Quantifiable breathing pattern components can predict asthma control: an observational

21 Background: Breathing pattern disorders are frequently reported in uncontrolled asthma. At 22 present, this is primarily assessed by questionnaires, which are subjective. Objective measures of 23 breathing pattern components can provide additional useful information about asthma control. This 24 study examined whether respiratory timing parameters and thoracoabdominal (TA) motion 25 measures could predict and classify levels of asthma control. Methods: 122 asthma patients at STEP 26 2- STEP 5 GINA asthma medication were enrolled. Asthma control was determined by the Asthma 27 Control Questionnaire (ACQ7-item) and patients divided into ‘well controlled’ or ‘uncontrolled’ 28 groups. Breathing pattern components (respiratory rate (RR), ratio of inspiration duration to 29 expiration duration (Ti/Te), ratio of ribcage amplitude over abdominal amplitude during expiration 30 phase (RCampe/ABampe), were measured using Structured Light Plethysmography (SLP) in a sitting 31 position for 5-minutes. Breath-by-breath analysis was performed to extract mean values and within- 32 subject variability (measured by the Coefficient of Variance (CoV%). Binary multiple logistic 33 regression was used to test whether breathing pattern components are predictive of asthma 34 control. A post-hoc analysis determined the discriminant accuracy of any statistically significant 35 predictive model. Results: Fifty-nine out of 122 asthma patients had an ACQ7-item < 0.75 (well- 36 controlled asthma) with the rest being uncontrolled (n= 63). The absolute mean values of breathing 37 pattern components did not predict asthma control (R 2 = 0.09) with only mean RR being a significant 38 predictor (p < 0.01). The CoV% of the examined breathing components did predict asthma control 39 (R 2 = 0.45) with all predictors having significant odds ratios (p < 0.01). The ROC curve showed that 40 cut-off points > 7.40% for the COV% of the RR, > 21.66% for the CoV% of Ti/Te and > 18.78% for the 41 CoV% of RCampe/ABampe indicated uncontrolled asthma. Conclusion: The within-subject 42 variability of timing parameters and TA motion can be used to predict asthma


Abstract 21
Background: Breathing pattern disorders are frequently reported in uncontrolled asthma. At 22 present, this is primarily assessed by questionnaires, which are subjective. Objective measures of 23 breathing pattern components can provide additional useful information about asthma control. This 24 study examined whether respiratory timing parameters and thoracoabdominal (TA) motion 25 measures could predict and classify levels of asthma control. Methods: 122 asthma patients at STEP 26 2-STEP 5 GINA asthma medication were enrolled. Asthma control was determined by the Asthma 27 Control Questionnaire (ACQ7-item) and patients divided into 'well controlled' or 'uncontrolled' 28 groups. Breathing pattern components (respiratory rate (RR), ratio of inspiration duration to 29 expiration duration (Ti/Te), ratio of ribcage amplitude over abdominal amplitude during expiration 30 phase (RCampe/ABampe), were measured using Structured Light Plethysmography (SLP) in a sitting 31 position for 5-minutes. Breath-by-breath analysis was performed to extract mean values and within-32 subject variability (measured by the Coefficient of Variance (CoV%). Binary multiple logistic 33 regression was used to test whether breathing pattern components are predictive of asthma 34 control. A post-hoc analysis determined the discriminant accuracy of any statistically significant 35 predictive model. Results: Fifty-nine out of 122 asthma patients had an ACQ7-item < 0.75 (well-36 controlled asthma) with the rest being uncontrolled (n= 63). The absolute mean values of breathing 37 pattern components did not predict asthma control (R 2 = 0.09) with only mean RR being a significant 38 predictor (p < 0.01). The CoV% of the examined breathing components did predict asthma control 39 (R 2 = 0.45) with all predictors having significant odds ratios (p < 0.01). The ROC curve showed that 40 cut-off points > 7.40% for the COV% of the RR, > 21.66% for the CoV% of Ti/Te and > 18.78% for the 41 CoV% of RCampe/ABampe indicated uncontrolled asthma. Conclusion: The within-subject 42 variability of timing parameters and TA motion can be used to predict asthma control. Higher 43 breathing pattern variability was associated with uncontrolled asthma suggesting that irregular 44 resting breathing is an indicator of poor asthma control. Breathing pattern disorders (also known as dysfunctional breathing) are commonly reported in 67 patients with uncontrolled asthma, even though their relationship (causal or coincidental) has not 68 been clearly determined yet (1,2). Dysfunctional breathing has been characterised as a change in 69 the biomechanical and physiological components of breathing, resulting in intermittent or chronic 70 respiratory and non-respiratory symptoms, which worsens asthma patients' quality of life (3). The 71 most commonly reported respiratory symptoms of dysfunctional breathing are predominant upper 72 thoracic breathing, asynchrony between ribcage and abdominal motion, breathlessness, chest 73 tightness, wheezing and deep sighing (4). However, most of these have been described subjectively 74 through clinicians' observations or using symptom questionnaires, such as the Nijmegen 75 Questionnaire (NQ) (5). The use of the NQ in this way has been criticised due to its reliance on 76 patients' perceptions, and its subjectivity (6,7). Objective measures of quantifiable breathing pattern 77 components are needed to increase our understanding of the complex relationship between 78 breathing patterns, symptoms and asthma control. 79 Breathing pattern comprises components of volume, timing and thoracoabdominal (TA) movements 80 (8). Breathing pattern components, such as tidal volume (Vt), timing parameters (inspiration and 81 expiration duration or their ratio, respiratory rate (RR)) and TA motion, can now be measured non-82 invasively without requiring patients' cooperation as traditional lung function tests do (9,10). 83 Although changes in some of these quantifiable breathing pattern components among asthma 84 patients have been previously reported (11), any relationship of them with different levels of asthma 85 control have not been examined thoroughly. This may lead to a current lack of use of quantifiable 86 breathing pattern components in the evaluation process of asthma control. A positive weak 87 correlation (r=0.33) has been previously reported between TA asynchrony, as measured using 88 Respiratory Inductive Plethysmography (RIP), and Asthma Control Questionnaire (ACQ7-item) (12). 89 In addition, Raoufy et al. (2016) has previously reported that within-subject variability of Vt and 90 breath cycle duration as measured by the RIP, could differentiate uncontrolled asthma patients 91 (n=10) from patients with well-controlled asthma (n=10) as determined by the presence of asthma 92 symptoms. However, there is still a lack of information about the use of other quantifiable breathing 93 pattern components to indicate levels of asthma control. 94 To date, traditional lung function tests primarily provide information about airway calibre and lung 95 volume during single forced expiratory maneuvers. Dynamic breathing pattern measures during 96 resting breathing over time may provide additional information to increase our understanding of 97 their physiological role in the evaluation process of asthma control. Thus, the aim of this study was 98 to establish whether respiratory timing parameters and/ or respiratory TA movements measured 99 using Structured Light Plethysmography (SLP) during resting breathing, could predict asthma control. 100

Methods 101
This observational cross-sectional study recruited 122 adult asthma patients with a range of asthma 102 severity from a difficult-to-treat outpatient clinic at the University Hospital Southampton and from 103 staff and students at the University of Southampton. Individuals with a medical diagnosis of asthma 104 without any other chronic respiratory disease or any upper respiratory tract infection on the day of 105 data collection were eligible for this study. Levels of asthma control were determined by the ACQ7-106 item, and cut-off points < 0.75 and > 1.50 were used to define well-controlled and uncontrolled 107 asthma respectively. Asthma patients with partially-controlled asthma (ACQ7-item scores between 108 0.75 and 1.50) were not included in this study. All participants were between STEP 2 and STEP 5 109 asthma medication according to GINA guidelines (14). 110 After obtaining informed consent, participants' demographic data and medication history were 111 collected. Asthma medication data was used to determine asthma severity. Participants' breathing 112 pattern components were recorded during resting breathing in a seated position and then 113 spirometry (Vitalograph) was performed to evaluate lung function. 114 Breathing pattern components were recorded using the SLP (Thora-3Di TM , Pneumacare Ltd) 115 according to manufacturers' guidelines (15). This is a non-invasive motion-analysis recording 116 system. It comprises a contactless device which projects a grid pattern of light onto an individual's 117 chest wall covering the area between the clavicles and the umbilicus. The distortion of the grid 118 pattern intersection points caused by the displacement of the anterior surface of the chest wall is 119 recorded by two digital cameras. The two digital cameras are attached on the SLP which generates a 120 time-varying output trace. The manufacturer's own software did not allow direct breath-by-breath 121 estimations of ribcage and abdominal amplitudes (RCampe and ABampe). Thus, an automatic peak 122 detection algorithm written in Matlab code and used in our previous research (16)  The automatic algorithm identified local minima and maxima of the inspiration phase for each 126 breath cycle. The RR was defined as the number of complete breath cycles in one minute and the 127 inspiratory/ expiratory phase ratio (Ti/Te) was defined as the proportionality between inspiratory 128 and expiratory phases. The inspiratory time (Ti) was calculated as the time between a minimum in 129 the sum SLP output trace and the next peak. The expiratory time (Te) was calculated as the time 130 between a peak and the next minimum. The ribcage and abdominal amplitudes (RCampe and 131 ABampe) were defined as the vertical distances between a trough and the next peak on the SLP's 132 output as derived from the different SLP's traces used to record the motion of the ribcage and 133 abdomen separately. The within-subject variability of the breathing pattern components was 134 calculated as the Coefficient of Variance expressed as a percentage (CoV%). 135 The patients' breathing pattern components were recorded for 5 minutes at the sitting position. The 136 participants were requested to stay still and quiet during the whole recording procedure. This was to 137 minimise external body movement artefacts on the SLP's output trace as this could bias values of 138 breathing pattern components during data extraction. When patients were ready to be recorded, 139 they were falsely informed about the start of breathing pattern recording. The actual recording time 140 started one minute after the initial notification. This was to eliminate any impact of the patients' 141 awareness on breathing pattern measurements whilst recording natural behavior of their breathing. 142 Descriptive statistics were used to summarise demographic data and lung function measurements 143 Comparisons of the breathing pattern components between well-controlled and uncontrolled 144 asthma groups were made using the Mann-Whitney U test (significance level p < 0.01) as normal 145 distribution of the data was not found. Multiple binary logistic regression, using the forced method, 146 was performed to predict uncontrolled asthma (ACQ7-item > 1.50). Two regression models were 147 applied, one using absolute mean values of RR, Ti/Te and RCampe/ABampe as predictors. The other 148 one involved the within-subject variability measures (Cov%). Both regression models met the 149 assumption of multicollinearity (Variance Inflation Factor < 10). When all predictors of a regression 150 model significantly predicted uncontrolled asthma, a post-hoc analysis using a Receiver Operating 151 Characteristic curve (ROC) was used to identify cut-off points for changes in breathing pattern 152 components distinguishing well-controlled and uncontrolled asthma. 153

154
One hundred twenty two adult asthma patients (75 females) were recruited and completed the 155 study (mean age (sd) 44.75 years (15.98 years). Sixty-three participants had an ACQ score of > 1.5 156 (uncontrolled asthma), whereas 59 participants scored < 0.75 (well-controlled asthma). Thirty-three 157 participants had mild asthma (STEP 2 on GINA asthma medication), with 29 of these being in the 158 well-controlled group while the rest of them had moderate-to-severe asthma (STEP 3, 4 and 5 on 159 GINA asthma medication). There were similar numbers of males and females in both groups (Table  160 1). Both groups also had similar average body mass index (BMI). Those in the uncontrolled asthma 161 group had reduced average lung function compared to the well-controlled asthma group (Table 1). 162 Although those in the uncontrolled asthma group had significantly higher median RR than those in 163 the well-controlled group, no significant differences were found for the other absolute mean values 164 of breathing pattern components (Ti/Te and RCampe/ABampe) ( Table 2). On the other hand, the 165 within-subject variability measures (CoV%) of all the breathing pattern components were found to 166 be significantly increased in the uncontrolled asthma group compared to the well-controlled group 167 (Table 2). 168 When mean values of RR, Ti/Te and RCampe/ABampe were entered into the regression model 169 asthma control was not predictable with only the beta coefficient of RR being significantly greater 170 than zero ( Table 3). When within subject variability measures (CoV%) of breathing pattern 171 components were entered into the model, a good fit was found (Table 4). This accounted for 45% of 172 the variance in the ACQ7-item scores. The beta coefficients of the CoV% of all breathing pattern 173 components were found to be significantly greater than zero suggesting that increased within-174 subject variability of RR, Ti/Te and RCampe/ABampe predicts uncontrolled asthma. A linear 175 relationship was found between the CoV% of all breathing pattern components and the log of the 176 ACQ7-item score with no more than 5% of the total cases being considered as influential cases 177 (standardised residuals > 2) in the specific regression model. the RR discriminated well-controlled from uncontrolled asthma. Optimal cut-off points for the CoV% 185 of Ti/Te and RCampe/ABampe were estimated to be > 21.66% and > 18.96% respectively (Table 5). 186

Discussion 187
The study aimed to examine whether respiratory timing parameters and/ or respiratory TA 188 movements could predict and classify levels of asthma control. The within-subject variability of 189 breathing pattern components, such as RR, Ti/Te and RCampe/ABampe, was found to predict 190 asthma control, but their absolute mean values did not. Based on these findings, the within-subject 191 variability of breathing pattern components is likely to be a better indicator of asthma control than 192 their mean values when measured in a single occasion. This may be because the within-subject 193 variability can efficiently reflect the natural behaviour of tidal breathing over time compared to the 194 absolute mean values of the same respiratory parameters. Therefore, the study's findings suggest 195 that the regularity of resting breathing can be considered as another physiological marker which 196 reflects levels of asthma control. The importance of measuring the natural behaviour of breathing 197 patterns has been previously highlighted as this may reflect better the adaptability of the respiratory 198 system occurred during symptomatic periods of asthma (17). 199 On the other hand, the limited variance we found in the absolute mean values of Ti/Te and 200 RCampe/ABampe may have biased the asthma control prediction. Although the RR was found to be 201 a significant predictor of asthma control, there was a lack of a linear relationship between mean RR 202 and asthma control. All these may be attributed to the presence of confounders previously reported 203 in cross-sectional observational study designs (18,19). Authors' expect that examples of such 204 confounders could be a postural effect, the patients' asthma complexity, the underlying patients' 205 anxiety levels, and an effect of rescue medication usage prior to breathing pattern measurements. 206 Some of these, such as posture and emotions, have been clearly suggested to affect absolute mean 207 values of breathing pattern measurements (18,19,20), but the impact of asthma complexity and 208 medication usage on breathing patterns is not clear yet. 209 Respiratory rate is affected by many factors, and so there was no clear separation between the well-210 controlled and controlled groups for this parameter. Asthma patients frequently have co-existing 211 anxiety which can have an impact on RR (21). There is also a relationship between asthma and 212 obesity (22), and it is well known that BMI can have an impact not only on patients' asthma control 213 but also on timing components of breathing patterns (23). Although levels of anxiety were not 214 assessed in our study, our study's individuals with raised RR and well-controlled asthma were obese 215 (BMI >30 kg/m 2 ). The normal RR found in some individuals of the uncontrolled asthma group is 216 unexplained, but this may have been caused by the use of rescue medication prior to breathing 217 pattern recordings during this study. 218 Raoufy et al. (2016) have previously reported that the within-subject variability of Vt and breath 219 cycle duration can differentiate patients with well-controlled asthma from those with uncontrolled 220 asthma. Our findings are in agreement with Raoufy et al.'s work despite methodological differences, 221 such as the method used to determine asthma control (National Asthma Education and Prevention 222 program vs ACQ7-item), the breathing pattern recording time (60 minute vs 5 minutes), the 223 recording posture (supine vs sitting) and the equipment used to monitor breathing patterns (SLP vs 224 RIP) at rest. 225 The optimal time for recording variability within breathing pattern parameters is not known in the 226 literature. We measured within-subject variability over 5 minutes and found this was sufficient for 227 making significant predictions of asthma control using respiratory rate, proportionality of respiratory 228 phases, and TH motion. To the best of authors' knowledge, the study presented here also provides 229 for a first time specific cut-off points for the within-subject variability of the breathing pattern 230 components, which can be used to differentiate well-controlled from uncontrolled asthma. 231 However, more research is required to confirm the accuracy of our results in the future. 232 In addition, the different posture selected in our study compared to Raoufy et al. (2016) did not 233 seem to have an impact on the ability of within-subject variability of the breathing pattern 234 components to predict asthma control. However, more research involving different postures, such 235 as supine or standing, is required to check maintenance of the identified association between 236 asthma control and within-subject variability of breathing pattern components. 237 Some limitations underlie this research. We did not include patients with partially controlled asthma 238 (ACQ7-item score between 0.75 and 1.50) so that ACQ7-item score could be used as a binary 239 outcome within the recruited sample. A causal or coincidental relationship between within-subject 240 variability and asthma control could not be determined from our findings due to the selected study 241 design. It is not known whether uncontrolled asthma preceded the increased within-subject 242 variability of the breathing pattern components, or vice versa. However, it is assumed that increased 243 within-subject variability in the presence of uncontrolled asthma might be due to physiological, 244 psychological or biomechanical factors as previously observed in the literature (3). In any way, a 245 future prospective cohort study is required to examine the exact nature of the relationship between 246 the changes in quantifiable breathing pattern components and asthma control. 247

Conclusion 248
The study showed that within-subject variability of timing parameters and THA motion predicts and 249 classifies levels of asthma control, but same results were not found for mean values of them. It is 250 concluded that increased within-subject variability of RR, Ti/Te and RCampe/ABampe is associated 251 with uncontrolled asthma. This sheds a light on the use of stable resting breathing as another 252 important marker of asthma control. 253

Consent for publication 266
Patients' anonymous data were agreed to be published for maintaining anonymity and protecting 267 individuals' health data. 268

Availability of data and materials 269
The datasets used and analysed during the current study are available from the corresponding 270 author on reasonable request. 271      >18.78 0.773 0.044 0.686-0.859 0.000* ^Optimal cut-off points were selected as the closest points from the left corner of the individual ROC curves for the CoV% of each breathing parameter;* significant result was defined at p < 0.01