Acoustic biomarkers in asthma: a systematic review

Abstract

Objective

Current monitoring methods of asthma, such as peak expiratory flow testing, have important limitations. The emergence of automated acoustic sound analysis, capturing cough, wheeze, and inhaler use, offers a promising avenue for improving asthma diagnosis and monitoring. This systematic review evaluated the validity of acoustic biomarkers in supporting the diagnosis of asthma and its monitoring.

Data sources

A search was performed using two databases (PubMed and Embase) for all relevant studies published before November 2023.

Study selection

27 studies were included for analysis. Eligible studies focused on acoustic signals as digital biomarkers in asthma, utilizing recording devices to register or analyze sound.

Results

Various respiratory acoustic signal types were analyzed, with cough and wheeze being predominant. Data collection methods included smartphones, custom sensors and digital stethoscopes. Across all studies, automated acoustic algorithms achieved average accuracy of cough and wheeze detection of 88.7% (range: 61.0 − 100.0%) with a median of 92.0%. The sensitivity of sound detection ranged from 54.0 to 100.0%, with a median of 90.3%; specificity ranged from 67.0 to 99.7%, with a median of 95.0%. Moreover, 70.4% (19/27) studies had a risk of bias identified.

Conclusions

This systematic review establishes the promising role of acoustic biomarkers, particularly cough and wheeze, in supporting the diagnosis of asthma and monitoring. The evidence suggests the potential for clinical integration of acoustic biomarkers, emphasizing the need for further validation in larger, clinically-diverse populations.

Keywords:

• Asthma monitoring
• sound
• recording
• digital
• respiratory
• algorithm

Introduction

In recent years, new methods of diagnosis and monitoring have arisen, including the use of digital biomarkers, which are characterized as data derived from digital devices (1). This includes smartphones, sensors, wearables and implants, which record digital biomarkers that are objective, quantifiable and encompass both physiological and behavioral information (2). As medical devices are more integrated into clinical practice, digital biomarkers have become a new tool for studying precision medicine.

Digital biomarkers allow for long-term, noninvasive, automatic, remote, continuous data collection and monitoring. For example, wristwatches collecting heart rate data to predict and notify users about abnormal heart rates (3). Digital biomarkers have been identified and validated across different medical specialties for conditions such as schizophrenia (3), Alzheimer’s disease (4), neuromuscular disorders (5) and congestive heart failure (6).

Acoustic biomarkers are a subset of digital biomarkers which use sounds to identify a disease state or physiology. In cardiac monitoring, digital stethoscopes are now able to analyze heart sounds (7,8). Acoustic biomarkers have also been studied in gastrointestinal disease, where low-cost sensors can record and analyze bowel sounds to measure gastrointestinal motility (9). Acoustic biomarkers have been assessed in orthopedics, where the sounds produced by joint articulation could be used to assess joint health (10).

In the context of respiratory health, these acoustic biomarkers primarily involve the recording of respiratory sounds to analyze elements such as breathing patterns, coughs, wheezing and crackles (11). Research into acoustic biomarkers in respiratory medicine increased during the COVID-19 pandemic, when they were used to predict COVID-19 diagnosis (12–16). Since then, acoustic biomarkers have shown promise in monitoring and diagnosis in other respiratory diseases: COPD (17–20), pneumonia (21–23), sleep apnea (24–26), as well as asthma, which received particular attention in recent years (27–29). Remote monitoring of patients using acoustic technology in combination with artificial intelligence algorithms holds promise for continuous, remote patient monitoring (30).

Asthma requires regular monitoring to achieve good control and prevent exacerbations by appropriately escalating treatment (31). Despite the availability of effective treatments, optimal asthma control is seldom achieved (32–38). Some acoustic respiratory devices are already commercially available for testing and the scientific community is likely to see a growing trend in development of these type of devices (39). Automated acoustic monitoring could allow for long-term data collection about the patient’s asthma status, improve management and detect exacerbations early. For instance, a study of 89 asthmatics showed that 24-h cough frequency correlated with Asthma Control Questionnaire scores and FEV1 (27). Similarly, longitudinal nocturnal cough monitoring using sound recording in a 29-day-long study in asthmatic patients was used to effectively predict asthma exacerbations (28).

Acoustic monitoring has also been applied to wheeze analysis. In pediatric patients, breath sounds were recorded overnight continuously using a phono-pneumography sensor over the trachea. It was demonstrated that even with subjectively well-controlled asthma (based on asthma diary entries), children experience considerable amount of wheezing overnight, suggesting suboptimal management of asthma (40). Furthermore, computerized analysis of wheeze sounds during tidal breathing correlated with asthma severity (41).

To our knowledge, only specific reports and applications of acoustic biomarkers in asthma have been published, but no systematic review has been undertaken on the utility and application of the various aspects of acoustic biomarkers in asthma. This systematic review, therefore, aims to evaluate the validity of acoustic biomarkers in the diagnosis and monitoring of patients with asthma.

Methods

Inclusion criteria

Studies on the use of acoustic signals, such as cough or wheeze, as digital biomarkers in asthma were included. Studies used recording devices to register sound or performed analysis of prerecorded sounds. Studies tested the sound recording devices and compared them against manual sound annotation or other standards of care (such as clinician diagnosis or spirometry).

Exclusion criteria

Qualitative studies, animal studies, technical reports and non-English language studies were excluded. Studies relating to other respiratory diseases (such as COPD or COVID-19), studies including analysis based on other factors (such as patient-reported questionnaires) were excluded. In the first phase of abstract screening, technical reports were also excluded (engineering papers analyzing sound characteristics without testing in patient populations, as well as sound analysis protocols).

This systematic review was registered with the PROSPERO international database (PROSPERO ID: CRD42023473584) and followed the Preferred Reporting Items of Systematic Reviews and Meta-analyses guidelines (42).

Literature search

A search was performed using two databases: PubMed and Embase. All studies in English up to 1st November 2023 were included. The search was conducted based on the following terms and their MeSH terms: “asthma”, “lung disease”, “cough”, “wheeze”, “breath”, “acoustic signal”, “sound” and “cough”. The full search strategy, registered on PROSPERO, and excluded terms are available in the Appendix.

Study selection

Titles and abstracts of search results were screened independently for eligibility by two authors. Full texts were obtained and evaluated. References of these full texts and relevant literature reviews were screened to identify any studies not captured by the original search strategy.

PRISMA diagram

Data abstraction

Data was abstracted from full-text papers into Microsoft Excel spreadsheets under the following data categories: Study Selection, Data Abstraction and Data Analysis, Study Characteristics, Patient characteristics, Data collection methods, Acoustic signals, Asthma identification, Risk of Bias Assessment. A full strategy for data abstraction is available in Appendix Figure B1.

Risk of bias assessment

All studies included in the review were appraised using the QUADAS-2 tool for reviews to evaluate the risk of bias and applicability of studies (44).

Statistical analysis

Statistical analysis and calculations were done using Microsoft Excel (Microsoft Corporation). Average accuracy has been calculated for all studies. For sensitivity and specificity, ranges and medians have been provided due to the limited availability of specific statistical details in studies (such as confidence intervals and sensitivity thresholds).

Outcomes

The primary outcome of the study was to assess the validity of automated acoustic sound analysis in the diagnosis and monitoring of asthma, including detection of acute exacerbations. The secondary outcome of the study was to identify the different types of acoustic signals collected and characterize the types of algorithms.

Results

Study selection

7855 Studies were identified in the initial abstract screening. Duplicates (n = 300) and studies not relevant to the study question were excluded (n = 7456). The majority of the latter were engineering studies which did not test the acoustic biomarkers on patients. 99 Studies were identified, of which 71 studies were excluded because of no full text access, no clinical validation or no mention of asthma. 19 Studies remained, and a further 8 were identified through reference screening as shown in Figure 1, resulting in 27 studies for analysis.

Figure 1. Workflow representing phases of the systematic review process (PRISMA criteria (43)).

Data abstraction and data analysis

Table 1 describes the study details and patient characteristics. Table 2 summarizes data collection methods, study results, conclusions and quality of the studies.

Table 1. Study details and patient population characteristics.

No	PMID/doi	Author	Year	Country	Study setting	Study aim	Study measure	Study (days)	Endpoint	n	M/F	Patient population characteristics
1	33363391	Tinschert et al.	2020	Switzerland	Real-world	Validate cough frequency algorithm to identify and predict asthma control	Uncontrolled asthma detection and prediction of attacks using cough frequency	29	Primary endpoint–Asthma Control test–weekly, secondary–self reported asthma attacks	79	M = 41, F = 53	N = 79 (self-reported asthma)
2	23000725	Habukawa et al.	2013	Japan	Clinic-based study	Determine if breath sound analysis could be used to detect changes in transfer characteristics according to the control level of asthma	Pediatric respiratory sounds analysis	1	Asthma control assessment	139	M = 61, F = 78	N = 80 (physician reported asthma)
3	31167662	Porter et al.	2019	Australia	Clinic-based study	Validate an algorithm for pediatric respiratory diagnoses including asthma and reactive airway disease (RAD)	Pediatric cough sounds analysis	1	Automated diagnosis algorithm	585		N = 149
4	30439599	Nabi et al.	2019	Pakistan	Inpatient hospital appointment	Investigate and classify wheeze sounds of asthmatic patients according to their severity level (mild, moderate, severe).	Analysis of spectral integrated features	NR	Wheeze sound classification	55	M = 42, F = 21	MILD asthma = 17, moderate = 18, severe = 20
5	33018992	Nemati et al.	2020	USA	Hospital	Analyze cough sounds to estimate lung function and severity of the disease	Voluntary cough recording	NR	Analysis of cough sounds	70	M = 28, F = 42	N = 70 (asthmatic = 25)
6	33484516	Nabi et al.	2021	Pakistan	In-patient hospital	Investigate behavior of frequency related features in three severity levels of asthma patients (mild, moderate and severe)	Frequency based features and mean frequency	NR	Severity of asthma (clinical judgement)	55	M = 34, F = 21	Asthmatic patients (physical reported)
7	23366456	Holmes et al.	2012	Ireland	Real-world	An automatic detection algorithm to identify inhalations from acoustic recordings of inhaler use.	Inhalation recordings	92	Inhaler use over 3 months	20	M = 9, F = 11	Asthmatic patients
8	27170883	Holmes et al.	2014	Ireland	Real-word	An algorithm to automatically analyze patient inhaler use in order to evaluate adherence.	Inhalation recordings	30	Adherence over 1 month.	12		Asthmatic patients
9	10.1109/ICASSP40776.2020.9054062	Yadav et al.	2020	India	Inpatient hospital	Validity of an algorithm detecting asthma in spontaneous speech	Analysis of speech to recognize asthma patients	NR	Compare the use of speech and non-speech sound in asthmatic vs. healthy subject classification	95	M = 28, F = 19	N = 48 asthmatic, n = 47 healthy patients
10	10.1109/TED.2019.2951580	Khan et al.	2020	Saudi Arabia	Laboratory	Test the design of a body-worn sensor detecting wheezing	Analysis of wheezing episodes	NR	Wheeze detection (automated count)	4	M = 2 F = 2	N = 3 prerecorded wheeze sounds from asthmatic patients
11	10.1109/WH.2016.7764567	Ra et al.	2021	South Korea	Laboratory	Investigate and classify wheezing in respiratory sounds using an automated system	Wheezing sounds analysis	NR	Wheeze detection (automated count)	NR	NR	N = 31 wheezing asthmatic patients, n = 32 healthy
12	10.1016/j.bspc.2015.05.002	Mazić et al.	2015	Croatia	Inpatient hospital	Investigate and classify wheezing in children’s respiratory sounds	Children’s respiratory sounds	NR	Respiratory sound analysis	16	NR	N = 16 asthmatic and obstructive bronchitis patients
13	33299332	Rassouli et al.	2020	Switzer-land	Inpatient hospital recording	Investigate the prevalence and characteristics of nocturnal cough in asthmatic patients	Manual cough counting, pattern recognition	29	Primary–nocturnal cough rate manually labeled by human annotators. Secondary–cough severity.	94	M = 41, F = 53	N = 94 asthmatic patients
14	10.1109/EMBC.2018.8512452	Azam et al.	2018	Pakistan	Hospital recording	Analyze breath sounds to diagnose abnormalities present	Respiratory sound analysis	NR	Respiratory sound analysis	50	NR	N = 50 patients with COPD, asthma, or bronchitis
15	30927097	Grzywalski et al.	2019	Poland	Inpatient hospital	Analyze auscultatory sounds in four classes: wheezes, rhonchi, and fine and coarse crackles	Analysis of breath sounds	NR	Automated analysis of breath sounds	50	NR	NR
16	29346430	Taylor et al.	2018	Ireland	Inpatient hospital	Estimation of inhalation technique	Analysis of sound to assess inhaler use	NR	Estimation of inhalation flow profile to assess inhaler medication adherence	20	NR	N = 20 healthy participants
17	36699651	Kuhn et al.	2023	Switzer-land	Hospital, outpatient	Evaluate the performance of the cough detector	Number of time-stamped cough epochs	8	Cough counting	5	NR	N = 5 diagnosed with asthma by pulmonologist
18	35989223	Kuluozturk et al.	2022	Turkey	Clinic-based study	Training and validating a novel model designed for cough detection.	Accuracy for cough sound-based multiclass classification of diseases.	153	Automated cough detection	110	NR	Physician diagnosed
19	33875413	Habukawa et al.	2021	Japan	Clinic-based study	Validate an automatic wheeze recognition algorithm as a clinical medical device in children.	Wheeze sound analysis	153	Recognize wheeze from sound	374	M = 241, F = 133	N = 374 with mild bronchial asthma
20	33019267	Vatanparvar et al.	2020	USA	NA	Investigate unique features in a cough sound and validate a model for cough detection.	Cough sound detection	NR	Validation of algorithm to recognize coughs	69		N = 60 physician diagnosed asthma
21	35263208	Porter et al.	2023	Australia	Hospital	Develop and test the accuracy of a smartphone-based diagnostic algorithm that detects exacerbations.	Cough sound detection	NR	Exacerbation prediction compared with actual admissions	119	M = 33, F = 86	N = 119 physician diagnosed asthma
22	32459641	Barata et al.	2020	Switzerland	Home monitoring	Early recognition of exacerbations	Recognition and segmentation of nocturnal asthmatic coughs and epochs	28	Automated cough counting as predictor for exacerbations	94	M = 40, F = 54	NR
23	29990246	Oletic et al.	2018	Malaysia	NA	Algorithm description for wheeze quantification	Wheeze sound analysis	NR	Asthmatic wheeze detection	16		N = 8 asthmatic, n = 8 healthy patients
24	29650306	Islam et al.	2018	India	Research center	Classification of normal and asthmatic subjects using signal processing of posterior lung sound signals	Breath sounds analysis	NR	Classification of normal and asthmatic subjects	60		N = 30 healthy patients, n = 30 asthma (n = 8 mild, n = 11 moderate, n = 11 severe)
25	33400647	Sen et al.	2021	Turkey	Hospital	Developing a method to differentiate between asthma and COPD using breath sounds.	Breath sounds analysis	1	Classification of asthma and COPD	30		Diagnosed by FEV1/FVC values below 0.7 on the day.
26	19085584	Guntupalli et al.	2008	USA	Hospital	Evaluate the utility of vibration response imaging in differentiating between asthma and COPD.	Vibration response imaging validation (percussion-like device measurements)	1	Differentiating between COPD and asthma using vibration imaging	66		Diagnosed with medical history, physical exam, spirometry
27	10.1002/ppul.23125	Hirai et al.	2015	Japan	Hospital	Develop a new cough monitoring system for children.	Validate cough monitoring system	2	Cough counting algorithm	3	M = 6, F = 4	Clinical decision made based on previous records

NR: not reported.

Table 2. Data collection methods, study results, conclusions and risk of bias (SN: sensitivity; SP: specificity; ACC: accuracy; NR: not reported).

No	PMID/doi	Author	Data collection method	Device	Signal: type, number		Algorithm performance: SN, SP, ACC			Conclusion
1	33363391	Tinschert et al.	Smartphone	Smartphone	Cough	2291	54	67	61	Statistically significant association of nocturnal cough and sleep quality with asthma control
2	23000725	Habukawa et al.	Custom device	Trademarked device	Breath sounds	139	83.6	84.2	NR	Breath sound analysis was able to identify patients without asthma, with mild asthma or severe asthma
3	31167662	Porter et al.	Smartphone	Smartphone	Breath sounds	2925		*		Breath sound analysis was able to identify patients with asthma. Positive percent agreement in three age groups: 97% (29 days–12 years), 80% (29 days–2 years), 98% (2–12 years). Negative Percent Agreement was 91, 97, and 88%, respectively.
4	30439599	Nabi et al.	Audio auscultation device	Trademarked device (contact device)	Wheeze	775	95	99		The classifier exhibited good performance in terms of sensitivity, specificity, and PPV.
5	33018992	Nemati et al.	Smartphone, smartwatch, microphone	Smartphone, smartwatch, microphone	Cough	3695	NR	NR	95	Accurate estimation of lung obstruction severity based on acoustic features of cough.
6	33484516	Nabi et al.	Digital stethoscope	Trademark device	Wheeze	775		*		All wheeze samples indicate significant differences in severity levels.
7	23366456	Holmes et al.	Custom device	Trademarked device	Inhaler	255	95	94	89	The algorithm was able to detect inhalations from an inhaler.
8	27170883	Holmes et al	Custom device	Trademarked device	Inhaler	407	98.3	98.4	92.1	Cohen’s kappa agreement score was >0.81
9	10.1109/ICASSP40776.2020.9054062	Yadav. et al	Microphone	Microphone	Speech	2938	NR	NR	77.8	Features of speech can be used for asthmatic and healthy subject classification.
10	10.1109/TED.2019.2951580	Khan et al.	Custom device	Trademarked device (contact device)	Wheeze, speech, cough	NR		*		A flexible acoustic sensor can be used to detect wheezing in asthma. The gage repeatability and reproducibility calculates the repeatability as 92.12%.
11	10.1109/WH.2016.7764567	Ra et al.	Digital stethoscope	Trademarked device (contact device)	Breath sounds	103	100	95	98	An ecosystem that allows healthcare professionals to collect physiological data with good results.
12	10.1016/j.bspc.2015.05.002	Mazić et al.	Microphone	Microphone and computer	Wheeze	45	99.67	99.94	NR	The two-level algorithm for wheezing detection in children’s respiratory sounds can discriminate wheeze signals from non-wheeze sounds.
13	33299332	Rassouli et al.	Smartphone	Smartphone	Cough	704,698		*		Patients coughed in 53% of 2212 nights, The highest counts were measured in the first 30 min in bed.
14	10.1109/EMBC.2018.8512452	Azam et al.	Smartphone	Smartphone	Breath sounds	255	NR	NR	75.21	Human breath sound can be used to detect irregular patterns occurring in respiratory cycles due to asthma.
15	30927097	Grzywal-ski et al.	Digital stethoscope	Trademarked device (contact device)	Breath sounds	522	76.5	83.6	NR	The implementation of automatic sound analysis can significantly improve the efficiency of interpretation of auscultation sounds.
16	29346430	Taylor et al.	Custom device	Trademarked device	Inhalator	300	NR	NR	90.89	The method generated sufficient accuracy in estimating inhalation parameters, such as peak expiratory flow and inspiratory capacity.
17	36699651	Kuhn et al.	Smartphone	Smartphone	Cough	648	88.55	99.97	NR	The cough monitoring system was successful and showed similar metrics as other automated cough detection systems.
18	35989223	Kuluo-zturk et al.	Smartphone	Smartphone	Cough	787	NR	NR	99.39	The model possesses an excellent multiclass classification accuracy of 99.39% for cough sound-based diagnosis.
19	33875413	Habukawa et al.	Custom device	Custom device	Wheeze	1201	96.6	98.5	NR	The wheeze recognition algorithm was verified to identify wheezing with high accuracy.
20	33019267	Vatanpa-rvar et al.	Smartphone	Smartphone	Cough	5380	90.3	96.39	93.34	The method enables subject-specific collection of coughs with limited error.
21	35263208	Porter et al.	smartphone	NR	Cough	2925	63.8	94.9	73.95	Our diagnostic algorithm accurately detected the presence of an asthma exacerbation as a point-of-care test without requiring clinical examination or lung function testing.
22	32459641	Barata et al.	Smartphone	NR	Cough	30304	99.9	91.5	99.8	Our study showed longitudinal nocturnal cough and cough-epoch recognition from nightly recorded smartphone-based audio from adults with asthma.
23	29990246	Oletic et al.	Sound database recordings	Microphone and smartphone	Wheeze	7480	89.34	96.28	94.91	The proposed algorithm identifies occurrence and tracks multiple individual wheeze frequencies.
24	29650306	Islam et al.	Custom	4-channel data acquisition system	Breath sound	NR	NR	NR	93.3	The proposed multichannel asthma detection method yields good results in cough identification.
25	33400647	Sen et al.	Custom	14-channcel pulmonary sounds data acquision device	Breath sound	9600	NR	NR	100	Pulmonary sounds analysis may be a complementary tool in clinical practice for differential diagnosis of asthma and COPD, especially in the absence of reliable spirometry testing.
26	19085584	Guntupalli et al.	Vibration device	VRI Device System (contact device)	Vibration	NR	NR	NR	84	Combined qualitative and quantitative evaluations of lung sounds are sensitive in distinguishing between lung sound recordings of COPD and asthma individuals.
27	10.1002/ppul.23125	Hirai et al.	Video camera and microphone with accelerometer	Microphone: EM167NM2	Cough	NR	98,8	97,8	NR	Our method demonstrated good agreement with the video-audio recording method in children and is considered to be highly useful for the objective monitoring of overnight cough in children with respiratory diseases.

*Results in conclusions.

Study characteristics

Of the 27 studies reviewed, the following acoustic signals were assessed: breathing (n = 7), coughing (n = 9), wheezing (n = 6), inhaler use (n = 3), speech (n = 1) and chest percussion sounds (n = 1). Studies took place in the following settings: at home (n = 4), clinic-based (n = 5), hospital (n = 11), research laboratory (n = 3) and non-reported (n = 2). The average study length was 41 days (range: 1 – 153).

Patient characteristics

The total number of tested patients was 2290, with an average of 88 and median of 56 patients (range: 3–585). 52% of patients were male, among data for which gender information was available.

Data collection methods

Acoustic data was collected using various electronic devices, most commonly smartphones (n = 12), custom microphones and sensors (n = 8), digital stethoscopes (electronic stethoscopes with a microphone) (n = 4) and other devices such as vibration sensors and digital inhalers with sensors (n = 3). Studies either monitored patients in real-time using body-worn sensors and devices (e.g. smartwatch), or used microphones (e.g. smartphone) to collect data for retrospective analysis (analyzing sound signals to detect trends and specific features).

Acoustic signals

Overall, 73,750 acoustic signals were analyzed across all studies, with average of 3352 and median of 781 (range: 45–30,304).

Asthma identification

Across the 27 studies, the algorithms for automated asthma detection achieved an average accuracy of 88.7% (range: 61.0 − 100.0%) with a median of 92.0%. The sensitivity of asthma detection ranged from 54.0 to 100.0% with a median of 90.3%; specificity ranged from 67.0 to 99.7%, with a median of 95.0%.

Risk of bias assessment

All studies have been reviewed for risk of bias and applicability using the QUADAS-2 tool. 70.4% (19 out of 27) studies have risk identified in at least one domain in the QUADS-2 tool. Full results are detailed in Table 3. For both the “Risk of Bias” and “Applicability Concerns” categories, “Index Test” was the area with most studies demonstrating risk of bias (10 studies were marked as “high risk” of bias in both categories for “Index Test”).

Table 3. QUADAS-2 risk of bias and applicability of studies.

	Year	Author	Risk of bias				Applicability concerns
			Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
1	2020	Tinschert et al.
2	2013	Habukawa et al.
3	2019	Porter et al.
4	2019	Nabi et al.
5	2020	Nemati et al.
6	2021	Nabi et al.
7	2012	Holmes et al.
8	2014	Holmes et al
9	2020	Yadav et al
10	2020	Khan et al.
11	2021	Ra et al.
12	2015	Mazić et al.
13	2020	Rassouli
14	2018	Azam et al.			?	?
15	2019	Grzywalski et al.
16	2018	Taylor et al.
17	2023	Kuhn et al.	?				?
18	2022	Kuluozturk
19	2021	Habukawa et al.
20	2020	Vatanparvar et al.			?	?
21	2023	Porter et al.
22	2020	Barata et al.	?				?
23	2018	Oletic et al.			?	?
24	2018	Islam et al.
25	2021	Sen et al.			?	?
26	2008	Guntupalli
27	2015	Hirai et al.			?	?

Note. Orange—high risk, green—low risk, question mark—unclear risk.

Cough monitoring and predicting asthma exacerbations

Most studies in the review analyzed various forms of cough detection in order to diagnose asthma or predict asthma exacerbations. Smartphone-recorded nocturnal cough detected periods of uncontrolled asthma (28). Nocturnal cough and sleep quality predicted asthma exacerbations with 70.0–75.0% accuracy, 75.0–88.0% sensitivity and 57.0–72.0% specificity (28). Other studies used an algorithm to analyze overnight recordings, which predicted asthma exacerbations without the need for peak-flow monitoring or spirometry (45,46). Similarly, a study on 94 adult patients with physician-diagnosed asthma recorded cough sounds overnight with a mobile phone for 30 days. Coughs were recorded in 53% of nights and were 4.5 times more frequent in asthmatics (29). A clinic-based study on an automated cough sound analytic system for the identification of asthma in children demonstrated 97.0% agreement between the automated analyzer and the clinical asthma diagnosis (47).

Wheeze detection

Several studies in the review focused on wheeze detection as a way of monitoring and predicting exacerbations, with various approaches, such as using custom wearable sensors (48). A proof-of-concept study described a low-cost, flexible aluminum foil sensor, which was noninvasively attached to the human chest to detect wheezing. The device underwent limited testing on 4 participants, however, a precision measure of variability (gage repeatability) reached 92.1% (48). An integrated smartphone application system (which collected physiological, environmental and patient-reported data) demonstrated up to 96.0% accuracy in detecting patients’ wheezing from sound recording and accurately classifying them as wheezing or normal breathing (49). Digital stethoscopes are also used in asthma acoustic monitoring. Artificial intelligence (AI) was applied to neural networks for polyphonic sound detection to detect patterns in the recorded sounds, including any pathology in breathing cycles. 522 Auscultatory sounds from 50 pediatric asthmatic patients demonstrated that AI-powered digital stethoscopes can improve the accuracy of auscultatory examinations (13% improvement in sensitivity of wheeze detection by the algorithm in comparison to clinicians) (50). Another custom-made, hand-held device with a wheeze recognition algorithm identified wheezing with 96.6% sensitivity and 98.5% specificity (51).

Asthma severity

Studies recorded wheeze sounds at several positions on the chest- acoustic frequency features were analyzed, and there were significant differences between asthma severity levels, showing that the technology was able to classify patients according to severity level (41,52).

Inhaler use

By recording inhalation sounds during inhaler use, the patient’s inhalation flow profile can be estimated. Inhalation technique could be assessed by estimating inhalation flow profiles and other measures, such as inspiratory capacity, with 90.89% accuracy of estimating peak expiratory flow (53). In two studies, an algorithm based on inhalation sounds was developed to automatically assesses adherence to a dry powder inhaler by detecting the use of the inhaler automatically over a 3 month period (54,55).

Discussion

The aim of the systematic review was to assess the validity of acoustic biomarkers in the diagnosis and monitoring of asthma. Studies of acoustic biomarkers in asthma showed high average accuracy (88.7%) in the detection of asthma. Several methods of acoustic data collection have been identified, with cough and wheezing being present in most discussed studies as the main acoustic measures collected and analyzed by algorithms.

Standard asthma questionnaires rely on subjective reporting of a patient’s symptoms and have the potential to be replaced by long-term monitoring of acoustic biomarkers, which shows promise in monitoring asthma control. Moreover, the technology presented in the studies are accessible and almost fully automated, which allows for long-term remote monitoring. Many models integrate the use of AI-guided algorithms, which are trained on acoustic databases, which further facilitates decision making (12,13,56,57). Use of acoustic biomarkers can help to reduce some patient-related issues with asthma management, such as poor adherence or poor inhaler technique.

Similarly to studies in COVID-19¹³ (58), studies of acoustic biomarkers in asthma demonstrated high specificity and sensitivity in distinguishing coughs and wheezing from asthmatic and non-asthmatic patients, which raises the potential for automated support in reaching a clinical decision of asthma diagnosis using passive sound recording (49). Other studies, such as ones using acoustic features of cough in COPD, show promise in predicting exacerbations. For instance, a respiratory sensor (which recorded respiratory sounds) was able to predict 78.8% of COPD exacerbations on average 5 days earlier than when the patient sought medical attention (17). Detecting asthma exacerbations was, however, described in only one of the reviewed studies, where monitoring nocturnal cough predicted asthma exacerbations with 70–75% accuracy (28). Other studies analyzing the usefulness of acoustic biomarkers in respiratory medicine drew similar findings to this review.

Our results are similar to other studies of acoustic biomarkers in respiratory medicine. It is possible to diagnose COVID-19 with high sensitivity and specificity (98.5 and 94.2%) using cough sound analysis, whereas findings in asthma reached 97% (13,47). Several studies in our review utilized digital stethoscopes with good outcomes, similar to other applications of this technology other areas of medicine, such as cardiology. Real-time monitoring of heart disease using similar technology achieved 97 and 88% accuracy respectively for classifying abnormal and normal heart sounds in early detection of cardiac disease (59). Digital stethoscopes have similar limitations across fields of medicine, including the lack of randomized trials comparing them to gold-standard diagnostic techniques.

Another promising area of research described was the use acoustic biomarkers in assessing inhaler adherence and technique. Use of digital inhalers, which lack the ability to measure acoustic biomarkers, has led to improved inhaler adherence and reduced exacerbation rates, as outlined in a recent review of 20 studies of digital inhalers in asthma and COPD (60). The added benefit of acoustic biomarkers is that they can be used to evaluate inhaler technique, such as the study included in our review which estimated inhalation flow with 90.9% accuracy. A similar study in patients with various airways diseases (including COPD) showed that audio signal analysis estimated peak inspiratory flow rate through an inhaler mouthpiece with 88.2% accuracy. Furthermore, audio analysis detected inhaler technique errors at a similar rate compared to a clinical reviewer checklist for inhaler technique (85% for the audio analysis, compared to 90% for the checklist assessment) (61). Research is needed to assess if the use of acoustic biomarkers to evaluate inhaler technique results in improved asthma outcomes or sustained improvements in inhaler techniques.

The evolution of digital health data analytics is an opportunity for the optimization of outcomes in patients with respiratory disease. New literature suggests the potential of smartphone-based diagnostic tools that use audio recordings to monitor symptoms (62). It minimizes the limitations of patient-driven monitoring and leads to early escalation to a clinician when symptoms persist or deteriorate despite best medical management (63). Additionally, it is a way to ease the significant burden of respiratory disease on healthcare systems by introducing cost-effective solutions which reduce labor intensive management strategies. In many cases, most importantly, it also increases patient engagement in their own medication adherence and self-monitoring (64).

This systematic review has important limitations. Many studies chosen for the review focused on the technical validation of algorithms (which analyzed prerecorded sound patterns) with limited testing in real-time and in real-word situations. There was little to no mention of statistical power calculations in the studies included. The external validity of the results is limited by the relatively small sample sizes in the included studies (average of 88 patients per study). Data extraction was very heterogenous, due to limited methodological detail, such as lack of patient recruitment information. Studies in the review were also found through screening of medical databases only. Future studies should also screen engineering and computing study databases, in order to find missing datasets of acoustic biomarkers with larger sample sizes.

70.4% (19/27) of the included studies were identified as having a risk of bias using the QUADAS-2 tool, of which the most common area was the Index Test, where 10 studies had high risk of bias in the “Risk of Bias” and “Applicability Concerns” categories. A pre-specified index test threshold was not explicitly stated in some of the studies, possibly resulting in an overestimation of performance of some algorithms. This also resulted in limited comparison between studies, as they lacked standardization. Data analysis could not be performed uniformly on all studies in the review due to unavailability of sensitivity or specificity thresholds in some of the studies analyzed. Therefore, a range of results are provided in this review instead of average results and confidence intervals, which could be improved in future studies by explicitly stating performance thresholds for algorithms, enabling meta-analysis studies.

There is a need for a multidisciplinary approach where clinicians provide input into algorithm development, and engineers offer technical advice. Focus needs to be put on real-word applications of algorithms to check the reproducibility of those solutions. Large acoustic datasets should be validated clinically, so that gold-standard comparators can be used in future study design, as seen in COVID-19 acoustic testing (16).

Conclusions

Acoustic biomarkers are a promising new tool in the diagnosis and monitoring of asthma. However, trials of their use in asthma have important limitations. Therefore, randomized-control trials of acoustic biomarkers in asthma diagnosis and monitoring are warranted, before they can be used widely in clinical practice.

Acknowledgments

Many thanks to Dr. Paul Pffefer and Zofia Przypaśniak for their advice and help with project conceptualization and advice. The research did not receive any grant or form of funding from public, commercial or nonprofit sectors.

Disclosure statement

The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.

Appendix A

Full search strategy used in the systematic review.

#1 Asthma [MH]

#2 Lung Diseases, Obstructive [MH:NOEXP]

#3 Anti-asthmatic agents [MH]

#4 Asthm

#5 Bronchospas*

#6 Bronchoconstrict*

#7 Bronch* and (constrict*)

#8 Cough*

#9 Wheez*

#10 Breath*

#11 (bronchial* or (respiratory) or (airway*) or (lung*)) and (hypersensitiv* or (hyperreactiv*) or (allerg*))

#12 OR/1-11

#10 Acoustic signal [MH]

#11 Sound signal [MH]

#12 Sound recording [MH]

#13 Cough detection [MH]

#14 (Breath* and (sound*)) and (detect* or record* or analys* [tiab] or monitor* [tiab] or evaluat* [tiab])

#15 (Wheez* [tiab]) and (detect* [tiab] or record* [tiab] or analys* [tiab] or monitor* [tiab] or evaluat* [tiab])

#16 (Cough* [tiab]) and (detect* [tiab] or record* [tiab] or analys* [tiab] or monitor* [tiab] or evaluat* [tiab])

#17 OR/10-16

#18 Animals[mh] not (humans[mh])

#19 editorial[publication type]

#20 review[publication type] not (systematic review [publication type])

#21 Letter [publication type]

#22 case report

#23 report of a case

#24 a case of

#25 (mice or (mouse) or (rat) or (animal*) or (pig))

#26 ((COPD) or (chronic bronchitis) or (emphysema) or (chronic obstructive pulmonary disease) or (cancer) or (bronchiectasis) or (fibrosis) or (surgery) or (operation)) not (asthma)

#27 OR/18-26

#28 #12 and #17

#29 #28 NOT #27

SHORT VERSION:

(Asthma [MH]) or (Lung Diseases, Obstructive [MH:NOEXP]) or (Anti-asthmatic agents [MH]) or (Asthma [tiab]) or (Bronchospas* [tiab]) or (Bronchoconstrict* [tiab]) or (Bronch* [tiab] and (constrict* [tiab])) or ((bronchial* [tiab] or (respiratory[tiab]) or (airway* [tiab]) or (lung* [tiab])) and (hypersensitiv* [tiab] or (hyperreactiv* [tiab]) or (allerg* [tiab])))

AND

(Acoustic signal [MH]) or (Sound signal [MH]) or (Sound recording [MH]) or (Cough detection [MH]) or ((Breath* [tiab] and (sound* [tiab])) and (detect* [tiab] or record* [tiab] or analys* [tiab] or monitor* [tiab] or evaluat* [tiab])) or ((Wheez* [tiab]) and (detect* [tiab] or record* [tiab] or analys* [tiab] or monitor* [tiab] or evaluat* [tiab])) or ((Cough* [tiab]) and (detect* [tiab] or record* [tiab] or analys* [tiab] or monitor* [tiab] or evaluat* [tiab]))

NOT

(Animals[mh] not (humans[mh])) or (editorial[publication type]) or (review[publication type] not (systematic review [publication type])) or (Letter [publication type]) or (“case report”) or (“report of a case”) or (“a case of”) or ((mice or (mouse) or (rat) or (animal*) or (pig))) or (((COPD) or (chronic bronchitis) or (emphysema) or (chronic obstructive pulmonary disease) or (cancer) or (bronchiectasis) or (fibrosis) or (surgery) or (operation)) not (asthma))

Appendix B

Figure B1. Data extraction strategy followed by authors for the data extraction.