Breath characteristics and adventitious lung sounds in healthy and asthmatic horses

2.1 Study design and animal population

Our study was designed as an observational, analytical, prospective multicenter case-control study conducted at the veterinary medical teaching hospital of the University of Bern (Switzerland) and at the Faculty of Veterinary Medicine of the University of Montréal (Canada).

Suitable candidates were identified using the horse owner-assessed respiratory signs index (HOARSI).²⁵ Briefly, horses were graded on a scale from 1 to 4 based on the current owner-reported signs of respiratory disease (coughing, nasal discharge, breathing pattern, and performance). These grades represent unaffected (grade 1) to severely affected (grade 4) individuals. Horses with a HOARSI grade of 1 were included as controls (ctl), those with a grade of 2 or 3 entered the study as mild to moderate asthma (mEA) cases, and those with a grade of 4 as severe asthma (sEA) cases.

Based on data from a previous study with privately owned horses in Bern, power analysis indicated that 12 ctl and 12 mEA cases were needed to detect differences in adventitious lung sounds. Swiss client-owned, adult horses presented to the ISME equine clinic as well as teaching horses of the University of Bern were enrolled as ctl or mEA during July to December 2022. From the research herd of the Faculty of Veterinary Medicine of the University of Montréal, 10 sEA horses (5 in exacerbation [sEA+] and 5 in remission [sEA−]) were enrolled, and data were collected in September 2022. These horses were diagnosed based on history (HOARSI 4) and previous results of lung function testing (transpulmonary pressure change >15 cm of H₂O when stabled and fed hay) and bronchoalveolar lavage fluid (BALF) cytology (>25% neutrophils). To induce exacerbation, animals were stabled (wood shavings bedding) 4 weeks before the study and challenged by feeding hay. Management remained the same throughout the study period. No treatment was administered before collection of the data, except if horses showed inappetence, respiratory distress, or tachycardia (>60 beats per minute). Horses in remission were kept outside on pasture.

Clinical examination, thoracic ultrasound examination, airway endoscopy, BALF cytology and lung function (only in Montréal) were performed by the same clinician (EG). The results of these procedures were used to further characterize HOARSI-assigned groups. In Bern, examinations were performed on the same day, once in each horse. In Montréal, all examinations (except thoracic ultrasound examination, which was not repeated after the first examination) were performed twice at intervals of 7 to 10 days. This approach resulted in 44 examination sessions and breath sound recordings (ie, once in 12 ctl and 12 mEA and twice each in 5 sEA− and 5 sEA+).

2.2 Clinical examination

Horses were weighed and complete physical examinations including attitude, appetite, heart rate, body temperature, and mandibular lymph node palpation were performed. The respiratory system was specifically examined using the WCS23⁸ during spontaneous breathing. Subsequently, ultrasound examination, digital auscultation and recording, lung function measurements, endoscopic examination and BALF collection were performed in that order, except when otherwise stated.

2.3 Thoracic ultrasound examination

A thoracic ultrasound examination was carried out on all the horses to delimit the respiratory field, measure skin—pleural distance, detect signs of infectious disease, and detect fields potentially interfering with auscultatory findings (eg, scars, masses, free fluid). The examination was performed starting from the 17th to the 3rd intercostal space, and from dorsal to ventral on both sides (Bern: SonoSite Titan, Bothell, WA; Montréal: ExaPad, IMV Imaging, Bellshill, Scotland).

2.4 Digital auscultation device

Respiratory sounds were obtained from 3-dimensional (3D)-printed stethoscope heads using microphones with extended low-frequency response. Stethoscope heads were spring-suspended in 3D-printed enclosures pressed to the horse with elastic belts. This approach eliminated friction noise from the belts and standardized the contact pressure of the stethoscope membrane. Outside noise was further decreased by electronic noise canceling. Breath sounds were amplified, low pass filtered, and digitized within the recording heads with 16 bits per sample at a frame rate of 4410 Hz. Thoracic excursions during inhalation and exhalation were recorded by an inductive plethysmography belt located at the 15th intercostal space and digitized similarly. Samples from all recording heads and the plethysmography unit were transmitted to a central unit and saved on a Secure Digital (SD)-card in a multitrack pulse code modulation (PCM) WAV file. Subjective comparisons using a commercial recording stethoscope suggested the superior sound quality of the novel device.

2.5 Collection of respiratory sound data

For each recording session, 1 sensor was placed over the trachea, and 5 sensors were placed on each side of the thorax and held in place by 3 elastic straps (Figure 1). The plethysmography belt was placed caudally, behind the last elastic strap fixing the 2 caudodorsal sensors at the 15th intercostal space. To avoid pressure sores, a padded mat was placed on the horse's back. Cables connected each sensor to the central unit (placed on the right side, below the caudodorsal sensor) and the latter was powered by a battery attached to the right side of the 2nd ventral elastic belt. The gain was standardized for all recordings and the sound quality was evaluated using headphones before starting the recording. We recorded for 60 minutes in a quiet stable away from other horses and without sedation. The data recorded on the SD card then were transferred to a computer to visually assess the quality of the recording on spectrograms (Audacity version 3.1.1 for Windows).

2.6 Analysis and classification of the respiratory lung sounds

One-hundred random breathing cycles per recording were randomly selected for blind analysis by the same observer (EG). This sample was weighted with 20% for the tracheal sensor and the remaining 80% equally divided for each lung sensor (8% per sensor). A total of 4400 breaths from the 44 recording sessions were presented to the observer in random order using a web application, along with an input mask for their classification (Figure 2). For contextual reference, the observer visualized 2 adjacent breaths before and after the breath to be classified. Additional information, such as sensor location, a spectrogram of the breath sound, and the waveform of the plethysmography data, were provided. Simultaneously, the sound could be heard through headphones with good low-frequency capability. Wheeze, rattle (coarse crackle), crackle (fine crackle), cough, gastrointestinal sounds, heart sounds and environmental noise were assessed as absent, questionable, or present, and breath intensity as decreased, normal or increased. Multiple attributes could be selected for the same breath. Before the start of the analyses, the primary observer (EG) performed training classification of 800 breaths, of which 200 also were classified independently by a second observer (VG). Interobserver discrepancies were discussed until agreement was reached to minimize the effect of observer experience, presumably most pronounced at the beginning of the categorization phase.

2.7 Lung function tests

Lung function tests of sEA horses in Montréal were performed using the Equine MasterScreen impulse oscillometry system (Jaeger GmbH, Würzburg, Germany) in unsedated horses standing in stocks as previously described.^{26, 27} On each test day, the system was calibrated. Briefly, an airtight facemask was placed on the horse's head allowing multifrequency impulses produced by a loudspeaker to be superimposed to the tidal breathing of the horse with simultaneous acquisition of the pressure-flow signal response of the respiratory system by pressure transducers connected to a pneumotachograph. Three consecutive 30-second recordings were completed for each horse, except for 1 horse where recordings could not be made because of intolerance to the examination. Lung function data were acquired using LabManager (version 4.53, Jaeger, Würzburg, Germany) and analyzed with FAMOS (IMC, Meβsysteme, Berlin, Germany) using the fast Fourier transform method. Inspiratory (insp), expiratory (exp) and whole breath resistance (R), reactance (X), and coherence of respiratory system at frequencies of 5 and 10 Hz were analyzed. For the first assessment of 3 horses, the lung function was performed 1 day after the other examinations because of technical problems.

2.8 Endoscopic tracheal mucus scoring and BALF cytology

Horses were sedated with either xylazine (Montréal) (Nerfasin 100, Dechra, Pointe-Claire, QC, Canada; 0.5 mg/kg, IV) or detomidine (Bern) (Equisedan, Dr. E. Graeub AG, Bern, Switzerland; 10 μg/kg, IV) and butorphanol (Montréal: Dolorex, Merck & Co, Inc, Rahway, NJ; 20-30 μg/kg, IV; in Bern: Morphasol, Graeub AG, Bern, Switzerland; 10 μg/kg, IV). Tracheal mucus accumulation was scored along the entire length of the trachea using endoscopy (Bern: VET-OR1200HD, Medical Solution GMBH, Wil, Switzerland; Montréal: Evis Exera II CV-180, Olympus Canada Inc, Richmond Hill, ON, Canada) using a validated scoring system.²⁸

Bronchoalveolar lavage fluid (BALF) collection and cytology methods slightly differed between Bern and Montréal. In Montréal, bronchoalveolar lavage (BAL) was performed under endoscopic guidance as previously described.²⁹ Briefly, the endoscope was advanced down into the right lung until its tip was wedged in the wall of a bronchus. After topical anesthesia with 0.5% lidocaine (Lurocaine; lidocaine hydrochloride 20 mg/mL, Vétoquinol N.-A. Inc, Lavaltrie, QC, Canada), two 250-mL boluses of warm sterile isotonic saline then were sequentially instilled into the bronchus through the endoscope's biopsy channel and aspirated by a suction pump (10-20 mmHg). The samples were kept on ice until reaching the laboratory, where cytocentrifuged preparations of BALF (400 μL) were made and cells were stained with a modified Wright-Giemsa solution (DiffQuick, Fisher Scientific, Waltham, MA). A blinded investigator (SMS) performed differential leukocyte counts using 400 cells. In Bern, as previously described,³⁰ a sterile BALF tube (either 240 or 300 cm long depending on horse size; 10 mm outer diameter; Bivona Medical Technologies) was used for BALF collection. Briefly, the catheter was passed through the nostrils and trachea until it was wedged in a bronchus. Lidocaine (Lidocain 2% Streuli ad us. vet., Uznach, Switzerland) was administered, followed by instillation of 0.9% NaCl solution. The volume of lidocaine and NaCl used depended on the horse's weight, with <300 kg horses receiving 15 mL lidocaine and 200 mL NaCl, whereas horses weighing ≥300 kg received 20 mL of lidocaine and 250 mL of NaCl. The solutions were gently withdrawn manually using 60-mL syringes. The BALF was pooled in a 250-mL silicone-coated glass bottle (Fisher Scientific, Leicestershire, UK), and the amount of liquid collected was recorded. The BALF was filtered through a 40 μm cell strainer (Falcon, FisherScientific, Leicestershire, UK) and sent to the laboratory, where it was first centrifuged for 3 minutes at 500g. A volume of 100 μL of the sediment was combined with an additional 100 μL of phosphate-buffered saline (PBS). The solution was cytocentrifuged a second time for 8 minutes at 1000g. Finally, slides were automatically stained with Wright-Giemsa and differential leukocyte counts of 200 cells were performed by laboratory technicians.

2.9 Statistical analysis

Statistical analyses were done using R version 4.3.1 (R Core Team 2023).³¹ Group differences in means of age, weight, body condition score, skin—pleural distance, WCS23, tracheal mucus score and BALF cytology were assessed by Welch's analysis of variance (ANOVA) and Games-Howell post hoc tests for pairwise comparisons. Differences in corresponding medians were evaluated using heteroscedastic one-way ANOVA.

Except for the repeated measurements in the sEA cases, all statistical analyses were performed using only the first examination sessions, resulting in 34 complete examination sessions (3400 breaths).

The proportions of analyzable breaths that contained adventitious sounds (wheezes, crackles, and rattles) and the respiratory variables (breath intensity, mean and variance of breath duration) were compared across groups (ctl, mEA, sEA−, sEA+) using Kruskal-Wallis tests and Dunn tests for pairwise comparisons, with P-values adjusted with the Holm method. Coughing episodes from the 3400 samples did not present sufficient events for statistical analysis. Breath intensity was calculated as the difference between the proportions of increased and decreased breath sounds, leading to a value between −100% and +100%. The proportion of adventitious sounds was calculated by adding the proportions of sounds labeled as “present” with weight 1 and those labeled as “questionable” with weight 0.5. Mean and variance of breath duration were calculated from the plethysmography data of the entire recordings.

The differences between repeated measurements in the sEA cases (means and variance of respiratory duration, breath intensity, ratio of impulse oscillometry system resistance at 5 and 10 Hz, impulse oscillometry system reactance at 5 Hz, proportion of wheezes, crackles, and rattles, tracheal mucus score, neutrophilic percentage of BALF cytology and WCS23) were assessed using 2-sided 1-sample t-tests. Multiple comparisons were accounted for using Bonferroni correction.

The predictive value of auscultation findings was assessed using general linear models, with group, neutrophilic percentage on BALF cytology, tracheal mucus score, and WCS23 as response variables and breath characteristics (breath duration and variability, breath intensity, wheezes, crackles, rattles, cough) as predictors, controlling for skin-pleural distance, age, and sex. For these models, the proportions of abnormal sounds (wheezes, crackles, rattles, and coughs) were transformed to new variables with 3 levels (abnormal sound absent; at least 1 breath with questionable abnormal sound; at least 1 breath with present abnormal sound). Model selection was based on the Bayesian information criterion. A difference of 2 to 6 in the Bayesian information criterion is needed for positive evidence in favor of a specific model.³²

3.1 Horse population characteristics

Thirty-four horses were included in the study, 11 geldings and 23 mares between 5 and 32 years of different breeds: 12 American Quarter Horses, 9 Warmbloods, 5 Freibergers, 3 Thoroughbreds, 2 Trotters, 5 Paint Horses, 2 Canadians, 2 Grade horses, 1 Arabian, 1 Camargue, 1 Haflinger, and 1 Italian pony. All horses were in good body condition and had normal appetite and attitude, and normal body temperature and heart rate. No significant differences in age or sex were found among the groups, but WCS23, tracheal mucus score, and percentage of neutrophils, lymphocytes, and mast cells in BALF cytology differed significantly between groups (P < .05, Table 1). Comet-tail artifacts were seen on thoracic ultrasound examination in most horses regardless of clinical status (as previously described³³), but no other abnormalities such as consolidations or free fluid were observed.

In accordance with the inclusion criteria, all ctl horses had a HOARSI of 1, and all cases had a HOARSI ≥2 (7 mEA with HOARSI = 2, 5 mEA with HOARSI = 3, 10 sEA with HOARSI = 4). In sEA horses in Montréal, exacerbation was confirmed during lung function assessment based on the ratio of resistance at 5 and 10 Hz being ≥1 and the reactance at 5 Hz being ≤0. Of the 5 sEA+ horses, 1 mare did not tolerate the examination and required inhalation with 1000 μg of Salbutamol (Ventolin, GSK, Mississauga, ON, Canada) at both examinations because of respiratory distress. The status of all sEA− horses was further confirmed using lung function measurements with the ratio of resistance at 5 and 10 Hz being ≤1 and the reactance at 5 Hz being ≥0.²⁶

3.2 Quality assessment of the digital auscultation device and sound characterization

Each horse underwent simultaneous auscultation at 11 locations for 1 hour, resulting in a high number of breath recordings per horse, ranging from 5478 to 27 852. Of all recorded breaths, 96.3% were correctly identified from the plethysmography data and 85.9% were classified as usable, enabling characterization and categorization according to the predetermined criteria. Among nonrespiratory sounds, environmental noises were present in 30.9% and questionably present in 2.3% of the usable samples. Most of the environmental noise was caused by external air ventilation or road traffic. Gastrointestinal sounds also were present in 28.8% and questionably present in 11.1% of the samples. Heart sounds were noted less frequently, with presence at 1.9% and questionable presence at 0.8%.

Training classification of 800 breaths and comparisons with the second observer permitted characterization of the respiratory sounds before the start of scoring. Coughs were characterized by a wide vertical band with high intensity and a broad frequency spectrum >2000 Hz. Crackles and rattles consisted of sequences characterized by high-frequency and high-intensity peaks, occasionally presenting challenges in differentiation because of the similarities. Crackles tended to be present in the inspiratory phase, whereas rattles were present in both the inspiratory and expiratory phases. When heard through headphones, the sound of crackles was similar to Velcro opening, whereas rattles were reminiscent of a snoring sound. Both sounds also could resemble the rubbing of the stethoscope against the skin, leading to potential confusion. These similarities and artifacts led to a higher proportion of these sounds being categorized as questionable or potentially as false positive. Indeed, in the final sample, the proportions of questionable crackles (overall scorings: 5.4% questionable, 8.1% present) and rattles (3.2% questionable, 4.3% present) were considerably higher than of wheezes (1.9% questionable, 9.9% present).

Wheezes appeared as thin transverse bands between 200 and 1000 Hz, providing the most distinctive signal both audibly and visually. Most questionable wheezes were likely a result of potential confusion with intestinal sounds transmitted from the abdomen into the thoracic cavity. The high frequency of gastrointestinal sounds, which were noted in more than 33% of the scored breaths overall, necessitated a strategy to improve distinction. The training phase showed that wheezes typically occurred at the end of expiration, as previously described,³⁴ whereas gastrointestinal sounds were not associated with the respiratory phase. Here, parallel visual inspection of the spectrogram and plethysmography curve was particularly helpful. Additionally, wheezes often occurred in series (commonly present in the breaths preceding and following the analyzed breath), facilitating differentiation from gastrointestinal noises.

Examples of typical adventitious breath sounds can be found as visual representations and as audio files in Supplementary Item 1.

3.3 Respiratory variables and adventitious sounds in healthy and asthmatic horses

Groups differed significantly in the proportion of breaths with wheezes (P < .001), rattles (P = .01) and crackles (P < .01) and in breath intensity (P < .01), whereas mean or variance of breath duration were not significantly different among groups (Figure 3). Pairwise comparisons showed that these variables were consistently higher in sEA+ compared to all other groups. This was particularly marked for wheezes where sEA+ had a median at 68.6% (interquartile range [IQR], 27.5%-89.9%) whereas ctl, mEA and sEA− were at 0% (0%-0% each). Crackles and rattles followed the same pattern, with medians for sEA+ at 66.1% (IQR, 32.6%-68.2%) and 17.7% (IQR, 7.6%-20.6%), respectively, and medians for ctl, mEA and sEA− at 0.6% (IQR, 0%-1.2%) and 1.8% (IQR, 1.4%-3.0%), 1.0% (IQR, 0.4%-2.5%) and 2.4% (IQR, 0.6%-4.1%), and 0.7% (IQR, 0%-3.2%) and 1.3% (IQR, 1.2%-2.6%), respectively. The medians for respiratory intensity showed a progressive increase to −9.4% (IQR, −15.5% to 1.3%) for ctl, 1.7% (IQR, −5.0% to 11.9%) for mEA, 5.6% (IQR, −1.2% to 27.8%) for sEA− and a much higher value of 97.9% (IQR, 82.0%-100%) for sEA+, with significant differences between sEA+ and ctl.

3.4 Measurement repeatability in sEA cases

The differences between repeated measures of mean and variance of breath duration, breath intensity, lung function scores (ratio of impulse oscillometry system resistance at 5 and 10 Hz, impulse oscillometry system reactance at 5 Hz), proportions of adventitious sounds (wheezes, crackles, rattles), tracheal mucus score, and neutrophilic percentage on BALF cytology, and WCS23 were not significantly different from zero (all P = 1; except for variance of breath duration, P = .63), indicating stable clinical status for sEA− and sEA+. Statistical results and a graphical representation of this analysis can be found in Supplementary Item 2.

3.5 Linear regression analyses

Mean breath duration and breath intensity were positively associated with asthma severity (estimate = 0.53, P = .07; estimate = 0.07, P < .01, respectively).

We found no significant association of the tested predictors with neutrophil percentage on BALF cytology.

Wheeze was positively associated with tracheal mucus score (estimate = 0.63, P = .03). Wheeze explained 14% of the variance of the tracheal mucus score.

Breath intensity and wheeze were positively associated with WCS23 (estimate = 0.08, P < .01; estimate = 2.40, P = .06, respectively).