Use of voice signal analysis, an emerging noninvasive biomarker, may identify patients with congestive HF who are at highest risk for 20-month hospitalization or mortality, using voice analysis, according to research published in the Journal of the American Heart Association.
Researchers studied a cohort of 10,583 patients who were registered to a telemedicine call center in Israel and had chronic health conditions including congestive HF. Using voice processing techniques (Vocalis Health), the researchers extracted 223 acoustic features from 20 seconds of audio from each patient. The researchers then developed a vocal biomarker that was based on a training cohort of patients who did not have congestive HF (n = 8,316), and then evaluated use of the vocal biomarker in a cohort of 2,267 patients with congestive HF (mean age 77 years; 63% men) who were categorized into four biomarker quartiles.
The primary outcome was all-cause mortality, which occurred in 36% of patients over a median follow-up of 20 months.
Kaplan–Meier survival analysis demonstrated that the cumulative probability of death was greater with increasing quartiles, and ranged from 23% in quartile one, to 29% in quartile two, to 38% in quartile three to 54% in quartile four (P < .001)
Moreover, for each standard deviation increase in the vocal biomarker, researchers reported a 32% greater risk for death during follow-up (HR = 1.32; 95% CI, 1.24-1.41). Patients in the highest quintile were 96% more likely to die compared with patients in the first quintile (HR = 1.96; 95% CI, 1.59-2.42).
In other findings, for each standard deviation increase in the vocal biomarker in the congestive HF cohort, researchers observed a nearly 25% greater risk for hospitalization during follow-up (HR = 1.21; 95% CI, 1.15-1.28).
“This is a very straightforward, noninvasive, transparent tool. Patients are recording themselves all the time. They're making phone calls; they're doing telemedicine visits with clinics, especially now with the COVID-19 pandemic; patients record their voice when they use Siri, Alexa or Google to do searches,” Elad Maor, MD, PhD, associate professor of cardiology at Sheba Medical Center and Tel-Aviv university and also a former fellow at Mayo Clinic in Rochester, Minnesota, with which this study was a collaborative effort, told Healio.
For this study, the 20-second voice recordings were collected via periodic phone conversations with nurses at the telehealth center. Maor and colleagues explained that low-level acoustic features were extracted from the samples, using the voice processor, and included mel-frequency cepstrum representation, pitch and formant measures, jitter, shimmer and loudness. These features were extracted via a resolution of 100 points per second and formed a matrix with 2000 columns. Next, further high-level features were extracted from the matrix, resulting in 223 valid features for each patients’ voice sample. Then, a machine learning linear model was built based on the training cohort and blinded to the congestive HF cohort.
“Physicians and cardiologists who deliver health care from a distance can use voice as an additional tool. At this point, we are not saying that voice can be used in order to diagnose disease as an independent tool, but it can be an additional tool for the physician with the main advantage that it's noninvasive and transparent to the patient.
“Looking at this from the clinical standpoint, with a patient who is discharged home and then registered to a telehealth center, voice could be used to help the nurse or physician to identify high-risk patients and then invest more resources for these patients. It could help in better allocating limited resources in the field of telemedicine,” Maor told Healio. – by Scott Buzby
For more information:
Elad Maor, MD, PhD, can be reached at email@example.com.
Disclosures: Maor reports that he is a consultant for Vocalis Health and several other authors report ties with Vocalis Health.