An open-label randomized controlled trial was designed to assess the utility of real-time remote auscultation using a Bluetooth-connected electronic stethoscope. Direct auscultation using a traditional stethoscope was used as control. To standardize and enhance the reliability of the assessment, we used a lung simulator for lung auscultation and a cardiology patient simulator for cardiac auscultation [10]. All sessions in this study were conducted at the skills laboratory at Dokkyo Medical University. As study participants, we recruited senior residents and faculty in the Department of Diagnostic and Generalist Medicine, as they were considered representative of general physicians working in community hospitals or clinics, the main population expected to perform auscultation in routine clinical practice. The only exclusion criterion was refusal to participate in this study. At baseline, no physicians exhibited hearing loss at their previous annual health checkup. This study was conducted in accordance with the current version of the Declaration of Helsinki. The study protocols were approved by the institutional ethics committee of Dokkyo Medical University, Tochigi, Japan (No. R-33-20J and R-37-19J). Written informed consent was obtained from all participants after a detailed explanation of the study and before participation.

For lung auscultation, the same lung simulator (MW28) was used in the tutorial and test sessions. This simulator was designed for medical education training and includes 34 samples of lung sounds recorded from actual patients and reproduced using a high-quality sound system. These lung sounds were classified, according to the American Thoracic Society classification system, as continuous (wheezes, rhonchi, or stridor) or discontinuous (fine or coarse crackles) [11]. Fine crackles were defined as “relatively high-pitched sounds usually heard at the end of inspiration as air enters the acinar unit.” Coarse crackles were defined as “the low-pitched, bubbling sounds that result from the accumulation of secretions in larger bronchi and trachea” [11]. The default respiratory rate of 15 breaths/minute was used during the study.

For cardiac auscultation, the same cardiology patient simulator (MW10) was used in the tutorial and test sessions. This simulator was designed for medical education training and includes 88 cases of cardiac sounds recorded from actual patients and reproduced using a high-quality sound system.

The electronic stethoscope is equipped with pressure-sensitive sensors, and the signals are converted into sound waves. It is also equipped with a volume regulator and a frequency filter. The filter has a bell mode, diaphragm mode, and wide mode, which enhance the 20-100 Hz, 200-2000 Hz, and 20-2000 Hz frequency bands, respectively. In the lung and cardiac parts, we used the diaphragm mode and the bell mode, respectively. The transmitter transferred the sounds from the lung simulator to the headset via Bluetooth (A2DP: Advanced Audio Distribution Profile).

Age, sex, and years since obtaining a degree in medicine were collected from all participants as baseline demographic data. All participants’ answers for each sound in the test session were collected. The primary outcome measure was the test score in each group. The rates of correct answers for each sound were the secondary outcome measures.

The correct answer in each group was compared using the Fisher exact test for primary and secondary outcome measures. Continuous variables for participant baseline characteristics are presented as medians (IQRs) and were compared using the Mann-Whitney U test. Categorical and binary variables for participant baseline characteristics are presented as numbers (percentages) and were compared using the Fisher exact test. A P value <.05 was considered statistically significant. All statistical tests were performed using R 3.6.0 for MacOS X (The R Foundation for Statistical Computing, Vienna, Austria).

In total, 20 physicians in the Department of Diagnostic and Generalist Medicine of Dokkyo Medical University were enrolled in the final analysis (Figure 3). Of these, 7 (35%) were senior residents (3-5 years after graduation), and 13 (65%) were faculty (≥6 years after graduation). The median age of all participants was 32 (IQR 8.5) years; 16 (16/20, 80%) were male, and the median time since graduation was 7.5 (IQR 7.5) years. In the lung part, 11 and 9 participants were assigned to the intervention and control groups, respectively. There were no statistically significant differences in participant age (P=.25), sex (P=.82), and years since graduation (P=.15) between the groups (Table 1). In the cardiac part, 6 and 14 participants were assigned to the intervention and control groups, respectively. There were no statistically significant differences in participant age (P=.99), sex (P=.99), and years since graduation (P=.78) between the groups (Table 1).

Test scores and rates of correct answers for each lung sound are summarized in Table 2. The total combined test score was 80/110 (72.7%) in the intervention group and 71/90 (78.9%) in the control group, with no differences between the groups (P=.32). There were no differences between the groups for normal lung sounds, wheezes, rhonchi, coarse crackles, fine crackles, or stridor. Only 3/11 (27%) participants in the intervention group correctly auscultated pleural friction rubs, whereas 7/9 (78%) in the control group did (P=.03).

Details of the answers for lung auscultation are supplied in Table 3 and Table 4. During exploratory data analysis, we determined that 4/22 (18%) instances of normal lung sounds were assigned as pleural friction rubs in the intervention group. On the other hand, 5/11 (45%) instances of pleural friction rubs were assigned as normal lung sounds in the same group.

Test scores and rates of correct answers for each cardiac sound are summarized in Table 5. The total combined test score was 50/60 (83.3%) in the intervention group and 119/140 (85.0%) in the control group, with no differences between the groups (P=.77). There were no differences between the groups for normal cardiac sounds, S₃, aortic stenosis, aortic regurgitation, mitral stenosis, mitral regurgitation, and atrial fibrillation. Although not statistically significant, there was over a 30% difference in the score for mitral stenosis between the 2 groups: While 5 of 6 (84%) participants in the intervention group correctly auscultated, only 7 of 14 (50%) participants in the control group did (P=.19).

Details of the answers in the remote cardiac auscultation group are provided in Table 6. Misinterpretations between normal cardiac sounds and S₃ sounds were frequently observed in the intervention group: 2 of 12 (17%) instances of normal cardiac sounds were assigned as S₃ sounds. On the other hand, 4 of 12 (33%) instances of S₃ sounds were assigned as normal cardiac sounds. Compared to the intervention group, participants in the control group made fewer misclassifications between normal cardiac sounds and S₃ sounds (Table 7): Misclassifications of normal cardiac sound as S₃ sounds occurred in 2 of 28 (7%) sounds, and misclassifications of S₃ sounds as normal cardiac sounds occurred in 5 of 28 (17%) sounds. On the other hand, misinterpretations between mitral stenosis sounds and mitral regurgitation sounds were frequently observed in the control group: 6 of 14 (43%) instances of mitral stenosis sounds were assigned as mitral regurgitation sounds. In comparison, 4 of 28 (14%) instances of mitral regurgitation sounds were assigned as mitral stenosis sounds.

In this study, there were 4 main findings. First, using a simulator, we demonstrated that the utility of real-time remote auscultation using a Bluetooth-connected electronic stethoscope was comparable to that of direct auscultation using a classic stethoscope. From previous finding of lung auscultation, coarse crackles, fine crackles, wheezes, and stridor are useful for diagnosing bronchitis or pneumonia [12-14], interstitial pulmonary fibrosis or pneumonia [12,15-19], exacerbation of asthma or chronic obstructive pulmonary disease [12,19], and upper-airway obstruction [19,20], respectively. For cardiac auscultation, valvular cardiac diseases and irregular rhythm disease were not misclassified as normal cardiac sounds in real-time remote auscultation and classical groups. The correct detection of S₃ is also an essential skill in diagnosing various cardiac diseases such as congestive heart failure, ischemic heart disease, cardiomyopathies, myocarditis, cor pulmonale, high-output states, left-to-right intracardiac shunts, and complete atrioventricular block [21]. Therefore, the results of this study suggest that real-time remote auscultation would be useful for the diagnosis of these diseases. At the bedside, this would allow medical staff to screen patients for adventitious sounds at a distance, protecting themselves from infectious diseases such as COVID-19.

Second, the rate of correct answers for pleural friction rubs was lower in the real-time remote lung auscultation group than in the direct lung auscultation group. Therefore, it would be challenging to diagnose pleuritis with real-time remote lung auscultation [12].

Third, in the real-time remote lung auscultation group, we observed a trend for confusion of normal lung sounds and pleural friction rubs. Respecting pleural friction rubs, 45% were misclassified as normal lung sounds in the remote auscultation group in this study. According to the participants, placement of the electronic stethoscope to the surface of the lung simulator caused a bit of noise. Electronic stethoscopes are sensitive to electronic and ambient noise, and this placement noise may be the cause of the difficulty observed in auscultation of pleural friction rub.

Fourth, the rate of correct answers for mitral stenosis was higher in the real-time remote cardiac auscultation than in the direct cardiac auscultation group. According to the participants, the monitoring screen of the simulator was on the caudal side. The screen showed a heartbeat icon regardless of whether it was the systolic or diastolic phase without any waveform. In the direct cardiac auscultation group, it was difficult to watch a display with auscultation. On the other hand, in the real-time remote auscultation group, the participants could watch the screen to detect the systolic or diastolic phase with auscultation. This may have been the cause for the misinterpretations between mitral stenosis sounds and mitral regurgitation sounds in the direct auscultation group.

There were 3 strengths of note in this study. First, use of simulators allowed us to gather standardized data, reducing bias. Second, the direct comparison of a novel auscultation technology with classical auscultation as control adds value to this randomized controlled study. Third, all participants were physicians who specialized in general internal medicine. Thus, the study participants were representative of general physicians working in community hospitals or clinics, the main population expected to perform auscultation in routine clinical practice.

This study was a pilot study and had several limitations. First, the sample size was small and did not include real patient data. Fully powered trials need to be conducted to better show equivalence. There was grouping variation through simple randomization, especially in the cardiac part, with 6 in the intervention group vs 14 in the control group. This grouping variation may have affected the detection power in this study. Therefore, in future studies, the efficacy of real-time remote auscultation has to be confirmed at the bedside with a larger sample size. Second, as the researcher who placed the electronic stethoscope on the simulators could not hear the sounds while doing so, the timing of the change in auscultation position could not be adjusted for optimum auscultation. Participants in the classical auscultation group were able to change the auscultation position on their own, which may have given this group an advantage over the remote auscultation group. Third, the respiratory rate and phase of crackle could not be adjusted with the lung simulator. The default respiratory rate was slower than that of real patients with respiratory diseases. Therefore, the results from this study may not be generalizable to auscultation at the bedside. Fourth, in this study, the technique used for classical auscultation was not standardized. This may have limited the reproducibility of our results. Fifth, there might be some dependences among answers within-subject.

To the best of our knowledge, there has been no other study in which real-time remote auscultation and classical auscultation were directly compared. In terms of the accuracy of classical auscultation using the simulator, our results, which showed variable accuracy depending on the type of sounds, were consistent with previous studies [10,22]. In both previous studies, the total accuracies varied from 62% to 89.7%, which may have depended on the difficulty of tests. For example, one study that showed a low accuracy rate included 3 types of S₂, S₄, and S₃+S₄, whereas another study that showed a high accuracy rate only included S₂ split, mitral regurgitation, and aortic stenosis. Regarding the difficulty of testing, our study may be close to the previous 2 studies.

In a previous study of the accuracy of identifying lung sounds using classical auscultation, stridor was not included in the evaluation [10]. Stridor can be identified without a stethoscope. However, we included stridor in this study, as its detection with electronic stethoscopes was reported in other studies [23].

We are aware of 1 study reporting results regarding the utility of real-time remote auscultation [7]. In that study, the interobserver concordance of remote auscultation via the internet (93.7%-94.0%) was lower than that of direct auscultation (98.4%-98.9%), but not statistically significant. Although remote auscultation in that study was not conducted in real time, the results are similar to ours in terms of a slightly lower accuracy in the remote auscultation group than in the conventional auscultation group.

This study demonstrates that the utility of a Bluetooth-connected, real-time remote auscultation system is comparable to that of classical, direct auscultation, except for pleural friction rubs. Future studies focused on real-time auscultation through Wi-Fi or the internet are warranted. Furthermore, this study leads the way for further studies in real patients with fully powered trials. In a future study, visualized waves of sounds [24] or artificial intelligence [25] would be supported to detect abnormal sounds.