User login
An artificial intelligence-enabled ECG model identified patients with intermittent atrial fibrillation in a 10-second test with 83% accuracy, based on data from more than 180,000 individuals.
“We have previously shown convolution neural networks can evaluate the resting ECG for detection of antiarrhythmic drug levels, abnormal electrolytes levels, and detection of asymptomatic left ventricular dysfunction, providing proof of concept that clinically important phenomena can be detected with artificial intelligence (AI) applications to the ECG,” wrote Zachi I. Attia, an electrical engineer and a primary author of the study, is with the Mayo Clinic, Rochester, Minn., and colleagues.
In a study published in the Lancet, the researchers reviewed data from 649,931 normal sinus rhythm ECGs collected from 180,922 adults between December 1993 and July 2017.
The ECGs were divided into three groups: training (454,789 ECGs from 126,526 patients) internal validation (64,340 ECGs from 18,116 patients) and testing (130,802 ECGs from 36,280 patients). The primary outcome was whether the AI-programmed ECG could identify AFib in a total of 3,051 patients in the testing data set who had verified AFib before being tested with the AI device. The AI-enabled ECG was designed to detect subtle changes using neural network technology previously used by the researchers to identify ventricular dysfunction.
Overall, a single ECG scan identified AFib with an accuracy of 79.4%, an area under the curve (AUC) of 0.87, sensitivity of 79.0%, and specificity of 79.5%. When researchers reviewed multiple ECGs from a 1-month window of either the study start date or 31 days before the first AFib, the accuracy increased to 83.3%, with an AUC of 0.90, sensitivity of 82.3%, and specificity of 83.4%.
The results support the use of subtle changes on normal sinus rhythm ECG to identify patient with potentially undetected AFib, and suggest that AI-enabled ECGs could be used at the point of care to identify patients at risk after unexplained strokes, also known as embolic stroke of undetermined source (ESUS), or heart failure, the researchers noted.
“Although it would require further study, it is possible that this algorithm could identify a high-risk subset of patients with ESUS who could benefit from empirical anticoagulation,” the researchers said.
The study findings were limited by several factors, including possible mislabeling of patients with unidentified atrial fibrillation who were classified negative. In addition, the prevalence of AFib in the study population may be higher than in the general population, they said.
However, the results suggest that use a noninvasive, widely available, point of care test to identify AFib “could have important implications for atrial fibrillation screening and for the management of patients with unexplained stroke,” they concluded.
This study was funded by internal resources of the Mayo Clinic. The researchers had no financial conflicts to disclose.
SOURCE: Attia ZI et al. Lancet. 2019 Aug 1. doi. org/10.1016/S0140-6736(19)31721-0.
This artificial intelligence-enabled ECG interpretation is groundbreaking in creating an algorithm to reveal the likelihood of atrial fibrillation in ECGs showing sinus rhythm.
AFib is now considered a global pandemic and needs to be detected not only to manage the arrhythmia but also to prevent comorbidities and death.
A 10-second, 12-lead ECG in current clinical practice is unlikely to reveal possible AFib if not present in this short monitoring time. However, the findings have clinical importance, particularly in identifying silent AFib and may have important implications for secondary prevention of patients with embolic stroke of undetermined source in terms of providing appropriate oral anticoagulation to prevent recurrences of stroke. The AI-enabled algorithm would require further validation in a different patient cohort, testing a healthier out-of-hospital population, as well as a rigorous prospective clinical trial assessment.
Future research areas include combining ECG algorithms with demographic variables, clinical features, and biomarkers, as well as exploring the use of wearable devices linking these variables and AI for smart monitoring to diagnose AFib.
Jeroen Hendriks, MD, of the University of Adelaide (Australia), and Larissa Fabritz, MD, of the University of Birmingham (England), made these comments in an accompanying editorial. Dr. Hendriks disclosed lecture or consulting fees from Medtronic and Pfizer/Bristol-Myers Squibb. Dr. Fabritz is the inventor of two patents and disclosed research grants and nonfinancial support from European research institutions and Gilead.
This artificial intelligence-enabled ECG interpretation is groundbreaking in creating an algorithm to reveal the likelihood of atrial fibrillation in ECGs showing sinus rhythm.
AFib is now considered a global pandemic and needs to be detected not only to manage the arrhythmia but also to prevent comorbidities and death.
A 10-second, 12-lead ECG in current clinical practice is unlikely to reveal possible AFib if not present in this short monitoring time. However, the findings have clinical importance, particularly in identifying silent AFib and may have important implications for secondary prevention of patients with embolic stroke of undetermined source in terms of providing appropriate oral anticoagulation to prevent recurrences of stroke. The AI-enabled algorithm would require further validation in a different patient cohort, testing a healthier out-of-hospital population, as well as a rigorous prospective clinical trial assessment.
Future research areas include combining ECG algorithms with demographic variables, clinical features, and biomarkers, as well as exploring the use of wearable devices linking these variables and AI for smart monitoring to diagnose AFib.
Jeroen Hendriks, MD, of the University of Adelaide (Australia), and Larissa Fabritz, MD, of the University of Birmingham (England), made these comments in an accompanying editorial. Dr. Hendriks disclosed lecture or consulting fees from Medtronic and Pfizer/Bristol-Myers Squibb. Dr. Fabritz is the inventor of two patents and disclosed research grants and nonfinancial support from European research institutions and Gilead.
This artificial intelligence-enabled ECG interpretation is groundbreaking in creating an algorithm to reveal the likelihood of atrial fibrillation in ECGs showing sinus rhythm.
AFib is now considered a global pandemic and needs to be detected not only to manage the arrhythmia but also to prevent comorbidities and death.
A 10-second, 12-lead ECG in current clinical practice is unlikely to reveal possible AFib if not present in this short monitoring time. However, the findings have clinical importance, particularly in identifying silent AFib and may have important implications for secondary prevention of patients with embolic stroke of undetermined source in terms of providing appropriate oral anticoagulation to prevent recurrences of stroke. The AI-enabled algorithm would require further validation in a different patient cohort, testing a healthier out-of-hospital population, as well as a rigorous prospective clinical trial assessment.
Future research areas include combining ECG algorithms with demographic variables, clinical features, and biomarkers, as well as exploring the use of wearable devices linking these variables and AI for smart monitoring to diagnose AFib.
Jeroen Hendriks, MD, of the University of Adelaide (Australia), and Larissa Fabritz, MD, of the University of Birmingham (England), made these comments in an accompanying editorial. Dr. Hendriks disclosed lecture or consulting fees from Medtronic and Pfizer/Bristol-Myers Squibb. Dr. Fabritz is the inventor of two patents and disclosed research grants and nonfinancial support from European research institutions and Gilead.
An artificial intelligence-enabled ECG model identified patients with intermittent atrial fibrillation in a 10-second test with 83% accuracy, based on data from more than 180,000 individuals.
“We have previously shown convolution neural networks can evaluate the resting ECG for detection of antiarrhythmic drug levels, abnormal electrolytes levels, and detection of asymptomatic left ventricular dysfunction, providing proof of concept that clinically important phenomena can be detected with artificial intelligence (AI) applications to the ECG,” wrote Zachi I. Attia, an electrical engineer and a primary author of the study, is with the Mayo Clinic, Rochester, Minn., and colleagues.
In a study published in the Lancet, the researchers reviewed data from 649,931 normal sinus rhythm ECGs collected from 180,922 adults between December 1993 and July 2017.
The ECGs were divided into three groups: training (454,789 ECGs from 126,526 patients) internal validation (64,340 ECGs from 18,116 patients) and testing (130,802 ECGs from 36,280 patients). The primary outcome was whether the AI-programmed ECG could identify AFib in a total of 3,051 patients in the testing data set who had verified AFib before being tested with the AI device. The AI-enabled ECG was designed to detect subtle changes using neural network technology previously used by the researchers to identify ventricular dysfunction.
Overall, a single ECG scan identified AFib with an accuracy of 79.4%, an area under the curve (AUC) of 0.87, sensitivity of 79.0%, and specificity of 79.5%. When researchers reviewed multiple ECGs from a 1-month window of either the study start date or 31 days before the first AFib, the accuracy increased to 83.3%, with an AUC of 0.90, sensitivity of 82.3%, and specificity of 83.4%.
The results support the use of subtle changes on normal sinus rhythm ECG to identify patient with potentially undetected AFib, and suggest that AI-enabled ECGs could be used at the point of care to identify patients at risk after unexplained strokes, also known as embolic stroke of undetermined source (ESUS), or heart failure, the researchers noted.
“Although it would require further study, it is possible that this algorithm could identify a high-risk subset of patients with ESUS who could benefit from empirical anticoagulation,” the researchers said.
The study findings were limited by several factors, including possible mislabeling of patients with unidentified atrial fibrillation who were classified negative. In addition, the prevalence of AFib in the study population may be higher than in the general population, they said.
However, the results suggest that use a noninvasive, widely available, point of care test to identify AFib “could have important implications for atrial fibrillation screening and for the management of patients with unexplained stroke,” they concluded.
This study was funded by internal resources of the Mayo Clinic. The researchers had no financial conflicts to disclose.
SOURCE: Attia ZI et al. Lancet. 2019 Aug 1. doi. org/10.1016/S0140-6736(19)31721-0.
An artificial intelligence-enabled ECG model identified patients with intermittent atrial fibrillation in a 10-second test with 83% accuracy, based on data from more than 180,000 individuals.
“We have previously shown convolution neural networks can evaluate the resting ECG for detection of antiarrhythmic drug levels, abnormal electrolytes levels, and detection of asymptomatic left ventricular dysfunction, providing proof of concept that clinically important phenomena can be detected with artificial intelligence (AI) applications to the ECG,” wrote Zachi I. Attia, an electrical engineer and a primary author of the study, is with the Mayo Clinic, Rochester, Minn., and colleagues.
In a study published in the Lancet, the researchers reviewed data from 649,931 normal sinus rhythm ECGs collected from 180,922 adults between December 1993 and July 2017.
The ECGs were divided into three groups: training (454,789 ECGs from 126,526 patients) internal validation (64,340 ECGs from 18,116 patients) and testing (130,802 ECGs from 36,280 patients). The primary outcome was whether the AI-programmed ECG could identify AFib in a total of 3,051 patients in the testing data set who had verified AFib before being tested with the AI device. The AI-enabled ECG was designed to detect subtle changes using neural network technology previously used by the researchers to identify ventricular dysfunction.
Overall, a single ECG scan identified AFib with an accuracy of 79.4%, an area under the curve (AUC) of 0.87, sensitivity of 79.0%, and specificity of 79.5%. When researchers reviewed multiple ECGs from a 1-month window of either the study start date or 31 days before the first AFib, the accuracy increased to 83.3%, with an AUC of 0.90, sensitivity of 82.3%, and specificity of 83.4%.
The results support the use of subtle changes on normal sinus rhythm ECG to identify patient with potentially undetected AFib, and suggest that AI-enabled ECGs could be used at the point of care to identify patients at risk after unexplained strokes, also known as embolic stroke of undetermined source (ESUS), or heart failure, the researchers noted.
“Although it would require further study, it is possible that this algorithm could identify a high-risk subset of patients with ESUS who could benefit from empirical anticoagulation,” the researchers said.
The study findings were limited by several factors, including possible mislabeling of patients with unidentified atrial fibrillation who were classified negative. In addition, the prevalence of AFib in the study population may be higher than in the general population, they said.
However, the results suggest that use a noninvasive, widely available, point of care test to identify AFib “could have important implications for atrial fibrillation screening and for the management of patients with unexplained stroke,” they concluded.
This study was funded by internal resources of the Mayo Clinic. The researchers had no financial conflicts to disclose.
SOURCE: Attia ZI et al. Lancet. 2019 Aug 1. doi. org/10.1016/S0140-6736(19)31721-0.
FROM THE LANCET