Advertisement
Mayo Clinic Proceedings Home

Hey Goglexiri, Do I Have Coronary Artery Disease?

      We want you to imagine what it was like for your parents or even grandparents 50 or 100 years ago when they needed a doctor. We picture our grandparents calling for a doctor on the “landline.” The doctor (if not already out on a house call) would drive to the house to see them in person, take a careful history, feel the pulse, listen for Korotkoff sounds, examine with a stethoscope, form a differential diagnosis, and cognitively try to deduce the most likely diagnosis. The physician would then rummage through his or her bag searching for medications that could help alleviate the symptoms and write a prescription to bring to the pharmacy. Now look ahead to the doctor-patient encounter 50 years from now. We imagine calling the doctor through a mobile device or even a television from the office or home. By the time we “see” the doctor, we have placed a temporary tattoo
      • Bandodkar A.J.
      • Wang J.
      Non-invasive wearable electrochemical sensors: a review.
      on our skin, and over 2-way video chat the history is taken. We then pull out a digital stethoscope,

      Fattah SA, Rahman NM, Maksud A, et al. Stetho-phone: low-cost digital stethoscope for remote personalized healthcare. Paper presented at: GHTC 2017: IEEE Global Humanitarian Technology Conference; October 19-22, 2017; San Jose, CA.

      and the doctor instructs us where to place it so that he or she can do a physical examination. While this is occurring, the temporary tattoo has determined our heart and respiratory rates, blood pressure, oxygen saturation, and temperature and has calculated our white blood cell count,
      • Bourquard A.
      • Pablo-Trinidad A.
      • Butterworth I.
      • et al.
      Non-invasive detection of severe neutropenia in chemotherapy patients by optical imaging of nailfold microcirculation.
      red blood cell count,
      • Edwards A.
      • Richardson C.
      • Van Der Zee P.
      • et al.
      Measurement of hemoglobin flow and blood flow by near-infrared spectroscopy.
      and glucose level.
      • Bauer A.
      • Hertzberg O.
      • Küderle A.
      • Strobel D.
      • Pleitez M.A.
      • Mäntele W.
      IR-spectroscopy of skin in vivo: optimal skin sites and properties for non-invasive glucose measurement by photoacoustic and photothermal spectroscopy.
      Armed with this information, artificial intelligence (AI) will note patterns and correlate symptoms, guiding the doctor to a diagnosis and a treatment plan; drones will complete the visit by delivering medications.
      Seems far-fetched, right? But you may be surprised how close we are to this scenario. At this moment, for a multitude of reasons, the medical community, and specifically cardiology, is witnessing exponential progress and amalgamation of telemedicine, AI, and noninvasive diagnostics. Telemedicine is thriving because of both resource need and patient preference.
      • Flodgren G.
      • Rachas A.
      • Farmer A.J.
      • Inzitari M.
      • Shepperd S.
      Interactive telemedicine: effects on professional practice and health care outcomes.
      • Ballester J.M.S.
      • Scott M.F.
      • Owei L.
      • Neylan C.
      • Hanson C.W.
      • Morris J.B.
      Patient preference for time-saving telehealth postoperative visits after routine surgery in an urban setting.
      It has enabled health care delivery to rural and underresourced areas and has allowed those with access to resources to see the doctor with minimal day-to-day interruption. In parallel, AI has already integrated impressively into our everyday lives, whether through searches on Google, Netflix suggestions for television shows or movies, or thermostats in our houses. Artificial intelligence is an incredibly powerful tool, with what seems like limitless capabilities at times, and it has already shown tremendous promise in medicine.
      • Krittanawong C.
      • Zhang H.
      • Wang Z.
      • Aydar M.
      • Kitai T.
      Artificial intelligence in precision cardiovascular medicine.
      In addition to telemedicine and AI, the rise in and development of noninvasive diagnostic testing has also been rapid, driven by physicians' desire to minimize risk and harm to the patient as well as patient preference for noninvasive diagnostics.
      In this issue of Mayo Clinic Proceedings, Maor et al
      • Maor E.
      • Sara J.D.
      • Orbelo D.M.
      • Lerman L.O.
      • Levanon Y.
      • Lerman A.
      Voice signal characteristics are independently associated with coronary artery disease.
      provide a glimpse of that future and the amalgamation of these intersecting platforms that will shape the future of health care delivery. These authors used voice (telecommunications) and AI (although they did not explicitly use AI, the discovery of the voice parameters implemented in this article did) to determine the presence of coronary artery disease (CAD) noninvasively. In this article, the authors recorded and analyzed the voices of 138 patients (37 controls and 101 patients who underwent coronary angiography) to assess whether there was an association between voice characteristics and CAD. The voice was recorded before any intervention in three different formats: reading a text and describing a positive and a negative emotional experience. The voice data were analyzed for multiple features of voice intensity and frequency with the use of mel frequency cepstral coefficients, and 81 prespecified voice features were extracted. The primary end point of this study was the presence of CAD, as detected by angiography, with further subclassification into mild (≥30% stenosis), moderate (≥50% stenosis), and severe (≥70%). Control subjects consisted of 2 groups: healthy volunteers who did not undergo any procedures (n=22) and patients free of known CAD who underwent noncardiac procedures (n=15). The study demonstrated that 5 of the extracted voice features were associated with CAD on univariate analysis. A multivariate model featuring age, sex, and other traditional cardiovascular risk factors concluded that 2 voice features (features 43 and 71) were independently associated with CAD. Notably, this association was significant only when patients were asked to describe an emotionally significant experience.
      Although there is no direct hypothesis, the authors are to be commended for uncovering an unexpected insight regarding what may be a new noninvasive method for diagnosis of CAD. Although the patient numbers are small, the large numbers used in previous voice analysis studies provide credible background and support that the observations of Maor et al are likely to be true. That said, there are a few elements of the article that require further discussion and thought. First, the voice analysis is complex and sophisticated. It is so far removed from normal clinical intuition that health care providers have to entirely depend on data-driven conclusions because of our inability to intuitively understand or interpret the core data itself or outliers. This absence of intuition means that implementation in clinical practice will require large volumes of data to concretely prove that the observed finding is a real effect. Furthermore, the control group is small and is enrolled from mixed methods: healthy volunteers who did not undergo angiography and patients who were known to be free of CAD and underwent noncardiac procedures. It is unclear whether control subjects were age and sex matched or controlled for other comorbidities and baseline demographic characteristics. Prospective control subjects (potentially sham controls) will be necessary for future studies. However, it should be underscored that the analysis was objective and blinded as the computer was unaware of the underlying diagnosis, bringing essential objectivity to the study. Last, it is unclear what prompted angiography in these patients. For example, were patients aware that they could have CAD? Were they given any medication preprocedurally, for example, aspirin, statin, β-blocker, or nitroglycerin? The psychological influence and potential for changes in a person’s voice from either medication or awareness that they have or are likely to have CAD may confound some of the analyses. Although this may be difficult to control for, clearly selected, prospectively matched controls may help in this regard.

      The Vagus Nerve and the Heart

      The creation and determination of voice result from a complex interaction among many elements, and there are biologically plausible associations that can support the findings of Maor et al. Simplistically, voice is created by 3 major components: the lungs (which provide the air to create sound), the larynx (which contains the vocal cords and other vital muscles), and the articulators (eg, the tongue, the palate, and the mouth muscles). Most of voice creation and control is subconscious. We generally are not consciously thinking about all the changes we have to make to our lungs, larynx, or articulators to produce or alter our voice. The nervous system is responsible for this subconscious control, and although there are many nerves involved in voice creation (cranial nerves 5, 7, 11, 10, 12), it is cranial nerve 10 (the vagus nerve) that catches our attention, especially considering its close relationship to the heart. The importance of this nerve in voice control is evident in patients with refractory epilepsy who have vagal nerve stimulators implanted: the most described consequence (complication) of such stimulators is voice change. The vagus nerve is critical for autonomic control, and it innervates the heart by the superior, inferior, and thoracic cardiac branches. Changes in the vagus nerve can cause fluctuation in the heart rate, often referred to as heart rate variability. Heart rate variability and CAD have a well-established relationship.
      • Tsuji H.
      • Larson M.G.
      • Venditti F.J.
      • et al.
      Impact of reduced heart rate variability on risk for cardiac events: the Framingham Heart Study.
      • Liao D.
      • Cai J.
      • Rosamond W.D.
      • et al.
      Cardiac autonomic function and incident coronary heart disease: a population-based case-cohort study: the ARIC Study.
      With this relationship in mind, it is not surprising that voice and CAD have a relationship, and it provides some biological basis for the underlying findings of this study.

      Where to Next?

      The future of voice analysis and human health is teeming with countless opportunities. Because all telecommunications today are generally digital, voice analysis could be easily integrated into current technological platforms (such as a smartphone) with either analysis by software on the platform or transmission of digital voice recordings to a central procession area. Artificial intelligence has the potential to learn your voice and its myriad variations and thereby determine whether substantial and subtle changes may correlate with disease that is either subacute or acute. There has also been recognition that the digital electrocardiogram can provide a substantial amount of noninvasive data, such as determining electrolyte levels
      • Attia Z.I.
      • DeSimone C.V.
      • Dillon J.J.
      • et al.
      Novel bloodless potassium determination using a signal-processed single-lead ECG.
      or even risk of cardiac events in selected populations.
      • Sugrue A.
      • Noseworthy P.A.
      • Kremen V.
      • et al.
      Identification of concealed and manifest long QT syndrome using a novel T wave analysis program.
      • Sugrue A.
      • Killu A.M.
      • DeSimone C.V.
      • et al.
      Utility of T-wave amplitude as a non-invasive risk marker of sudden cardiac death in hypertrophic cardiomyopathy.
      The combination of these elements creates a potent tool that is entirely noninvasive and provides an unbiased method to monitor cardiac health. The value of this technology is understated, and it could, indeed, have an impact on vulnerable populations, such as babies or the elderly, that may be limited in the voicing of symptoms. In babies, for example, a crib could measure heart and respiratory rates and analyze crying.
      Maor et al are to be congratulated for the stimulating findings in their manuscript and testing of a genuinely novel noninvasive method for the diagnosis of CAD. This article gives us a glimpse of the future, clarifies the potential power of the human voice in health, and highlights many future avenues that could transform health care.

      References

        • Bandodkar A.J.
        • Wang J.
        Non-invasive wearable electrochemical sensors: a review.
        Trends Biotechnol. 2014; 32: 363-371
      1. Fattah SA, Rahman NM, Maksud A, et al. Stetho-phone: low-cost digital stethoscope for remote personalized healthcare. Paper presented at: GHTC 2017: IEEE Global Humanitarian Technology Conference; October 19-22, 2017; San Jose, CA.

        • Bourquard A.
        • Pablo-Trinidad A.
        • Butterworth I.
        • et al.
        Non-invasive detection of severe neutropenia in chemotherapy patients by optical imaging of nailfold microcirculation.
        Sci Rep. 2018; 8: 5301
        • Edwards A.
        • Richardson C.
        • Van Der Zee P.
        • et al.
        Measurement of hemoglobin flow and blood flow by near-infrared spectroscopy.
        J Appl Physiol. 1993; 75: 1884-1889
        • Bauer A.
        • Hertzberg O.
        • Küderle A.
        • Strobel D.
        • Pleitez M.A.
        • Mäntele W.
        IR-spectroscopy of skin in vivo: optimal skin sites and properties for non-invasive glucose measurement by photoacoustic and photothermal spectroscopy.
        J Biophotonics. 2018; 11: e201600261
        • Flodgren G.
        • Rachas A.
        • Farmer A.J.
        • Inzitari M.
        • Shepperd S.
        Interactive telemedicine: effects on professional practice and health care outcomes.
        Cochrane Database Syst Rev. 2015; : CD002098
        • Ballester J.M.S.
        • Scott M.F.
        • Owei L.
        • Neylan C.
        • Hanson C.W.
        • Morris J.B.
        Patient preference for time-saving telehealth postoperative visits after routine surgery in an urban setting.
        Surgery. 2018; 163: 672-679
        • Krittanawong C.
        • Zhang H.
        • Wang Z.
        • Aydar M.
        • Kitai T.
        Artificial intelligence in precision cardiovascular medicine.
        J Am Coll Cardiol. 2017; 69: 2657-2664
        • Maor E.
        • Sara J.D.
        • Orbelo D.M.
        • Lerman L.O.
        • Levanon Y.
        • Lerman A.
        Voice signal characteristics are independently associated with coronary artery disease.
        Mayo Clin Proc. 2018; 93: 840-847
        • Tsuji H.
        • Larson M.G.
        • Venditti F.J.
        • et al.
        Impact of reduced heart rate variability on risk for cardiac events: the Framingham Heart Study.
        Circulation. 1996; 94: 2850-2855
        • Liao D.
        • Cai J.
        • Rosamond W.D.
        • et al.
        Cardiac autonomic function and incident coronary heart disease: a population-based case-cohort study: the ARIC Study.
        Am J Epidemiol. 1997; 145: 696-706
        • Attia Z.I.
        • DeSimone C.V.
        • Dillon J.J.
        • et al.
        Novel bloodless potassium determination using a signal-processed single-lead ECG.
        J Am Heart Assoc. 2016; 5: e002746
        • Sugrue A.
        • Noseworthy P.A.
        • Kremen V.
        • et al.
        Identification of concealed and manifest long QT syndrome using a novel T wave analysis program.
        Circ Arrhythm Electrophysiol. 2016; 9: e003830
        • Sugrue A.
        • Killu A.M.
        • DeSimone C.V.
        • et al.
        Utility of T-wave amplitude as a non-invasive risk marker of sudden cardiac death in hypertrophic cardiomyopathy.
        Open Heart. 2017; 4: e000561

      Linked Article