The present study design was based on the Cardiovascular Second Prevention Center (CSPC) of the Kitasato University Hospital. Patients who were admitted to this hospital due to acute CVD sought care at the CSPC outpatient clinic to prevent CVD recurrences every other year of their own volition. From this hospital-based cohort in which stableness and chronicity of CVD were ascertained, a cross-sectional single-center exploratory study was designed to assess the clinical usefulness of quantitative HS analyses.

An adequate number of participants were pre-calculated in previous studies describing an association of HF in the general population and patients with CVD (8, 9). Six to eight patients with HF are minimally required for the statistical analysis; however, we expected that forty patients with CVD would be needed for the present analysis. The Kitasato University Hospital Research Ethics Committee approved the present study protocol (Clinical Trial Registration No. B22-060). Between 4 August 2023 and 6 October 2023, 40 consecutive outpatients, aged between 46 and 87 years with chronic CVD, were enrolled after receiving written informed consent from each patient. In the present participants, S₃ and S₄ were not audible at the outpatient clinic. The mean duration from their discharge of acute illness to the present CSPC visit was 8.9 ± 6.3 years.

All participants underwent comprehensive clinical evaluations that included medical history updates, blood pressure measurements, electrocardiogram (ECG), echocardiography, chest x-ray, and blood tests on the same day. Blood collection for N-terminal pro-brain natriuretic peptide (NT-proBNP) was also performed during ordinary laboratory tests between 8:30 a.m. and 10:30 a.m. or 1:00 p.m. and 2:20 p.m.

HS collection was performed on the same day or within 17.0 ± 8.1 days, in between a tight schedule of the regular health insurance services using the handheld tri-antler chest piece (BMP-LD3, NTT Basic Research Laboratories, Japan, Figure 1A). This device has four microphones. Each microphone is placed inside the chest piece in each antler to record HS, and the fourth one is inside the housing to record ambient sounds (Figure 1B). BMP-LD3 was designed to attach three chest pieces firmly and easily to the anterior chest wall at the sitting or supine position of the participant per se. When the device was attached to the anterior chest, two chest pieces were perpendicularly placed on the second intercostal space (heart base) and the fourth intercostal space, along the left sternal border (Figure 1C). The last third one was at the apex (Figure 1D). Each patient sat freely in the daily examination room. Sounds from four microphones were recorded for at least 10 s. HS collected through the handheld tri-antler chest piece (BMP-LD3) were saved digitally at a rate of 16 kHz in a memory inside the connecting housing of BMP-LD3 (Figure 1A) (10). The block diagram of the procedure for S₁–S₄ extractions and their analyses is presented in Supplementary Figure S1.

Collected HS were analyzed by the audio editing and recording software (Audacity 3.3.3; Muse Group, Cyprus). Audacity, a sound recording software that is freely available, was used in the study for recording and analyzing HS (11, 12); signals captured through each chest piece were amplified by 30 dB, which was then simply subtracted by the level of the ambient sounds. Although the first (S₁) and second sounds (S₂) were easily recognized in all cases on the phonocardiogram, S₃ and S₄ were noted as tiny sound vibrations (Figure 2). To save S₃ and S₄ correctly, the signals at the fourth intercostal space chest piece along the left sternal border were chosen, and the other two signals at the base and apex chest pieces were used for identifying the timings of S₁, S₂, S₃, and S₄. We defined sound vibrations during protodiastole as S₃ equivalent (eS₃) and those during presystole as S₄ equivalent (eS₄). S₁ is mostly constructed by three peaks from the baseline (13, 14). S₂ is constructed by two peaks from the baseline (15, 16). From the end of S₂ to the start of eS₃, there is a sound-free pause as the early systolic baseline (17). Then, eS₃ was defined as shown in Figure 2. eS₄ is from the end of the mid-systolic baseline to the start of S₁ (2, 4). Thus, the range length was different between the first beat and the second one.

We chose two consecutive beats, which were recorded with clear ECG and with the absence of respiratory noise. On these two beats, the average and their mathematical difference (Δ) of intensity (dB) and audio frequency (Hz) were taken for the analysis of the clinical characteristics of patients. The RR interval was measured by an ECG attached to the device. All measurements were performed by an independent clinical laboratory technologist who was blinded to the clinical features of each patient.

HF and high-risk HF were defined as NT-proBNP ≥ 300 and ≥900 pg/ml, respectively (9). HF was also categorized on the levels of ejection fraction (EF). HF with reduced EF (HFrEF), mildly reduced EF (HFmrEF), and preserved EF (HFpEF) were defined as left ventricular ejection fraction (LVEF) <40%, 40%–50%, and ≥50%, respectively (18). Ischemic heart disease (IHD) included myocardial infarction, angina pectoris, and vasospastic angina. Atrial fibrillation (AF) was defined as absolute arrhythmia confirmed by ECG. The parameters of the echocardiogram follow the international guidelines for echocardiography (19). Valvular heart diseases were diagnosed by auscultation and echocardiography. The ECG wave graph and explanation of Q, R, and S waves at V1 and V5 of the ECG chest electrode are shown in Supplementary Figure S2. QRS duration was measured from the beginning of the Q wave to the end of the S wave (20). SV1 + RV5 voltage was defined as the sum of the S wave amplitude in V1 and the R wave amplitude in V5 (21). Cardiothoracic ratio (CTR) was defined as the ratio of the transverse dimension of the heart to the chest width measured on a chest radiograph of posteroanterior view (22).

Normally distributed continuous variables are expressed as mean ± standard deviation, whereas non-normally distributed values are expressed as median with interquartile range and were analyzed using Wilcoxon signed-rank test. Continuous variables were analyzed using unpaired t-tests. Categorical variables are reported as number (%) and were analyzed using the chi-squared test. Receiver operating characteristic (ROC) curves were constructed to assess the predicted occurrence of HF and high-risk HF and to determine the best cutoff of the sound intensity for detecting HF and high-risk HF. A multivariate regression analysis was performed to identify the factors associated with the presence of HF or high-risk HF among variables with values of p < 0.050 on the univariate logistic regression analysis. Statistical significance was defined as p < 0.050. JMP version 9.0 software (SAS Institute, Cary, NC, USA) was used to perform all statistical analyses.

The mean age of the participants was 71.6 ± 9.4 years, with males comprising 73% of the group and an average body mass index (BMI) of 24.2 ± 3.7 kg/m². NT-proBNP was 156 (92.8–464) pg/ml, and laboratory tests including ECG, chest X-P, and echocardiography were within or close to normal levels (Table 1). The number of preserved EF was 23, 7, and 4 cases in non-HF (n = 27), HF (n = 8), and high-risk HF (n = 5), respectively. Beta-blockers were prescribed to 35 of 40 patients (85%) and to 11 of 13 patients with HF (85%) (Supplementary Table S1).

Distributions of each sound intensity (dB) and frequency (Hz) and a difference between two beats in intensity and frequency of four components are shown in Figure 3. HS intensities of all four components (S₁–eS₄) were inversely related to BMI (p < 0.02–p < 0.001, Table 1, Supplementary Figure S3), and eS₄ was inversely related to CTR. There was no correlation between HS intensities and NT-proBNP (Supplementary Figure S4). The audio frequency of S₁ (fS₁) was positively related to BMI and CTR (BMI, p = 0.043; CTR, p = 0.010; Supplementary Table S2). There was no correlation between HS frequencies and NT-proBNP (Supplementary Figure S5). ΔS₁ positively correlated with NT-proBNP (p = 0.001) and was inversely correlated with deceleration time in echocardiography (p = 0.035). On the other hand, ΔfS₁ was positively correlated with NT-proBNP and CTR, respectively (NT-proBNP, p = 0.033; CTR, p = 0.018). As summarized in Table 2, intensities of S₁, S₂, eS₃, and eS₄ were not different in the presence or absence of HF, while ΔS₁ (p = 0.024) and ΔfS₁ (p = 0.042) between two consecutive beats were larger in patients with HF than that in patients without HF. These findings were observed where a preceding R–R interval was not different between HF and non-HF. ΔS₁ (p = 0.012) and ΔfS₁ (p = 0.012) were larger in patients with high-risk HF and were also correlated with AF (p = 0.001) or with E/e′ (echocardiographic parameters, p = 0.017, Table 3). Multivariate analysis showed that ΔS₁ was an independent factor from ΔfS₁ to predict HF (ΔS₁, OR = 1.659, p = 0.044). A group of high-risk HF was correlated significantly with AF (AF, p < 0.001, ΔS₁, p = 0.036, and ΔfS₁, p = 0.011, Supplementary Table S3). Multivariate analysis showed that ΔfS₁ was an independent factor from ΔS₁ to predict high-risk HF (ΔfS₁, OR = 1.690, p = 0.027). AF was a strongly independent factor from HS to predict HF or high-risk HF (Supplementary Table S4). The areas under the curve (AUC) of the ROC for the detection of high-risk HF assessed by ΔS₁ and ΔfS₁ were 0.706 and 0.843 (ΔS₁, p = 0.047; ΔfS₁, p = 0.031, Figure 4). The best cutoff values of high-risk HF for ΔS₁ and ΔfS₁ were 4.0 dB and 5.0 Hz, respectively (ΔS₁, sensitivity 40.0%, specificity 81.1%; ΔfS₁, sensitivity 100%, specificity 68.7%). However, the AUC of the ROC for the detection of HF assessed by ΔS₁ and ΔfS₁ was not significant (Supplementary Figure S6).

The incidence of HF was significantly higher in patients with ΔS₁ ≥ 4.0 dB than those with ΔS₁ < 4.0 dB (100 vs., 27.0%, p = 0.029). The incidence of HF was also significantly higher in patients with ΔfS₁ ≥ 5.0 Hz than that of those without (56.3 vs. 16.7%, p = 0.015). Only one participant met both ΔS₁ ≥ 4.0 dB and ΔfS₁ ≥ 5.0 Hz, and the incidence of HF tended to be high although the statistical difference was not performed (Figures 5A,B). The incidence of high-risk HF was significantly higher in patients with ΔS₁ ≥ 4.0 dB than those without (66.7 vs. 8.1%, p = 0.036). The incidence of HF was also significantly higher in patients with ΔfS₁ ≥ 5.0 Hz than those without (31.3 vs. 0.00%, p = 0.007). Among the three groups according to the NT-proBNP, ΔS₁ and ΔfS₁ in patients with NT-proBNP over 900 pg/ml were significantly larger than those in patients with NT-proBNP below 300 pg/ml (ΔS₁, p = 0.004; ΔfS₁, p = 0.011, Supplementary Figure S7) although statistical significance was not observed between the other groups.

There are many aspects in quantifying the clinical features of HS. For describing physical signs, physicians documented semiquantitatively by loudness, pitch, and tone with verbal expressions through the perception of the physician (23). A phonocardiogram is powerful in determining the timing of sound occurrence in relation to ECG; however, how to compare patients' variations remains unexplored. Reddy et al. (24) applied a calibrated pressure mechanocardiograph to measure sound strength, but pitch or tone was not measured. To overcome subjective differences in audibility or visual recognition of phonocardiogram, we introduced a digital acquisition and processing of HS.

Physical appropriateness of our auscultatory device, handheld tri-antler chest piece (BMP-LD3) was supported firstly by assessing the influence of BMI on the HS and secondly by recordings of tiny and barely audible vibrations around the timings of S₃ and S₄. It is known that the acoustic intensity of HS declines by the distance between the heart and the thorax surface, the volumes of the lung (25), and the amount of other sound absorbers such as connective, adipose, and muscle tissues (26). Luo et al. demonstrated a significant intensity drop in HS recordings in cases with BMI > 30 kg/m², when physicians applied smartphones in the hospital for auscultation (27). In the present study, we observed an inverse relation between the intensity of HS and the levels of BMI (BMI between 16.2 and 33.0 kg/m², n = 40, aged from 46 to 87 years). In addition, HS intensity in aged patients was weak (24), and stored data were augmented in the digital analysis in the present study. Accordingly, we understand that the present device has sufficient resolution to quantify HS acoustically.

HS have been accepted as a cardiac biomarker in patients with HF (28). To detect occurrences of HF, wearable devices such as a smartwatch or a smartphone (29, 30) are available along with an implanted long-term (31) and continuous monitoring device to observe ECG and cardiac sounds for the management of HF (32). Artificial intelligence analysis of ECG and phonocardiogram was challenged to predict the occurrence of HF (33–35). These studies focused on the presence of diastolic HS such as S₃ or S₄; however, S₃ or S₄ was not evident in the present participants; they were all ambulatory and did not exhibit overt HF upon visiting our outpatient clinic, and 13 of them (32.5%) had HF defined by NT-proBNP level ≥ 300 pg/ml.

Although HF in the present participants was hemodynamically and clinically mild and 85% of HF participants were under the control of beta-blockades, the quantitative sound analysis revealed that (1) a beat-to-beat alteration of S₁ intensity (dB) and its audio frequency (Hz) was significantly associated with the level of NT-proBNP and (2) the size of CTR correlated negatively with eS₄ intensity (p = 0.035) and positively with S₁ audio frequency (p = 0.010) (Table 1). In addition, highly significant differences in ΔS₁ intensity and frequency were noted between HF and non-HF. Since such peculiar findings were not observed in S₂, eS₃, or eS₄, these findings may suggest a latent cardio-mechanistic derangement in HF.

In previous case reports in patients with severe HF, S₁ was documented to be soft and muffled because of congestive blood flow and poor heart pumping (13, 14). S₁ appears during the isovolumic contraction when the left ventricular (LV) dP/dt reaches its peak level. A tight relationship between the maximum positive LVdP/dt and the amplitude of S₁ was repeatedly demonstrated in animal studies (36, 37). The level of peak LV dP/dt has been accepted as a standard index of contractility (38). Then, the relationship between S₁ sound characteristics and LV contractility may be described by a simple first-order system. However, a highly accurate pressure measurement was not realistic for participants in the outpatient clinic. To prove how well a simple first-order system fits to the relationship between S₁ alteration and LV contractility remains one of the important future studies.

In the Framingham Heart Study, AF occurs in more than half of individuals with HF, and HF occurs in more than one-third of individuals with AF (39). HF and AF also frequently coexist. AF precedes and follows HF with both preserved and reduced EF (40). A previous study also reported that AF is highly predictive of underlying HFpEF (41). H₂FPEF score was calculated by clinical variables including heavy, hypertensive, AF, pulmonary hypertension, and elder and filling pressure. Subsequently, AF accounts for 3 of 9 points in the H₂FPEF score (42). HFA-PEFF score was also calculated by functional, morphological, and biomarker domains. AF is a main factor in the initial evaluation of the HFA-PEFF diagnostic algorithm (43), although both scores have good sensitivity and specificity to detect HFpEF (44). In the present study, AF was noted in 11 of 13 participants with HF and they were classified as HFpEF. Equal RR interval in these participants may indicate dysfunction of the autonomic nervous system (45). Accordingly, a beat-to-beat S₁ alteration may represent mechanisms similar to pulsus alternans (46).

An audible range of humans is at least ≥0 dB at the audio frequencies between 20 Hz and 20 kHz (47). In human ears, the audible minimum change of sound pressure is at least ≥2 dB (23) although the audible threshold worsens according to age. In the present study, the detected sound signal was augmented electrically, which enabled us to recognize slight changes in HS on PCG recordings. In the present augmented signals, absolute differences in S₁ intensity and its peak audio frequency between consecutive beats were 4.0 dB and 5.0 Hz, and such tiny changes were assessable due to the digital quantification of sound signals.

There are several limitations: (1) the number of participants was limited in the present study; however, subjects from the real world of daily clinical activities provide the present research eliminating biases in the selection of participants, (2) participants with the absence of CVD were not included, (3) we did not study the difference between physician's audibility and digital acquisition of HS, and (4) since the present study is based on a small number of participants, the contribution of this single-center exploratory study remains a hypothesis generation for future further studies.