Vocal Biomarker Predicts Hospitalization and Mortality among Heart Failure Patients

Aim: This study aimed to develop a non-invasive vocal biomarker and evaluate its association with adverse outcome among CHF patients.

Methods and results: Study cohort included 10,467 patients who were registered to a call center of patients suffering from chronic conditions including CHF in Israel between June 2013 and October 2018. A total of 223 acoustic features were extracted from 20 seconds of speech for each patient. A biomarker was developed based on a training cohort of non-CHF patients (N=8,200) using machine learning techniques. The biomarker was tested on a mutually exclusive CHF study cohort (N=2,267) and was evaluated as a continuous and dichotomized variable (higher quartile vs. lower three quartiles). Mean age of the CHF study population was 75 (IQR 68-83) and 63% were men. During a median follow-up of 20 months (IQR 9-34), 824 (36%) patients died. Kaplan Meier survival analysis showed that the cumulative probability of death among high biomarker group was significantly higher (64%±14% vs. 56%±15%, p<.001). Multivariate Cox regression analysis with adjustment to known predictors of poor survival, demonstrated that each standard deviation increase in the biomarker was associated with a significant 28% increased risk of death during follow-up (95% CI 1.37 - 1.78, p < .001). The model consistently demonstrated an independent association of the biomarker with hospitalizations during follow-up (P<.001).

Conclusion: This is the first study to document a relationship between a non-invasive vocal biomarker and adverse outcome among CHF patients. The results have important clinical implications for telemedicine and CHF patient care.