Meeting News

Smartwatch technology can detect AF, hyperkalemia

Joseph Bumgarner

ORLANDO, Fla. — An accessory and algorithm used in conjunction with a smartwatch accurately differentiated atrial fibrillation from sinus rhythm, and another algorithm detected hyperkalemia, according to data presented at the American College of Cardiology Scientific Session.

Heart rhythm

The technology (KardiaBand, AliveCor) is the first FDA-cleared Apple Watch accessory that allows a patient to record a rhythm strip for 30 seconds and is paired with an app that includes an algorithm that instantly adjudicates the recorded rhythm.

“Like other wearable devices, this novel technology needed to be vetted to assess its accuracy and capability in accomplishing the task promised in a clinical setting,” Khaldoun G. Tarakji, MD, MPH, staff physician in the section of electrophysiology in the Robert and Suzanne Tomsich Department of Cardiovascular Medicine, Sydell and Arnold Miller Family Heart & Vascular Institute at Cleveland Clinic and principal investigator of the study, told Cardiology Today. “We decided to conduct the study to assess the accuracy of the algorithm through the paired app of the KardiaBand to detect AF compared with simultaneous 12-lead ECG, to assess the fidelity of the KardiaBand recording and the ability of an independent reader of the recording to accurately detect AF compared with the simultaneous 12-lead ECG and to explore a clinical setting where the use of the wearable device could impact health care delivery of patients with AF.”

Joseph Bumgarner, MD, electrophysiology fellow at Cleveland Clinic, and colleagues enrolled 100 consecutive patients (mean age, 68 years) and took 169 simultaneous ECG and Kardia\Band recordings, of which 57 were not interpretable by the KardiaBand.

Compared with the ECG, the KardiaBand interpreted AF with 93% sensitivity, 84% specificity and a kappa coefficient of 0.77, according to the researchers.

When a physician interpreted the KardiaBand recordings, sensitivity was 99%, specificity was 83% and the kappa coefficient was 0.83. When a physician interpreted the recordings that were not interpretable by the KardiaBand, AF was diagnosed with 100% sensitivity, 80% specificity and a kappa coefficient of 0.74, according to the researchers.

Agreement was strong in the 113 cases in which KardiaBand and physician readings of the same recording were interpretable (kappa coefficient, 0.88).

Tarakji noted that the patients were all scheduled for elective cardioversion by a cardiologist or electrophysiologist, but 8% of them turned out to be in sinus rhythm.

“This is one simple example of how a tool like this, if used intelligently, could enable us to avoid bringing patients in for unnecessary cardioversions, saving both patients and the health care system a significant amount of time and resources,” he told Cardiology Today.

Although the technology is already available to the public, “clinician adoption of any wearable technology is dependent on multiple factors, including more evidence assessing these products, guidance and data about potential applications and, most importantly, studies showing outcome data for both our patients and our health care system. Our adoption in clinical practice is also dependent on the right platform and setting that offers clinical service to patients without overwhelming physicians with the flow of data.”

Hyperkalemia

In another presentation, researchers reported that an artificial intelligence algorithm used with the KardiaBand detected hyperkalemia from a two-lead ECG.

Conner D. Galloway, senior data scientist with AliveCor, and colleagues analyzed 709,000 patients who underwent 2.1 million ECGs and had 4 million serum potassium values at Mayo Clinic between 1994 and 2017; two-thirds were used as the training group and one-third were used as the testing group. All patients had a serum potassium value acquired within 12 hours of an ECG.

Hyperkalemia was defined as serum potassium > 5 mmol/L. The final analysis included 20,196 positive samples and 2 million negative ones. The researchers tested whether the AI algorithm using a deep neural network could detect hyperkalemia using only ECG leads I and II.

According to the researchers, in the testing group of 698,939 ECGs from 224,690 patients, the area under the curve was 0.87, with sensitivity of 85% and specificity of 72%. Use of an algorithmic certainty threshold improved the AUC to 0.91 and sensitivity to 91%; specificity remained at 72%.

The researchers also conducted a prospective analysis in 10 patients undergoing hemodialysis. The patients used an early version of the KardiaBand to acquire a 4-hour ECG reading during two dialysis sessions, and also underwent bloodwork.

In that analysis, sensitivity by duration was 94% and specificity was 74%. – by Erik Swain

References:

Bumgarner J, et al. Abstract 1183-018.

Galloway CD, et al. Highlighted Original Research: Arrhythmias and Clinical EP and the Year in Review. Both presented at: American College of Cardiology Scientific Session; March 10-12, 2018; Orlando, Fla.

Bumgarner J, et al. J Am Coll Cardiol. 2018;doi:doi:10.1016/j.jacc.2018.03.003.

Disclosures: The monitors were provided by AliveCor. Bumgarner reports no relevant financial disclosures. Tarakji reports he serves on a medical advisory board for AliveCor and Medtronic. Galloway reports he is an employee of AliveCor.

 

Joseph Bumgarner

ORLANDO, Fla. — An accessory and algorithm used in conjunction with a smartwatch accurately differentiated atrial fibrillation from sinus rhythm, and another algorithm detected hyperkalemia, according to data presented at the American College of Cardiology Scientific Session.

Heart rhythm

The technology (KardiaBand, AliveCor) is the first FDA-cleared Apple Watch accessory that allows a patient to record a rhythm strip for 30 seconds and is paired with an app that includes an algorithm that instantly adjudicates the recorded rhythm.

“Like other wearable devices, this novel technology needed to be vetted to assess its accuracy and capability in accomplishing the task promised in a clinical setting,” Khaldoun G. Tarakji, MD, MPH, staff physician in the section of electrophysiology in the Robert and Suzanne Tomsich Department of Cardiovascular Medicine, Sydell and Arnold Miller Family Heart & Vascular Institute at Cleveland Clinic and principal investigator of the study, told Cardiology Today. “We decided to conduct the study to assess the accuracy of the algorithm through the paired app of the KardiaBand to detect AF compared with simultaneous 12-lead ECG, to assess the fidelity of the KardiaBand recording and the ability of an independent reader of the recording to accurately detect AF compared with the simultaneous 12-lead ECG and to explore a clinical setting where the use of the wearable device could impact health care delivery of patients with AF.”

Joseph Bumgarner, MD, electrophysiology fellow at Cleveland Clinic, and colleagues enrolled 100 consecutive patients (mean age, 68 years) and took 169 simultaneous ECG and Kardia\Band recordings, of which 57 were not interpretable by the KardiaBand.

Compared with the ECG, the KardiaBand interpreted AF with 93% sensitivity, 84% specificity and a kappa coefficient of 0.77, according to the researchers.

When a physician interpreted the KardiaBand recordings, sensitivity was 99%, specificity was 83% and the kappa coefficient was 0.83. When a physician interpreted the recordings that were not interpretable by the KardiaBand, AF was diagnosed with 100% sensitivity, 80% specificity and a kappa coefficient of 0.74, according to the researchers.

Agreement was strong in the 113 cases in which KardiaBand and physician readings of the same recording were interpretable (kappa coefficient, 0.88).

Tarakji noted that the patients were all scheduled for elective cardioversion by a cardiologist or electrophysiologist, but 8% of them turned out to be in sinus rhythm.

“This is one simple example of how a tool like this, if used intelligently, could enable us to avoid bringing patients in for unnecessary cardioversions, saving both patients and the health care system a significant amount of time and resources,” he told Cardiology Today.

PAGE BREAK

Although the technology is already available to the public, “clinician adoption of any wearable technology is dependent on multiple factors, including more evidence assessing these products, guidance and data about potential applications and, most importantly, studies showing outcome data for both our patients and our health care system. Our adoption in clinical practice is also dependent on the right platform and setting that offers clinical service to patients without overwhelming physicians with the flow of data.”

Hyperkalemia

In another presentation, researchers reported that an artificial intelligence algorithm used with the KardiaBand detected hyperkalemia from a two-lead ECG.

Conner D. Galloway, senior data scientist with AliveCor, and colleagues analyzed 709,000 patients who underwent 2.1 million ECGs and had 4 million serum potassium values at Mayo Clinic between 1994 and 2017; two-thirds were used as the training group and one-third were used as the testing group. All patients had a serum potassium value acquired within 12 hours of an ECG.

Hyperkalemia was defined as serum potassium > 5 mmol/L. The final analysis included 20,196 positive samples and 2 million negative ones. The researchers tested whether the AI algorithm using a deep neural network could detect hyperkalemia using only ECG leads I and II.

According to the researchers, in the testing group of 698,939 ECGs from 224,690 patients, the area under the curve was 0.87, with sensitivity of 85% and specificity of 72%. Use of an algorithmic certainty threshold improved the AUC to 0.91 and sensitivity to 91%; specificity remained at 72%.

The researchers also conducted a prospective analysis in 10 patients undergoing hemodialysis. The patients used an early version of the KardiaBand to acquire a 4-hour ECG reading during two dialysis sessions, and also underwent bloodwork.

In that analysis, sensitivity by duration was 94% and specificity was 74%. – by Erik Swain

References:

Bumgarner J, et al. Abstract 1183-018.

Galloway CD, et al. Highlighted Original Research: Arrhythmias and Clinical EP and the Year in Review. Both presented at: American College of Cardiology Scientific Session; March 10-12, 2018; Orlando, Fla.

Bumgarner J, et al. J Am Coll Cardiol. 2018;doi:doi:10.1016/j.jacc.2018.03.003.

Disclosures: The monitors were provided by AliveCor. Bumgarner reports no relevant financial disclosures. Tarakji reports he serves on a medical advisory board for AliveCor and Medtronic. Galloway reports he is an employee of AliveCor.

 

    See more from American College of Cardiology Annual Scientific Session