Learn more: PMC Disclaimer | PMC Copyright Notice
Original research
Quantification of breathing irregularity for the diagnosis of dysfunctional breathing using proportional tidal volume variation: a cross-sectional, retrospective real-world study
1,2 Abstract
Objectives
To develop a statistical approach that provides a quantitative index measuring the magnitude of the irregularity of the breathing response to exercise for the diagnosis of dysfunctional breathing.
Design
Cross-sectional, retrospective, real-world study.
Setting
Single-centre study.
Participants
A population of 209 patients investigated with cardiopulmonary exercise testing in our institution for unexplained or disproportionate exertional dyspnoea between January and July 2022.
Primary and secondary outcome measures
A novel statistical approach providing a quantitative index—proportional tidal volume variation (PTVV)—was developed to measure the magnitude of the irregularity of the breathing response to exercise.
Results
PTVV provided a reliable statistical readout for the objective assessment of DB with a prediction accuracy of 78% (95% CI: 72 to 83%). The prevalence of DB in the investigated population was high with more than half of the patients affected by moderate-to-severe DB.
Conclusions
PTVV can easily be implemented in the clinical routine. Our study suggests a possible further simplification for the diagnosis of DB with two objective criteria including PTVV and one single criterion for hyperventilation.
Introduction
Dysfunctional breathing (DB) refers to several disorders in breathing patterns that result in chronic—typically exertional—dyspnoea and other non-respiratory symptoms. DB can occur in the presence or absence of cardiorespiratory diseases.1 Non-respiratory symptoms include dizziness, palpitations, tachycardia and paresthesia. So far, there is neither a generally accepted definition nor a well-established diagnostic method for DB. Since there is no gold standard diagnostic test, it has been difficult to determine the prevalence of DB. In a recent review, Vidotto et al estimated the prevalence of DB in UK adults in primary care to be around 9.5%.2 Since the pathophysiology of DB consists of biomechanical, psychological and physiological aspects of breathing, different forms of DB can be observed. Several studies suggested different categorisations for DB.3 The most commonly known form of DB is the hyperventilation syndrome (HVS).4 Other forms and categorisations of DB have been proposed, including the study from Boulding et al 1 suggesting a categorisation of DB into five categories, of which HVS and periodic deep sighing (PDS) are the main components. In this study, PDS is defined as frequent sighing combined with irregular breathing patterns, making erratic breathing (EB) a central component of DB and an intrinsic element for its diagnosis. A recent publication from Genecand et al also refers to the terms HVS and PDS/DB to categorise DB in patients who had previously suffered from a COVID-19 infection.5
Different methods have been used to diagnose DB. One of the most common ways is the use of the Nijmegen questionnaire.6 However, its value has been disputed by Boulding et al.1 Other diagnostic methods include different questionnaires, such as the Manual Assessment of Respiratory Motion and the Self Evaluation of Breathing Questionnaire, hyperventilation provocation testing, respiratory muscle function testing and cardiopulmonary exercise testing (CPET).1 7–10
CPET is a broadly available diagnostic tool for the assessment of cardiopulmonary diseases.11 12 It seems to be particularly useful for the diagnosis of DB, since it is able to distinguish specific response patterns to exercise for cardiac, respiratory, metabolic and muscular abnormalities13 14 and potentially for DB. A recent study from Frésard et al used CPET to identify long COVID patients with and without DB.15 The study indicated that an erratic breathing frequency (BF) and/or tidal volume (VT) in response to exercise was highly suspicious of DB. CPET was able to distinguish among HVS, EB and mixed types of DB.
In a recent review, Watson et al suggested CPET-based diagnostic criteria for the diagnosis of DB.3 These included two quantitative criteria of hyperventilation and one qualitative, thus subjective, criterion about the irregularity of the respiratory response to exercise. The quantitative criteria to detect hyperventilation are an elevated ventilatory equivalent of CO2 (VeqCO2) of over 35 at 40 watt during exercise and/or a decreased end tidal partial pressure of CO2 (PET CO2) of under 4.0 kPa both at rest and during exercise. The third criterion is the subjective assessment of the irregularity of VT response and BF response to exercise. Even though the diagnostic algorithm of Watson et al showed some promising results regarding the diagnosis of DB, the subjective criterion of breathing irregularity leaves room for interpretation and is, therefore, affected by interexaminer discrepancies. Recent works reported the use of approximate entropy as an objective marker of breathing irregularity potentially useful to detect breathing pattern disorder and DB.16 17 These studies were conceived as proof of concepts and were based on small sample size, needing further validation and additional statistical developments.
The primary objective of the current study was to develop a statistical index for the objective quantification of the irregularity of the breathing response to exercise as a basis for the systematic diagnosis of DB and to compare its performance to standard experts’ evaluations. The secondary objective was to assess the usefulness of this index in a real-world population of patients referred for unexplained or disproportionate dyspnoea.
Material and methods
Study population
The study population consisted of 224 consecutive individuals who underwent CPET at the Cantonal Hospital St. Gallen for unexplained or disproportionate exertional dyspnoea with or without presence of cardiopulmonary conditions between January and June 2022 (figure 1). Data from 15 CPETs (7%) were excluded due to missing raw data or insufficient quality. Thus, 209 CPETs were retained for further analyses. Data used in the current study were extracted retrospectively from electronic clinical reports under general informed consent in place in our institution since 2019. All patients treated at our institution may consent to the further use of their data for research projects. In the current study, data from patients who gave general consent to the further use of their data were extracted and analysed anonymously.
Cardiopulmonary exercise testing
Spirometry was performed according to the American Thoracic Society (ATS) guidelines18 (Jäeger Vyntus CPX, CareFusion, Vyaire Medical, Germany; and SentrySuite V.3.10.8, Vyaire Medical, Germany). The patients were seated on a cycloergometer (ergoselect 200, Typ P, ergoline GmbH, Bitz, Germany), and a 12-lead ECG (ASSY CAM 14 V.2, GE Medical Systems, Wisconsin, USA) was applied (GE CardioSoft V.6.73, GE Medical Systems, Wisconsin, USA). After adapting the mouthpiece, parameters at rest were determined. Following a period of 2 min of unloaded pedalling, a ramp exercise test with a constant workload increment of 10–30 W/min was performed until exhaustion. The aim was to have a test duration of 8–12 min. Capillary blood samples were drawn from earlobes at rest and at peak exercise and analysed immediately. Variables during exercise were analysed breath by breath (Jäeger Vyntus CPX, CareFusion, Vyaire Medical, Germany; and SentrySuite V.3.10.8, Vyaire Medical, Germany) in accordance with the ATS statement on CPET.19
Breath-by-breath data were retrospectively extracted from the CPET software SentrySuite (V.3.10.8). Patient baseline characteristics, including demographics, comorbidities and various lung parameters, were extracted from the patient information system of the Cantonal Hospital St. Gallen. The parameters of interest included VT, BF, VE, P ET CO2 and the partial pressure of carbon dioxide at rest and peak exercise.
Assessment of DB by independent experts
Two senior clinical experts in the field of exercise testing rated the individual CPETs independently and blinded to the results of the statistical methodology. In the current study, DB was diagnosed based on the criteria of Watson et al 3 including on the one hand the components of ventilatory irregularity and on the other hand the component of hyperventilation. The irregularity component was classified subjectively by the experts as no, mild, moderate and severe and was further dichotomised as yes (severe, moderate) and no (no, mild). The categorisation of the breathing irregularity was strictly based on qualitative considerations. Experts evaluated the graphical representation of the ventilation slopes (VT vs minute ventilation (VE)) and subjectively assessed the level of irregularity. The experts also evaluated the presence or absence of hyperventilation following Watson et al’s criteria.3 In a second step, a consensus conference between the two experts took place for the assessment of presence/absence of DB. Both experts agreed on a consensus for the cases with differing assessments. This consensus assessment was assumed as the current clinical gold standard for DB evaluation. The consensus DB diagnosis was compared with the proposed diagnostic algorithm.
Statistical considerations
The irregularity of the respiratory response to exercise—thereafter named proportional tidal volume variation (PTVV)—was defined from the normalised residual variation of a non-linear regression model fit to the ventilation slopes describing the relationship between the VT and the VE. The general form of the non-linear regression model is described as follows:
with the function being a three-parameter log-logistic non-linear regression model, the vector of model parameters and the residual error term. The three-parameter log-logistic model is defined as follows:
The vector of model parameters includes the following three parameters: the steepness of the curve, the lower asymptote and the upper asymptote.
PTVV is defined as the normalised root mean squared error defined as follows:
where is the number of observations.
The DB classification accuracy of PTVV in comparison to the clinical gold standard (expert assessment) was evaluated using receiver operating characteristics (ROC) curves. The optimal PTVV cut-off was extracted using the Youden statistics (Youden Index) defined as followed:
The current sample size was justified based on the expected precision of the true positive rate estimate of PTVV compared with the clinical gold standard. Assuming a true positive rate for PTVV of 80%, a sample size of 224 patients would provide a CI around the parameter estimate with a half-width of approximately 5%, which was deemed acceptable.
All analyses were done using the R statistical software20 including the extension packages drc,21 INDperform 22 and ROCit.23
Patient and public involvement
Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.
Results
Characteristics of the study population
CPET breath-by-breath data from 209 patients were analysed. Patient’s characteristics are shown in table 1. Patients had a median age of 56 years, 50% were men with a median body mass index of 26 kg/m2 and had a median forced expiratory volume in the first second (FEV1) of 79% predicted. The context of CPET included a majority of patients with long COVID.
Table 1
Patients characteristics.
| Characteristics | |
| Cases, n | 209 |
| Age (years), median (range) | 56 (17–87) |
| Gender, male/female | 104/105 |
| BMI (kg/m2), median (IQR) | 26 (23–30) |
| FEV1 (% predicted), median (IQR) | 79 (60–94) |
| Context of CPET | |
| Long COVID | 89 (45%) |
| Lung cancer | 31 (15%) |
| Sleep apnoea | 38 (18%) |
| Asthma | 35 (17%) |
| COPD | 45 (22%) |
Results are reported as median and IQR, unless otherwise specified.
BMI, body mass index; COPD, chronic obstructive pulmonary disease; CPET, cardiopulmonary exercise testing; FEV1, forced expiratory volume in the first second.
Expert evaluation
The assessment of two independent experts were in agreement in 81% of cases with respect to the presence or absence of DB. From the consensual evaluation of 209 CPET reports, the experts diagnosed 119 participants (57%) with moderate-to-severe breathing irregularity in response to exercise. In addition, among the 209 patients, 45 patients (22%) were identified with hyperventilation. The overall prevalence of DB (based on irregularity and hyperventilation) was 62% (95% CI: 55 to 69%). We observed a large overlap between the patients showing a significant irregularity and those having hyperventilation (figure 2A). 76% (34/45) of patients with hyperventilation also had breathing irregularities. The number of patients with hyperventilation without breathing irregularity in the response to exercise was 11 (5%). Thus, the irregularity criterion was the main criterion for the diagnosis of DB affecting 92% (119 out of 130) of patients with significant DB. In the current cohort of patients, all 45 patients with a significant hyperventilation fulfilled simultaneously both hyperventilation criteria (figure 2B) which may indicate a possible redundancy among the two hyperventilation criteria suggested by Watson et al.3 Finally, the diagnoses of patients with unexplained dyspnoea without DB included a majority of the patients with long COVID (37%), asthma (20%) and chronic obstructive pulmonary disease (17%).
Venn diagram and UpSet plots of the different diagnostic criteria for dysfunctional breathing. (A) The Venn diagram of the dysfunctional breathing status in our cohort of patients. The number of patients with a significant irregularity and/or hyperventilation is represented. (B) An UpSet plot to visualise the three intersecting criteria for the assessment of dysfunctional breathing.
DB prediction by PTVV
Log-logistic non-linear regression models were fitted to the 209 ventilation slopes (VT vs VE) and the normalised root mean square errors (PTVV) were derived from the model fits.
Two examples of non-linear regression models fitted to the ventilation slopes of patients with or without significant DB are shown in figure 3.
Examples of non-linear regression model fits to ventilation slopes. (A) A typical example of a patient with a significant moderate-to-severe dysfunctional breathing. (B) An example of a patient without significant dysfunctional breathing. The proportional tidal volume variation values and scales are shown in the lower right corner of the plots. VE, minute ventilation; VT, tidal volume.
The distribution of PTVV is shown in figure 4A. The classification accuracy of PTVV in comparison to the expert assessment is graphically summarised in the ROC curve, representation shown in figure 4B. The area under the curve was 0.835. The optimal PTVV cut-off defined by the Youden index was 0.154. The prediction accuracy of DB obtained using the optimal PTVV cut-off was 78% (95% CI: 72% to 83%) for the expert consensus gold standard.
Distribution and classification accuracy of proportional tidal volume variation (PTVV). (A) The histogram distribution of PTVV. (B) The receiver operating characteristics curve (ROC) for the evaluation of the classification accuracy of PTVV with an optimal cut-off value of 0.154.
The overall prevalence of DB assessed by PTVV was 55% (95% CI: 48% to 62%), that is, slightly less than what was assessed by the experts. This was to be expected considering that PTVV only assesses the irregularity of the ventilatory response to exercise independently of the hyperventilation criteria. Notice that PTVV still predicted a significant DB in 3 out 11 patients (27%) with pure hyperventilation as assessed by the experts.
Discussion
In the absence of a diagnostic gold standard, DB is difficult to diagnose and often unrecognised. The assessment of DB by two independent experts is still highly prone to subjectivity. With PTVV, we developed and validated an objective methodology to quantify breathing irregularity as a basis for the diagnosis of DB. The methodology underlying PTVV could lay the ground for an easier recognition and quantification of DB, which could also be used for follow-up measures and treatment response. PTVV, as a quantitative biomarker, could also set the foundation for the scientific refinement of today’s relatively sparse treatment options.
In our preselected population of patients undergoing CPET for unexplained or disproportionate dyspnoea and in a variety of clinical contexts, DB was identified in about 60% of cases. As expected, this is much higher than reported from prevalence studies investigating the general population or cohorts with specific diseases. The identification of DB not only lent an explanation for their symptoms, obviating the need for further diagnostic investigations, but also represented a starting point for targeted treatment interventions.
From the three diagnostic criteria for DB described by Watson et al,3 the irregularity of the ventilatory response plays the most important role. In our study, roughly 92% of DB cases could be diagnosed using this single irregularity criterion. Hyperventilation rarely occurred without breathing irregularity, giving less weight to the two hyperventilation criteria. In addition, in the current dataset, the two hyperventilation criteria (high VeqCO2 and low PET CO2) were redundant—all patients having the first hyperventilation criterion (ventilatory equivalent CO2) also had the second criterion (pressure of end-tidal CO2) fulfilled. Thus, the number of criteria used to diagnose DB could be reduced to PTVV and one single hyperventilation criterion, among which the low level of PET CO2 at rest and exercise seems to be the simplest to assess.
The use of PTVV offers several advantages to the examiner who wants to quantify the magnitude of the irregularity of the ventilatory response to exercise. First, it is easier to quantify DB by the means of a single index and cut-off value rather than interpreting ventilation slopes, which require a trained eye and is prone to subjectivity. Second, it saves time, because all DB criteria can be measured or calculated and therefore, do not need to be derived from graphs. At last, PTVV minimises interexaminer discrepancies, as there is no room for interpretation.
The main limitation of the current study is the lack of gold standard for the diagnosis of DB. There are no definitive recommendations and the current study strictly relied on the diagnostic criteria proposed by Watson et al which only reflect the authors’ own opinions. Another limitation of the study is the lack of a healthy control group. Since the study population consists only of patients referred for a CPET in a single lung centre, it mostly consists of participants with cardiopulmonary diseases. A healthy control group of significant size would have helped to better distinguish between normal breathing patterns and DB breathing patterns. The abnormality of EB is still debated, and the definition of the normality of tidal breathing has remained a challenging issue over the last decades.24 Furthermore, the prevalence of this study does not represent the prevalence of DB in the general population due to selection bias—patients were preselected for unexplained and/or disproportionate dyspnoea. Another limitation was the setting of cut-offs. The cut-offs were set based on the study population. As the study population is not representative of the general population, the chosen cut-offs may be associated with a risk of overfitting. In order to prevent overfitting, the analysis of additional independent datasets may be needed.
In conclusion, the CPET-based PTVV index provides a powerful, easy and objective tool to identify ventilatory irregularities, one important cornerstone to diagnose DB. The prediction accuracy of PTVV for the identification of ventilatory irregularities was high and comparable to the concordance rate of independent clinical experts. Only few individuals diagnosed with DB according to Watson et al’s criteria exclusively had a positive hyperventilation criteria, and the two hyperventilation criteria were redundant. If confirmed by further studies, we propose to reduce the diagnostic criteria for DB to PTVV and a single criterion of hyperventilation. PTVV can be easily implemented as a diagnostic plugin for CPET analyses. The prevalence of DB in individuals with unexplained or disproportionate dyspnoea referred for CPET in our institution was about 60%. Considering its burden for everyday life, DB merits more attention by healthcare professionals. Once diagnosed, patients have an explanation for their symptoms and can profit from a targeted therapeutic intervention. In a follow-up project, we are planning to investigate a much larger retrospective cohort of patients over the last decade and to evaluate the evolution of the prevalence of DB over the years prior, during and after the COVID-19 pandemic, in patients with different cardiopulmonary diseases.
Supplementary Material
Footnotes
Contributors: MB conceived the study. MB acted as the guarantor of the study. MB and FB designed the study. GK and FU conducted the data collection. FB and MB contributed in the funding of the study. FB and GK drafted the manuscript together with MB and LK. MB and LK provided their clinical expertise. All authors contributed to the interpretation of data and approved the final version of the manuscript.
Funding: The study was supported by an unconditional grant from the Lungenliga St. Gallen—Appenzell and an institutional grant from the Cantonal Hospital St. Gallen.
Competing interests: None declared.
Patient and public involvement: Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review: Not commissioned; externally peer reviewed.
Data availability statement
Data are available on reasonable request. Anonymised data underlying the results reported in this article will be available to researchers upon reasonable request. Proposals should be directed to the corresponding author.
Ethics statements
Patient consent for publication
Not applicable.
Ethics approval
This study involves human participants and was approved by Ethikkommission Ostschweiz (EKOS 22/063). Data used in the current study were extracted retrospectively from electronic clinical reports under general informed consent in place in our institution since 2019. All patients treated at our institution may consent to the further use of their data for research projects. In the current study, data from patients who gave general consent to the further use of their data were extracted and analysed anonymously. The study was conducted in accordance with the principles enunciated in the Declaration of Helsinki and the guidelines of Good Clinical Practice.