and into high-risk red areas of our country. People need the tools to save themselves and others because some Australians will always be isolated from medical mental health facilities. Van Gogh’s brother Theo held him in his arms as he succumbed to the injuries inflicted in that rural paddock and lamented that he hadn’t recognised the signs of depression in time. We will keep sending our letters and visiting those paddocks and towns because we know this is our best line of defence.  But we need our leaders to recognise the signs and invest beyond the blue.

Outcomes of Patients With ST-Segment Elevation Myocardial Infarction Admitted During COVID-19 Pandemic Lockdown in Germany – Results of a Single Center Prospective Cohort Study

Manuel Rattka, Lina Stuhler, Claudia Winsauer, Jens Dreyhaupt, Kevin Thiessen, Michael Baumhardt, Sinisa Markovic, Wolfgang Rottbauer and Armin Imhof

1Clinic for Internal Medicine II, University Hospital Ulm – Medical Center, Ulm, Germany
2Institute of Epidemiology and Medical Biometry, Ulm University, Ulm, Germany


Front. Cardiovasc. Med., 20 April 2021 |


Objective: Since the outbreak of the COVID-19 pandemic, healthcare professionals reported declining numbers of patients admitted with ST-segment myocardial infarction (STEMI) associated with increased in-hospital morbidity and mortality. However, the effect of lockdown on outcomes of STEMI patients admitted during the COVID-19 crisis has not been prospectively evaluated.

Methods: A prospective, observational study on STEMI patients admitted to our tertiary care center during the COVID-19 pandemic was conducted. Outcomes of patients admitted during lockdown were compared to those patients admitted before and after pandemic-related lockdown.

Results: A total of 147 patients were enrolled in our study, including 57 patients in the pre-lockdown group (November 1, 2019 to March 20, 2020), 16 patients in the lockdown group (March 21 to April 19, 2020), and 74 patients in the post-lockdown group (April 20 to September 30, 2020). Patients admitted during lockdown had significantly longer time to first medical contact, longer door-to-needle-time, higher serum troponin T levels, worse left ventricular end-diastolic pressure, and higher need for circulatory support. After a median follow-up of 142 days, survival was significantly worse in STEMI patients of the lockdown group (log-rank: p = 0.0035).

Conclusions: This is the first prospective study on outcomes of STEMI patients admitted during public lockdown amid the COVID-19 pandemic. Our results suggest that lockdown might deteriorate outcomes of STEMI patients. Public health strategies to constrain spread of COVID-19, such as lockdown, have to be accompanied by distinct public instructions to ensure timely medical care in acute diseases such as STEMI.

Speech-Induced Atrial Tachycardia

André Zimerman, M.D., and Andre d’Avila, M.D., Ph.D.

Speech-Induced Atrial Tachycardia | NEJM

A 58-year-old man presented to the emergency department with a 1-month history of intermittent dizziness and palpitations that occurred when he spoke. Continuous electrocardiographic monitoring revealed atrial premature beats when he was speaking isolated words (Panel A) and sustained atrial tachycardia up to 167 beats per minute when he was speaking full sentences (Panel B). The arrhythmias promptly subsided when he stopped talking (Panel C). Arrhythmias were reproducibly triggered by any spoken syllable but not by soundless mouth movement, deep breathing, or an inspiratory breath hold. A transthoracic echocardiogram showed a structurally normal heart, and findings on computed tomography of the chest were unremarkable. Speech-induced atrial tachycardia may represent transient aberrant vagal stimulation from the recurrent laryngeal nerve to the atrial autonomic ganglia, resulting in abnormal automaticity or triggered activity. The patient’s symptoms resolved promptly with oral metoprolol treatment, which was continued for 3 months. The symptoms did not recur, and no catheter ablation was performed. At 3 years of follow-up, the patient remained asymptomatic.

May 29, 2021
DOI: 10.1056/NEJMicm2030596

Diastolic Dysfunction and Heart Failure With Preserved Ejection Fraction: Understanding Mechanisms by Using Noninvasive Methods

Masaru Obokata, Yogesh N.V. Reddy, and Barry A. Borlaug

J Am Coll Cardiol Img. 2020 Jan, 13 (1_Part_2) 245–257


Research in the last decade has substantially advanced our understanding of the pathophysiology of heart failure with preserved ejection fraction (HFpEF). However, treatment options remain limited as clinical trials have largely failed to identify effective therapies. Part of this failure may be related to mechanistic heterogeneity. It is speculated that categorizing HFpEF patients based upon underlying pathophysiological phenotypes may represent the key next step in delivering the right therapies to the right patients. Echocardiography may provide valuable insight into both the pathophysiology and underlying phenotypes in HFpEF. Echocardiography also plays a key role in the evaluation of patients with unexplained dyspnea, where HFpEF is suspected but the diagnosis remains unknown. The combination of the E/e′ ratio and right ventricular systolic pressure has recently been shown to add independent value to the diagnostic evaluation of patients suspected of having HFpEF. Finally, echocardiography enables identification of the different causes that mimic HFpEF but are treated differently, such as valvular heart disease, pericardial constriction, and high-output heart failure or infiltrative myopathies such as cardiac amyloid. This review summarizes the current understanding of the pathophysiology and phenotyping of HFpEF with particular attention to the role of echocardiography in this context.


HFpEF is a heterogeneous syndrome, and categorizing patients based upon pathophysiology may provide phenotype-specific therapies.
Echocardiography provides valuable information for assessing pathophysiological mechanisms, phenotyping, and diagnosis in cases of HFpEF.
Further study is needed to establish the HFpEF phenotype and roles of noninvasive imaging in it.


Heart failure with preserved ejection fraction (HFpEF) is a common clinical syndrome that is increasing in prevalence. Rather than having an isolated abnormality in left ventricular (LV) diastolic function, patients with HFpEF display multifaceted limitations in cardiac, vascular, and peripheral functions (1). Phenotyping based upon pathophysiology, comorbidities, or some combination may provide targeted therapies for the specific HFpEF subpopulations that are positioned to derive the greatest benefit (2).

Cardiovascular imaging plays a key role in the diagnosis and evaluation of HFpEF, particularly echocardiography, which allows for assessment of cardiac structure, function, and hemodynamics (3). This review summarizes the current understanding of the pathophysiology and phenotypes of HFpEF, with a focus on the essential role of imaging for the evaluation and care of patients with HFpEF.

Pathophysiology of HFpEF: Beyond Diastolic Dysfunction

LV diastolic dysfunction plays a fundamental, overarching role in the pathophysiology of HFpEF (1). LV diastolic dysfunction is defined by an impairment in relaxation, an increase in viscoelastic chamber stiffness, or some combination of the 2 (4,5) and leads to symptomatic HF by causing elevated filling pressures at rest or with exertion (6). Elevated filling pressures promote symptoms of dyspnea (7), impair exercise capacity (7,8), increase risk for HF hospitalization (9), and decrease survival in HFpEF (10). The importance and assessment of LV diastolic dysfunction in HFpEF are reviewed in detail in other articles in this issue.

Although diastolic dysfunction is central to HFpEF, it is also important to acknowledge that there are declines in LV relaxation and compliance with normal aging or with cardiometabolic comorbidities such as obesity, insulin resistance, and hypertension (11–13). Not all patients with diastolic dysfunction have or will develop clinical HFpEF (14,15). Research in the past decade has demonstrated that, in addition to LV diastolic dysfunction, multiple nondiastolic abnormalities in cardiovascular system contribute to the syndrome of HFpEF. These abnormalities include subtle LV systolic dysfunction, left atrial (LA) impairment, relative pericardial restraint, abnormal right ventricular-pulmonary artery coupling, pulmonary vascular disease, systemic vascular stiffening, coronary and peripheral microvascular dysfunction, and chronotropic incompetence…

Association of Echocardiographic Left Ventricular End-Systolic Volume and Volume-Derived Ejection Fraction With Outcome in Asymptomatic Chronic Aortic Regurgitation

Li-Tan Yang, MD1,2Vidhu Anand, MBBS1Elena I. Zambito, RDCS1Patricia A. Pellikka, MD1Christopher G. Scott, MS3Prabin Thapa, MS3Ratnasari Padang, MBBS, PhD1Masaaki Takeuchi, MD4Rick A. Nishimura, MD1Maurice Enriquez-Sarano, MD1Hector I. Michelena, MD1

JAMA Cardiol. 2021;6(2):189-198. doi:10.1001/jamacardio.2020.5268

Key Points

Question  Are disk-summation method–derived left ventricular (LV) end-systolic volume index and LV ejection fraction (LVEF) associated with mortality in asymptomatic patients with hemodynamically significant chronic aortic regurgitation?

Findings  In this cohort study of 492 asymptomatic patients with moderately severe to severe aortic regurgitation, besides conventional linear LVEF and LV end-systolic dimension index, LV end-systolic volume index and volume-derived LVEF were robust independent factors associated with mortality. Thresholds of increased mortality risk for linear LVEF and volume-derived LVEF were 60%, and for LV end-systolic dimension index and volume index, 21 to 22 mm/m2 and 40 to 45 mL/m2, respectively; previously reported LV end-systolic volume index threshold of 45 mL/m2 was a robust marker of increased risk of death.

Meaning  In asymptomatic low-risk patients with aortic regurgitation, LV end-systolic volume index and volume-derived LVEF provided similar risk-stratifying power as conventional linear LVEF and LV end-systolic dimension index.


Importance  Volumetric measurements by transthoracic echocardiogram may better reflect left ventricular (LV) remodeling than conventional linear LV dimensions. However, the association of LV volumes with mortality in patients with chronic hemodynamically significant aortic regurgitation (AR) is unknown.

Objective  To assess whether LV volumes and volume-derived LV ejection fraction (Vol-LVEF) are determinants of mortality in AR.

Design, Setting, and Participants  This cohort study included consecutive asymptomatic patients with chronic moderately severe to severe AR from a tertiary referral center (January 2004 through April 2019).

Exposures  Clinical and echocardiographic data were analyzed retrospectively. Aortic regurgitation severity was graded by comprehensive integrated approach. De novo disk-summation method was used to derive LV volumes and Vol-LVEF.

Main Outcome and Measures  Associations between all-cause mortality under medical surveillance and the following LV indexes: linear LV end-systolic dimension index (LVESDi), linear LVEF, LV end-systolic volume index (LVESVi), and Vol-LVEF.

Results  Of 492 asymptomatic patients (mean [SD] age, 60 [17] years; 425 men [86%]), ischemic heart disease prevalence was low (41 [9%]), and 453 (92.1%) had preserved linear LVEF (≥50%) with mean (SD) LVESVi of 41 (15) mL/m2. At a median (interquartile range) of 5.4 (2.5-10.1) years, 66 patients (13.4%) died under medical surveillance; overall survival was not different than the age- and sex-matched general population (P = .55). Separate multivariate models, adjusted for age, sex, Charlson Comorbidity Index, and AR severity, demonstrated that in addition to linear LVEF and LVESDi, LVESVi and Vol-LVEF were independently associated with mortality under surveillance (all P < .046) with similar C statistics (range, 0.83-0.84). Spline curves showed that continuous risks of death started to rise for both linear LVEF and Vol-LVEF less than 60%, LVESVi more than 40 to 45 mL/m2, and LVESDi above 21 to 22 mm/m2. As dichotomized variables, patients with LVESVi more than 45 mL/m2 exhibited increased relative death risk (hazard ratio, 1.93; 95% CI, 1.10-3.38; P = .02) while LVESDi more than 20 mm/m2 did not (P = .32). LVESVi more than 45 mL/m2 showed a decreased survival trend compared with expected population survival.

Conclusions and Relevance  In this large asymptomatic cohort of patients with hemodynamically significant AR, LVESVi and Vol-LVEF worked equally as well as LVESDi and linear LVEF in risk discriminating patients with excess mortality. A LVESVi threshold of 45 mL/m2 or greater was significantly associated with an increased mortality risk.

Long Working Hours and Risk of Recurrent Coronary Events

Xavier Trudel PhD; Chantal Brisson PhD; Denis Talbot PhD; Mahée Gilbert-Ouimet PhD; Alain Milot MD

Journal of the American College of Cardiology
Volume 77, Issue 13, 6 April 2021, Pages 1616-1625



Evidence from prospective studies has suggested that long working hours are associated with incident coronary heart disease (CHD) events. However, no previous study has examined whether long working hours are associated with an increased risk of recurrent CHD events among patients returning to work after a first myocardial infarction (MI).


The purpose of this study was to examine the effect of long working hours on the risk of recurrent CHD events.


This is a prospective cohort study of 967 men and women age 35 to 59 years who returned to work after a first MI. Patients were recruited from 30 hospitals across the province of Quebec, Canada. The mean follow-up duration was 5.9 years. Long working hours were assessed on average 6 weeks after their return to work. Incident CHD events (fatal or nonfatal MI and unstable angina) occurring during follow-up were determined using patients’ medical files. Hazard ratios were estimated using Cox proportional hazard regression models. Splines and fractional polynomial regressions were used for flexible exposure and time modeling.


Recurrent CHD events occurred among 205 patients. Participants working long hours (≥55 h/week) had a higher risk of recurrent CHD events after controlling for sociodemographics, lifestyle-related risk factors, clinical risk factors, work environment factors, and personality factors (hazard ratio vs. 35 to 40 h/week: 1.67; 95% confidence interval: 1.10 to 2.53). These results showed a linear risk increase after 40 h/week and a stronger effect after the first 4 years of follow-up and when long working hours are combined with job strain.


Among patients returning to work after a first MI, longer working hours per week is associated with an increased risk of recurrent CHD events. Secondary prevention interventions aiming to reduce the number of working hours among these patients may lower the risk of CHD recurrence.

Collaboration during Crisis: A Novel Point-of-Care Ultrasound Alliance among Emergency Medicine, Internal Medicine, and Cardiology in the COVID-19 Era

Nova Panebianco MD, MPH; Cameron M.Baston MD; Mili Mehta MD; Victor A.Ferrari MD; Dinesh Jagasia MD; Marielle Scherrer-Crosbie MD, PhD; Srinath Adusumalli MD, MSHP

Journal of the American Society of Echocardiography
Volume 34, Issue 3, March 2021, Pages 325-326

To the Editor:

The coronavirus disease 2019 (COVID-19) pandemic may be the greatest public health emergency we will experience in our lifetimes. It has both exposed major shortcomings in the American medical system and revealed our capacity for innovation and collaboration. Early in disaster planning at our institution, we identified several issues regarding echocardiography: (1) personal protective equipment shortages, (2) the infection control risk posed by large ultrasound machines, (3) heterogenous knowledge of basic point-of-care ultrasound (POCUS) echocardiography, and (4) a need for cardiac diagnostics beyond the scope of basic POCUS (e.g., regional wall motion abnormalities).12345

Before COVID-19, an enterprise-level multidisciplinary POCUS committee had been organized to address POCUS training, credentialing, and image archival. With multispecialty agreement, including members of this committee, the default method of cardiac ultrasound imaging became POCUS in patients with or suspected of having COVID-19. Echocardiography laboratory sonographers were available to remotely support and direct frontline providers during bedside echocardiographic image acquisition using either in-room intensive care unit cameras when the provider was using a cart-based machine or the teleguidance feature on the handheld ultrasound systems. Echocardiography faculty members, with access to the POCUS image archive, offered remote real-time image interpretation assistance (Figure 1). This initiative minimized the number of providers exposed to patients with COVID-19 and maximized infection control precautions, while appropriately triaging the need for comprehensive echocardiography.

We have identified the following points as instrumental for success: (1) multispecialty collaboration, (2) repurposing of existing technology, (3) acquisition of handheld ultrasound devices with teleguidance capabilities, (4) dedicated informatics support, (5) scalable education through virtual meetings and recorded sessions, and (6) frequent communication. Penn Health Tech, an interdisciplinary center dedicated to innovation, was instrumental in acquiring handheld ultrasound devices. Our chief medical informatics officer deputized an information services project manager, who had frequent communication with providers and industry to understand and build the POCUS image archive. Our handheld ultrasound device vendor provided virtual product training sessions, and recorded sessions were posted on the institutional website. The Division of Cardiovascular Medicine educated staff members about the work flow and took call from 9 am to 5 pm on business days. The University of Pennsylvania institutional review board reviewed the data collection policy and approved the protocol.

In spring 2020, we averaged a daily census of 60 to 90 patients with COVID-19, with an average of three requests for cardiology support daily. In the 151 cardiac POCUS studies performed among patients with COVID-19, there were 26 echocardiography laboratory overreads, demonstrating a utilization rate of 17%. Of these, 23 (88%) were diagnostic and prevented sonographer exposure to infectious patients. On the first day of deployment, the protocol enabled the detection of occult systolic heart failure in an ill patient with COVID-19, with subsequent diuresis that reversed multisystem organ failure. In another patient suspected to have COVID-19, pericardial tamponade was suspected on emergency department POCUS, with overread by cardiology, and the patient was taken directly to the catheterization laboratory for drainage.

Novel technology and a health care crisis made it possible to break down existing silos and inspired innovation while maintaining the independence of the participating disciplines. This collaboration is generalizable to other institutions, which can adapt this work flow to fit their technological capabilities, and to other patients in need of urgent POCUS.

Prediction of atrial fibrillation by 12-lead electrocardiogram parameters in patients without structural heart disease

N Hirota, S Suzuki, T Arita, N Yagi, T Otsuka, H Semba, H Kano, S Matsuno, Y Kato, T Uejima, Y Oikawa, J Yajima, T Yamashita

European Heart Journal, Volume 41, Issue Supplement_2, November 2020, ehaa946.0536, Published: 25 November 2020



Recently, the analysis of electrocardiogram (ECG) waveform by artificial intelligence has been reported to pick out those who have atrial fibrillation (AF) or have a high potential of developing AF, which, however, cannot explain the mechanisms or algorisms for the prediction from its nature.Purpose

The purpose of this study is to conduct a comprehensive analysis to investigate the difference of weighting in predicting capability for AF among hundreds of automatically-measured ECG parameters using a single ECG at sinus rhythm.

Methods and results

Out of Shinken Database 2010–2017 (n=19170), 12825 patients were extracted, where those with ECG showing AF rhythm at the initial visit (including all persistent/permanent AF and a part of paroxysmal AF) and those with structural heart diseases were excluded. Out of 639 automatically-measured ECG parameters in MUSE data management system (GE Healthcare, USA), 438 were used. [Analysis 1] A predicting model for paroxysmal AF were determined by logistic regression analysis (Total, n=12825; paroxysmal AF, n=1138), showing a high predictive capability (AUC = 0.780, p<0.001). In this model, the relative contribution of ECG parameters (by coefficient of determination) according to the time phase were P:72.4%, QRS:32.7%, and ST-T:13.7%, respectively (Figure A). [Analysis 2] Excluding AF at baseline, a predicting model for new-developed AF were determined by Cox regression analysis (Total, n=11687; new-developed AF, n=87), showing a high predictive capability (AUC = 0.887, p<0.001). In this model, the relative contribution of parameters (by log likelihood) according to the time phase were P:40.8%, QRS:42.5%, and ST-T:24.9%, respectively (Figure B).


We determined ECG parameters that potentially contribute to picking up existing AF or predicting future development of AF, where the measurement of P wave strongly contributed in the former whereas all time phases were similarly important in the latter.