Original article
Voice Signal Characteristics Are Independently Associated With Coronary Artery Disease

https://doi.org/10.1016/j.mayocp.2017.12.025Get rights and content

Abstract

Objective

Voice signal analysis is an emerging noninvasive diagnostic tool. The current study tested the hypothesis that patient voice signal characteristics are associated with the presence of coronary artery disease (CAD).

Methods

The study population included 138 patients who were enrolled between January 1, 2015, and February 28, 2017: 37 control subjects and 101 subjects who underwent planned coronary angiogram. All subjects had their voice signal recorded to their smartphone 3 times: reading a text, describing a positive emotional experience, and describing a negative emotional experience. The Mel Frequency Cepstral Coefficients were used to extract prespecified voice features from all 3 recordings. Voice was recorded before the angiogram and analysis was blinded with respect to patient data.

Results

Final study cohort included 101 patients, of whom 71 (71%) had CAD. Compared with subjects without CAD, patients with CAD were older (median, 63 years; interquartile range [IQR], 55-68 years vs median, 53 years; IQR, 42-66 years; P=.003) and had a higher 10-year atherosclerotic cardiovascular disease (ASCVD) risk score (9.4%; IQR, 5.0-18.7 vs 2.7%; IQR, 1.6-11.8; P=.005). Univariate binary logistic regression analysis identified 5 voice features that were associated with CAD (P<.05 for all). Multivariate binary logistic regression with adjustment for ASCVD risk score identified 2 voice features that were independently associated with CAD (odds ratio [OR], 0.37; 95% CI, 0.18-0.79; and 4.01; 95% CI, 1.25-12.84; P=.009 and P=.02, respectively). Both features were more strongly associated with CAD when patients were asked to describe an emotionally significant experience.

Conclusion

This study suggests a potential relationship between voice characteristics and CAD, with clinical implications for telemedicine—when clinical health care is provided at a distance.

Section snippets

Study Population

The study population included a total of 166 patients who were enrolled between January 1, 2015, and February 28, 2017, including 129 patients who presented for coronary angiography, 22 apparently healthy control volunteers, and 15 control subjects who were referred to noncardiac procedures (including hernia operations, varicose vein procedures, and dermatologic and ophthalmologic procedures). We enrolled patients who were referred to the chest pain clinic and were not known to have preexisting

Results

Of 166 patients enrolled in the study, 28 (17%) patients had poor baseline voice recording due to background noise or multispeakers that did not allow voice feature extraction. Final study cohort included 138: 37 control subjects and 101 study subjects with available voice recordings who underwent a diagnostic coronary angiogram. Demographic, clinical, and laboratory data are summarized in Table 1. Median age of the study population was 61 years (IQR, 51-67 years) and 54 (54%) were men.

Discussion

The current study has 3 important observations. First, this is the first study to describe an association between voice characteristics and CAD: we identified 5 voice features that were associated with CAD. Voice analysis was performed in a blinded fashion, all voice features were prespecified, and CAD was confirmed with coronary angiogram. Second, the strongest association between voice and CAD was observed when patients were requested to record their voice while describing an emotional

Conclusion

This is the first study to suggest an association between voice characteristics and CAD. Voice features analysis holds the potential to assist physicians in estimating the pretest probability of CAD among patients presenting with chest pain, especially in the setting of telemedicine—when clinical health care is provided at a distance.

References (25)

  • A. Israel et al.

    Use of exercise capacity to improve SCORE risk prediction model in asymptomatic adults

    Eur Heart J

    (2016)
  • Y. Levanon et al.

    Method and system for diagnosing pathological phenomenon using a voice signal

    (2008)
  • Cited by (50)

    • Artificial intelligence in the catheterization laboratory

      2023, Intelligence-Based Cardiology and Cardiac Surgery: Artificial Intelligence and Human Cognition in Cardiovascular Medicine
    • Computerized analysis of speech and voice for Parkinson's disease: A systematic review

      2022, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      There were differences in crater and intensity features between voice signals of HC and people with heart disease [10]. In a study on the voices of 101 patients, Maor et al. showed a significant correlation between MFCCs features and coronary artery disease (CAD) [191]. In another study, MFCCs, pitch, formant, jitter, shimmer and loudness were found to be correlated with hospitalization and mortality rate among patients with congestive heart failure [192].

    • Noninvasive Voice Biomarker Is Associated With Incident Coronary Artery Disease Events at Follow-up

      2022, Mayo Clinic Proceedings
      Citation Excerpt :

      For all analyses, the type I error rate was .05 in a 2-sided test, and P values and CIs were calculated and presented at the 95% CI. The statistical analyses were performed using JMP 9 software15,17 (SAS Institute). The initial study population included a total of 117 participants who underwent baseline coronary angiography between January 1, 2015, and February 28, 2017.

    View all citing articles on Scopus

    For editorial comment, see page 818

    Potential Competing Interests: The authors report no competing interests.

    View full text