Changes in vocal features predict AF in novel screening method
A novel screening method using vowel sounds accurately predicted atrial fibrillation in a cohort of patients referred for cardioversion, researchers reported.
“This research demonstrates the ability of the developed ‘AF indicator,’ derived from voice analysis features, to detect atrial fibrillation,” Gregory Golovchiner, MD, head of the electrophysiology and pacing unit in the department of cardiology at Rabin Medical Center in Petah Tikva, Israel, told Healio. “The findings of this study open horizons for noninvasive, low-cost, age-friendly, prolonged and systematic monitoring of patients with known AF, as well as screening for AF in populations at risk.”
In a prospective study, Golovchiner and colleagues analyzed data from 158 patients aged 35 to 85 years with persistent AF admitted for nonurgent cardioversion at two centers between August 2016 and June 2020. The mean age was 71 years and 43% of participants were women. Patients pronounced the vowels “Ahh” and “Ohh” for at least 1 minute and were recorded synchronously with an ECG tracing. The recordings were performed before the cardioversion and then repeated after successful cardioversion. Researchers developed a speaker-dependent algorithm to provide an "AF indicator" for detection of AF from the speech signal, based on detecting changes in the set of vocal features using common features used in voice analysis.
“Due to the heart’s location inside the chest cavity, in close contact with the lungs and respiratory tract, heart contractions change the dynamics of air flow through the vocal cords,” the researchers wrote. “These periodic changes of the air flow affect the spectral properties of the voice signal. Irregular heart contraction, such as that occurs during AF episodes, is expected to modulate voice signals differently than a regular heart contraction.”
The findings were published in the Journal of Cardiovascular Electrophysiology.
The final analysis of “Ahh” and “Ohh” syllables was performed on 143 and 142 patients, respectively. The developed AF indicator was reliable. Its numerical value decreased in sinus rhythm after the cardioversion, with “Ahh” falling from a mean of 13.98 to 7.49, and “Ohh” falling from a mean of 11.39 to 2.99. The values at sinus rhythm were more homogenous compared with AF as indicated by a lower SD.
The area under the receiver operating characteristic curve was greater than 0.98 and 0.89 for the “Ahh” and “Ohh” sounds, respectively (P < .001 for both). The AF indicator sensitivity was 95% and its specificity was 82%.
The researchers noted that the study was conducted in an intensive cardiac care unit and the results may not be replicable in everyday situations with different background noises.
“The research is now extended to use the developed indicator in an outpatient environment,” Golovchiner told Healio. “Further research will evaluate the ability of AF detection using voice analysis of free speech.”
The data adds to growing research using vocal sounds to predict CV risk. As Healio previously reported, an app used to make 30-second voice recordings to establish a “voice biomarker” score successfully predicted patients who were more likely to present to the ED with chest pain or be admitted for ACS.
For more information:
Gregory Golovchiner, MD, can be reached at doctor@rhythm-heart.com; Twitter: @greg10838698.
Collapse
Read more about