Smartphone voice analysis can identify lung congestion in acute decompensated HF

According to a study published in JACC: Heart failure.

“Most patients with HF present to the hospital with fluid retention, which manifests as worsening dyspnea caused by pulmonary edema” Amir, MD Offer, director of the Heart Institute at Hadassah Medical Center in Jerusalem, Israel, and colleagues wrote. “Because pulmonary congestion is not only the main factor in hospitalization for heart failure, but also a major predictor of poor post-discharge outcomes, frequent monitoring of pulmonary congestion has been proposed as a way to keep patients. healthy and out of hospital. “

Researchers assessed whether speech measurements were indicative of specific clinical states of pulmonary congestion in patients with decompensated acute heart failure using a new speaker verification, processing and analysis app speech (HearO, Cordio Medical or Yehuda). They recruited 40 adult patients with acute decompensated HF who were asked to record five sentences, repeated three to four times each in their native language (Hebrew, Arabic or Russian).

The recordings were collected on admission and on discharge from the hospital.

“The premise behind the HearO system is that the subtle physiological changes associated with HF decompensation affect the patient’s speech and make them a ‘different person’ (voice print),†the researchers wrote. “These changes are much more subtle than those found between different speakers, but are nonetheless detectable using algorithms derived from those used in verifying text-dependent speakers.”

Pulmonary congestion speech measurement

Researchers evaluated five unique measurements of speech that assessed different combinations of characteristics, including high temporal resolution, high spectral resolution, autoregressive model for the spectrum, nonlinear amplitude and frequency mapping, symmetric version of a nonlinear spectral ratio and a Euclidean distance.

A total of 1,484 records were analyzed.

According to the study, recordings taken on discharge from hospital were successfully identified as being significantly different from baseline in 94% of cases. In addition, in 87.5% of cases, distinct differences from baseline were detected in all five speech measurements.

In a separate analysis, the researchers collected 72 additional voice recordings from nine patients, which were then reanalyzed using the blinded smartphone algorithm. According to the study, the system isolated the records into two separate unknown sets that, when opened, 97.8% of cases successfully matched admission or discharge records.

Differences in the recordings made on admission to hospital versus discharge were found for all five speech measurements, with the largest observed difference (218%) being found using the second speech measurement, which included high temporal resolution, an autoregressive model for the spectrum, and a symmetric version of a nonlinear spectra ratio.

“Current observations have provided substantial proof of concept that this new automated approach to speech processing and analysis can reliably identify these differences between two states of pulmonary congestion in patients with HF at the time of hospitalization for acute decompensated CI and after comprehensive in-patient treatment, â€the researchers wrote. “In this context, this speaker verification-based concept has the potential to serve as a novel tool in the hospital and remote arsenal for the assessment of pulmonary congestion in patients with HF.”

“An important step forward”

In a related editorial, Neal G. Ravindra, PhD, postdoctoral fellow in the cardiovascular medicine section of the Yale School of Medicine, and David P. Kao, MD, associate professor of medicine (cardiology) at the University of Colorado’s Anschutz Medical Campus, explained how the study could pave the way for future research on this new method of detecting pulmonary congestion.

“Active speech analysis as described by Dr Amir et al is an important step towards expanding the tools available to assess patients with HF,” wrote Ravindra and Kao. “Although nascent, the use of commonly available mobile technologies suggests a potential for broader use compared to highly invasive strategies requiring dedicated hardware. Thorough development and validation is needed before clinical use, but success in a use case like HearO may pave the way for even more practical and generalizable strategies.

Reference: