Introduction: Heart failure (HF) is a common reason for recurrent hospitalization. Daily smartphone-based recording of vocal biomarkers appears promising to remotely monitor patient well-being. The connection of self-reported quality of life and vocal biomarkers in patients with HF is unknown.
Methods: AHF-Voice is an ongoing monocentric prospective cohort study, focusing on voice alterations in patients hospitalized with acutely decompensated HF. Exclusion criteria are: listed for heart transplantation, high output HF, history of vocal fold disease/surgery, life-expectancy <6 months. Quality of life is assessed by the 23-item Kansas City Cardiomyopathy Questionnaire Overall Summary Score (KCCQ-OSS). Voice recordings (sustained vowels) are sampled iteratively using a specially developed smartphone app. The current analysis focuses on recordings collected at hospital admission and after 6 weeks of follow-up. Feature extraction of voice recordings were performed using python. Vocal biomarkers associated with changes in quality of life were identified using correlation and regression analysis.
Results: In this analysis, 72 patients hospitalized with acute HF were included (n=4 withdrew consent; n=6 excluded due to missing data). Hence, 62 patients were analyzed: mean age 76±10 years, 68% men, 81% in NYHA class III/IV, 47% de novo HF, mean LVEF 48±17%, 44% ischemic cause. Between admission and 6 weeks of follow-up, median KCCQ-OSS improved markedly, i.e. by +25 (4; 41) score points. Selected vocal biomarkers also changed materially (Table) and explained to a large degree the observed change in KCCQ over time: the vocal biomarker panel collected during admission explained 33%, whereas the vocal biomarker panel covering changes over time explained 39%, respectively. Overall, the vocal biomarkers explained 61% of the variance observed for change in KCCQ-OSS (Table).
Conclusion: In patients with acute HF, vocal biomarkers substantively explain the variation in quality of life. Tracking vocal biomarkers seems to be a promising low-barrier tool to non-invasively assess patient well-being and holds potential to effectively support heart failure management.
Table: Predictive utility of vocal biomarkers to explain changes in KCCQ-OSS.
| |
At hospital
admission
n=62
|
6-week
follow-up
n=62
|
Univariable R2 (%)
for changes in KCCQ
|
|
Admission
|
Δ admission
to 6 weeks
|
|
F1 mean, × 10²
|
6.8 (5.9, 7.7)
|
6.8 (6.0, 7.6)
|
6.5
|
12.1
|
|
F2 mean, × 10²
|
12.8 (11.2, 15.5)
|
12.1 (10.9, 14.1)
|
2.4
|
3.4
|
|
MFCC 2 median
|
5 (29)
|
1 (32)
|
3.2
|
3.5
|
|
MFCC 13 median
|
-1 (12)
|
-2 (14)
|
1.2
|
6.4
|
|
MFCC 13 mean
|
-1 (12)
|
-2 (13)
|
1.4
|
7.0
|
|
ΔMFCC 11 mean, × 103
|
-7 (-22, 9)
|
3 (-14, 20)
|
4.9
|
6.0
|
|
ΔMFCC 15 mean, × 10²
|
10 (3)
|
0 (3)
|
7.0
|
9.3
|
|
ΔMFCC 22 mean, ×103
|
5 (-18, 18)
|
2 (-11, 17)
|
4.9
|
6.0
|
|
ΔΔ MFCC 3 mean, × 103
|
-23 (-47, 16)
|
1 (-29, 46)
|
8.0
|
11.3
|
|
Zero-crossing median, × 10²
|
11.8 (8.3, 15.8)
|
13.6 (10.9, 16.6)
|
1.2
|
10.2
|
|
Zero-crossing mean, × 10²
|
11.9 (8.7, 15.9)
|
13.7 (10.8, 16.9)
|
0.8
|
3.3
|
|
F2 bandwidth mean, × 10²
|
10.2 (1.7)
|
9.8 (1.5)
|
10.7
|
14.3
|
|
Δ T0 mean, × 107
|
1 (-4, 8)
|
1 (-4, 4)
|
4.4
|
3.3
|
|
|
|
|
33
|
39
|
|
|
|
|
61
|
KCCQ-OSS was measured and vocal biomarkers were extracted both during hospitalization and after 6 weeks of follow-up. Data are mean (SD) or median (quartiles).
R2 = explained variance of vocal biomarker(s) for 6-week change in KCCQ-OSS, using either admission values, or 6-week changes, or the overall information. MFCC = Mel Frequency Cepstral Coefficients; KCCQ-OSS = Kansas City Cardiomyopathy Questionnaire Overall Summary Score.
