Discussion

This study showed that the ability of staff physicians, residents and medical students to describe five common lung sounds was similar across the groups and varied with the nature of the sound: it was high for the wheezes and the stridor, fair for the crackles and poor for the normal breath sound and the pleural friction rub. Second, the study showed that, for a given sound, whatever the group, the number of terms used by incorrect raters was almost three times that used by correct raters. Third, the study found that the terms used in roughly one-third of instances of complete intergroup agreement were incorrect. Finally, it showed that correct raters used the recommended terminology rather parsimoniously.

The high observer ability to identify the wheezes and the stridor was not surprising. By their long duration (typically >80–100 ms) tonal quality, and relatively high amplitude12–15 adventitious musical sounds are easier to recognise by the human ear than non-musical sounds of shorter duration (typically 5–15 ms) and lower amplitude like, for instance, the crackles.16 17

Several factors could have hampered the observers’ auscultatory skill. First and foremost was the approach to auscultation. Use by some observers of curious terms such as ‘cat meow’, ‘whistling’, ‘gurgling’ and ‘good air entry’ is reminiscent of the onomatopoeic approach used by Laennec,1 although less ingeniously. More rewarding could have been the approach proposed by Forgacs,12 whereby the auscultated sounds are related with other clinical parameters (eg, body position, phases of the respiratory cycle, site of auscultation and so on). By adopting it, our observers might have noticed that sound energy of the normal breath sound was essentially inspiratory, thus avoiding its mistaken classification—made by 17 observers—as bronchial breathing, which is usually perceived in the expiratory phase as well.18 19 Similarly, the description of the crackles could have been more precise had the observers considered that fine crackles are typically heard on mid-to-late inspiration, while coarse crackles tend to predominate on early inspiration,11 20 a feature unnoticed by the 134 observers classifying that sample. Finally, the performance in the identification of the pleural friction rub could have been improved, had the raters noticed that this sound (i) was biphasic, with the inspiratory and expiratory sequence of sounds mirroring one another; (ii) was of much higher amplitude than the crackles and (iii) had been recorded over the axillary region, three characteristic features of the pleural friction rub.11 20 Another factor that might have hampered the accuracy to identify the sounds was the fact that audio listening deprives the observers of the clinical context. Indeed, the possibility to auscultate over different chest regions, during different manoeuvres (eg, deeper inspirations) could have helped in the identification of certain sounds, especially the normal breath sound. A third and final factor was the possibility of artefacts—eg, hissing sounds or ambient sounds commonly heard in the clinical setting (eg, ventilators and so on)—that could have caused confusion. In this respect, our sound files had been filtered previously for the purpose of another publication.11 Furthermore, had they been present, artefacts could not entirely explain the results: for instance, while they could have caused the mistaken classification of the normal breath sound as crackles, they could hardly explain the classification of the crackles as normal sound.

Use of incorrect terms by the three groups was consistent with data from the literature. In an earlier study, Wilkins and colleagues6 found that only 32% of pulmonary physicians (n=233) and 11% of physicians with other specialties (n=54) were able to identify the pleural friction rub of their sound sample #6. That study also found that 20% of the physicians surveyed failed to identify the normal breath sound of sample sound #7. However, the authors gave no information about the incorrect terms used. Most recently, Hafke-Dys and colleagues10 asked 185 observers, including physicians and medical students, to listen to audio recordings of 24 respiratory sounds and match their assessment with descriptions provided by a team of specialists. Consistent with our findings, they found that the percentage of correct rating was low, the best results being noticed for wheezes (40%) and the stridor (30%). Interestingly, their observers used practically the whole list of proposed terms to describe the sound samples. As a result, on many occasions, the sound recordings were matched with incorrect terms. To quote but a few examples, sound samples #3 and #6—labelled by the experts as ‘normal breath sound’—were matched by many observers with terms like ‘wheeze’, ‘rhonchi’ or ‘coarse crackles’; similarly, samples #2 and #17—labelled as ‘inspiratory wheezes’—were matched with several terms, including ‘normal breath sound’, ‘prolonged expiration’ and ‘crackles’.

The high rate (35%) of three-group agreement on incorrect terms has clinical relevance. In daily practice, especially in ward visits, a circumstance where sound recording and analysis are usually not readily available, auscultatory doubts are usually solved by consensus among two or more observers, especially senior physicians. In this setting, complete agreement on incorrect terms can carry detrimental consequences to the patients. For instance, had it occurred in real patients, our observers’ agreement on the occurrence of non-existent crackles and bronchial breathing—reported on 15 occasions (table 1)—could have motivated the ordering of unnecessary and costly examinations—including chest X-rays and CT scans—or treatments.

Use of recommended terms3 4 by correct raters was suboptimal. The best results involved the use of the terms ‘wheeze’ and ‘stridor’ by virtually 100% of raters describing sample sounds #2 and #4. In contrast, the recommended term ‘normal breath sound’ was used by only 55% of correct raters, 35% of them preferring the old, tenacious ‘vesicular sound’. In the same vein, the traditional term ‘pleural friction rub’ was used by only 2 of the 14 correct raters, the others preferring combinations of the terms ‘pleural’, ‘friction’ and ‘rub’ and, on one occasion, ‘crackle’. Finally, more problematic was the use of no fewer than six correct terms to describe the crackles. Illustrating the need for more effort to encourage the use of standardised terminology, the inappropriate term ‘crepitation’ was preferred by a majority of 50% of raters, while the recommended term ‘crackle’ and the traditional term ‘rale’ were each used by merely 20% of observers.

It may be argued that our definition of correct answer artificially decreased the number of terms used by correct raters. We do not believe this to be so. First, the only way to define a correct answer is by comparison with some sort of gold standard. Instead of expert opinion, preferred by others,8 we opted by a comparison with computer-based sound analysis.11 Second, in our view, the smaller number of terms used by correct raters resulted at least partly from the fact that terminology is not totally independent from auscultatory skill. Conceivably, since, by definition, correct raters are familiar with the sounds, it would be plausible to postulate they picked their descriptive terms from the recommended list likely known to them, which is fairly small;3 4 this explanation fits the high rates of correct use of the terms ‘wheezes’ and ‘stridor’ quoted above. Similarly, as they were likely unaware of recommendations, less skilled observers would tend to choose terms from a virtual, imaginary list. Incidentally, our (intentional) choice of spontaneous sound identification may have maximised this tendency. Finally, it may be argued that, because it was recruited from the same hospital, our group of observers was relatively homogeneous, weakening the results. However, to compensate for this factor, we took care to choose a group quite heterogeneous in terms of background and clinical experience, not to mention education in medical schools from different countries.

The current data have clinical relevance. They show that in a group of observers typical of a university-based hospital, lack of auscultatory skill surpassed observers’ preferences as a factor determining the choice of sound terminology. If generalisable, this finding would suggest that recommendations for terminology alone are unlikely to ensure the universal adoption of a standardised nomenclature, even one as simple as that of ‘crackles’ and ‘wheezes’ for discontinuous and continuous adventitious sounds respectively, proposed by Robertson and Coope more than 60 years ago.21 Instead, our data strongly suggest that, to improve their efficiency, recommendations for terminology should be coupled with measures aiming to improve education on lung auscultation. In practice, this could be achieved, for instance, via the development and wide dissemination of computer-assisted learning tools (CALTs) for use by educators and in self-directed learning. Ideally, these CALTs should address the basic concepts of normal and adventitious sounds, and contain real clinical cases enriched by sound recordings and graphic display of waveforms for analysis. Among the existing tools, the recently developed Computerized Lung Auscultation—Sound Software (CLASS)22 meets the required features and, apparently, has the potential to achieve this goal.