Abstract
Variability of peak flow measurements has been related to clinical outcomes in asthma. We hypothesised that the entropy, or information content, of airway impedance over short time scales may predict asthma exacerbation frequency.
66 patients with severe asthma and 30 healthy control subjects underwent impulse oscillometry at baseline and following bronchodilator administration. On each occasion, airway impedance parameters were measured at 0.2-s intervals for 150 s, yielding a time series that was then subjected to sample entropy (SampEn) analysis.
Airway impedance and SampEn of impedance were increased in asthmatic patients compared with healthy controls. In a logistic regression model, SampEn of the resistance at 5 Hz minus the resistance at 20 Hz, a marker of the fluctuation of the heterogeneity of airway constriction over time, was the variable most strongly associated with the frequent exacerbation phenotype (OR 3.23 for every 0.1 increase in SampEn).
Increased airway impedance and SampEn of impedance are associated with the frequent exacerbation phenotype. Prospective studies are required to assess their predictive value.
Acute exacerbations of asthma account for much of the morbidity and mortality associated with this condition [1]. However, there is no currently available biomarker that can accurately predict the risk of future exacerbations. Previous studies have suggested that a geometrically self-similar airway tree may confer increased risk of asthma exacerbations and that fatal asthma is associated with a reduction in the structural complexity of the airway tree [2]. Similarly, the ventilation heterogeneity observed in asthma follows power law behaviour, which predicts catastrophic closure of small airways [3]. Therefore, characterising structural complexity may have utility in predicting asthma exacerbations.
It has been speculated that the temporal variability in lung function may also exhibit self-similarity at multiple time scales [4]. This would suggest that monitoring lung function over short time scales may provide insights into lung function variability over longer time scales of weeks to months, thus providing a more practical predictive tool for exacerbations. A number of tools have been utilised to characterise time series properties of physiological signals, including those that predict scaling and power law behaviour of information over multiple time scales and those that predict the probability of information repeating itself within a time series [5, 6]. Fluctuations and power law behaviour observed in a time series of lung function measurements such as peak expiratory flow (PEF) may predict poor asthma control or exacerbations [7, 8]. Thamrin et al [9] found that the degree of long-range correlation (self-similarity at different temporal length scales) in PEF measurements appeared to provide additional predictive information with respect to exacerbations in mild-to-moderate asthma, but less so in severe asthma.
The forced oscillation technique (FOT) [10] provides an ideal tool for measuring airway function over time, as it allows the respiratory system to be interrogated at a high temporal resolution by delivering forced oscillations to the airways and measuring the impedance (incorporating resistance and reactance components) of the respiratory system. Dynamics of FOT time series over very short time scales (e.g. minutes) may provide additional information that predicts the behaviour of the airways over longer time scales. For instance, Que et al. [11] plotted frequency distributions of the natural logarithm of respiratory system impedance (lnZrs), measured six times per second over a 15-min period, and found that both the mean and standard deviation of the lnZrs were higher in patients with asthma than in healthy controls. Furthermore, in healthy controls, unloading of the airway smooth muscle (ASM) induced by adoption of a supine posture in conjunction with increased ASM activation induced by methacholine (MCh) challenge led to a significant increase in the standard deviation of lnZrs, recapitulating the fluctuating behaviour observed in the asthmatic airway [11]. The authors concluded that asthma may be associated with not only generalised airway narrowing but also an increased appearance of statistically unlikely airway configurations.
Entropy, a measure of increased irregularity and statistically unlikely configurations, has been utilised to characterise a variety of physiological signals [12]. Entropy measurements give a statistical probability that a series of points within a physiological signal will repeat themselves at a subsequent time-point, within a given tolerance [13]. The entropy of FOT time series was recently investigated by Veiga et al. [14], who found that the entropy of airflow time series was reduced in patients with asthma compared with healthy controls, and that reduced entropy was associated with increased severity of airflow obstruction. However, the possible associations between the entropy of airway impedance measurements and other clinically important patient-centred outcomes in asthma, including asthma control and exacerbations, have not been evaluated.
We hypothesised that: 1) severe asthma is characterised by altered entropy of airway resistance and reactance; and 2) entropy of impedance time series is related to exacerbation frequency in severe asthma.
METHODS
Subjects
66 patients with severe asthma fulfilling the American Thoracic Society (ATS) criteria for asthma, and with no other respiratory disease, were recruited from Glenfield Hospital (Leicester, UK) outpatients. 33 patients were at Global Initiative for Asthma (GINA) treatment step 4 and 33 at GINA treatment step 5, as previously defined [15]. 30 healthy control subjects with similar demographics were recruited from hospital staff and by local advertising. Healthy subjects had no history of respiratory disease, and had normal spirometry and MCh responsiveness. All subjects were over the age of 18 yrs. The Leicestershire and Rutland ethics committee (Leicester) approved the study and all subjects gave written informed consent.
Subject characterisation
Patients with asthma completed the Asthma Control Questionnaire (ACQ) and Asthma Quality of Life Questionnaire (AQLQ). The number of severe asthma exacerbations suffered in the previous year was recorded. Severe exacerbations were defined as an acute worsening of asthma symptoms requiring treatment with high-dose systemic corticosteroids for ≥3 days [16]. Spirometry was performed according to ATS/European Respiratory Society (ERS) guidelines [17]. In particular, long-acting bronchodilators were withheld for 12 h prior to testing and short-acting bronchodilators for 4 h. Sputum induction and cell counting were performed as previously described [18]. Exhaled nitric oxide fraction at an expiratory flow rate of 50 mL·s−1 (FeNO50) was measured using a chemiluminescence analyser (NIOX; Aerocrine, Stockholm, Sweden), according to ATS/ERS guidelines [19].
Measurement of respiratory impedance
Impedance testing was undertaken using a Jaeger MasterScreen Impulse Oscillometry (IOS) system (Viasys Healthcare GmbH, Hoechberg, Germany), according to standard guidelines [20]. IOS was performed in the 5–35-Hz frequency range, with impulses triggered every 0.2 s for 150 s. At each time point, the resistance (R) and reactance (X) components of impedance were recorded at multiple frequencies from 5 to 35 Hz, yielding time series of each variable containing ∼750 data points. IOS was performed both before and after the administration of inhaled salbutamol (400 μg), delivered via a metered-dose inhaler and spacer.
A variety of measures of airway calibre were derived from the mean values of the 750 data points captured (table 1): 1) R at 20 Hz (R20), which we interpret as a measure of the mean level of airway constriction within the bronchial tree; 2) R at 5 Hz (R5) minus R20 (R5–R20), which we interpret as a measure of the heterogeneity of airway narrowing throughout the bronchial tree [21]; and 3) the area under the curve of the reactance spectrum between 5 Hz and resonant frequency (reactance area; AX), which we interpret as a measure of the heterogeneity of airway closure throughout the bronchial tree [22].
Examples of time series of R20 and R5–R20 at baseline are shown in figure 1, in a patient with severe asthma and a healthy control subject.
Examples of airway impedance time series. a) Resistance at 20 Hz (R20) in a healthy control subject; b) resistance at 5 Hz minus R20 (R5–R20) in a healthy control subject; c) R20 in a patient with severe asthma; d) R5–R20 in a patient with severe asthma.
Sample entropy analysis
We used the sample entropy (SampEn) algorithm to evaluate complexity in the respiratory impedance time series, using a custom program downloaded from the PhysioNet online resource [23]. This measure relies on the identification of recurrent patterns within a nonstationary dynamic time series, as described in detail in the online supplementary material. Within a highly regular system, sequence matches are of greater frequency, implying lower entropy and less complexity. SampEn has emerged as a less biased metric of variability than the alternative measure approximate entropy, and is relatively independent of record length, since it does not incorporate self-matches within the time series into the calculation of conditional probability [13]. We interpret SampEn of the core airway impedance markers (R20, R5–R20 and AX) to be a measure of fluctuation as a function of time (table 1). A higher value for SampEn equates to greater levels of temporal fluctuation and vice versa. Further details may be found within the online supplementary material.
Statistical analysis
Statistical analysis was performed using Prism version 5 (GraphPad, San Diego, CA, USA) and SPSS version 16.0 (SPSS Inc., Chicago, IL, USA). Colour maps were produced using Matlab R2007b (Mathworks Inc., Boston, MA, USA). Parametric data are presented as mean±sem, data that were log-normally distributed were log transformed and are presented as geometric mean (95% CI), and nonparametric data are presented as median (interquartile range). Unmatched groups were compared using one-way ANOVA with Bonferroni’s correction or unpaired t-tests for normally distributed data, Chi-squared test or Fisher’s exact test for ratios, unpaired t-tests of log-transformed data for log-normally distributed data, and the Kruskal-Wallis test with Dunn’s correction or the Mann–Whitney U-test for nonparametric data. Matched groups were compared using Friedman’s test with Dunn’s correction. A p-value of <0.05 was taken as the threshold for statistical significance. Spearman’s correlation coefficient was used to determine the degree of correlation between impedance parameters, and principal component analysis with varimax rotation and Kaiser normalisation was used as a data reduction tool. Logistic regression analysis was performed using block entry with all selected independent variables entered at the first step.
RESULTS
Subject demographics
The clinical and demographic characteristics of the study population are shown in table 2. The groups were well-matched for sex but differed significantly with respect to age and body mass index (BMI). However, there were no significant differences between the two asthmatic GINA 4 and GINA 5 cohorts.
Severe asthma is characterised by increased heterogeneous airway constriction and closure
Median values of heterogeneous airway constriction (R5–R20) and closure (AX) at baseline were significantly raised in both asthma groups compared with healthy controls, as shown in table 3 (R5–R20: healthy controls 0.035 kPa·L−1·s, GINA 4 asthma 0.08 kPa·L−1·s (p<0.01), GINA 5 asthma 0.14 kPa·L−1·s (p<0.001); AX: healthy controls 0.33 kPa·L−1, GINA 4 asthma 0.935 kPa·L−1 (p<0.05), GINA 5 asthma 1.8 kPa·L−1 (p<0.001)). The mean level of airway constriction (R20) was also raised in the asthma groups compared with controls, but this effect was much less pronounced and only reached statistical significance for healthy controls versus GINA 5 asthma patients. The GINA 4 and 5 asthma groups did not differ significantly with respect to any parameter.
Severe asthma is characterised by increased fluctuation (entropy) of airway impedance
Results relating to the entropy of impedance measurements closely mirrored those of the impedance measurements themselves, as shown in table 4. There was a progressive increase in the median baseline values of each parameter moving from the control to the GINA 4 asthma to the GINA 5 asthma groups. For instance, median SampEn of AX at baseline was 0.42 in controls, 1.05 in GINA 4 asthma and 1.19 in GINA 5 asthma (p<0.0001). These observations suggest that the heterogeneity of both airway constriction and closure persists over short time scales in patients with asthma. In contrast, the mean level of airway narrowing did not fluctuate significantly in patients with asthma when compared with healthy controls.
Airway impedance and entropy measures are associated with frequent exacerbations in severe asthma
The patients with asthma were divided into those who had infrequent exacerbations, defined as fewer than two exacerbations during the previous year (n=25), and those who had frequent exacerbations, defined as two or more exacerbations during the previous year (n=41), as shown in table 5. The two groups did not differ significantly with respect to age, GINA category, smoking history, duration of disease, post-bronchodilator forced expiratory volume in 1 s (FEV1) % predicted or FEV1/forced vital capacity ratio, sputum eosinophil or neutrophil counts, FeNO50 or AQLQ score. However, frequent exacerbators were significantly more likely to be female than infrequent exacerbators, had a significantly increased mean BMI and had a higher mean ACQ score.
All six impedance and impedance entropy parameters were significantly raised at baseline in frequent exacerbators compared with infrequent exacerbators, as shown in table 6. For example, median R5–R20 was 0.07 kPa·L−1·s in infrequent and 0.13 kPa·L−1·s in frequent exacerbators (p=0.0065), while median SampEn R5–R20 was 0.014 in infrequent and 0.114 in frequent exacerbators (p=0.0004). However, these differences were much less pronounced following bronchodilator administration (table S1).
SampEn R5–R20 is independently associated with frequent exacerbations in severe asthma
Close correlations were found between impedance and impedance entropy measurements, with r>0.5 and p<0.01 for every combination of parameters, as shown in table S2 and figure S4. Principal component analysis of all the continuous variables that were associated with frequent exacerbations showed that most impedance and impedance entropy measurements loaded onto the same factor, as shown in table S3.
A logistic regression model was constructed with the presence or absence of frequent exacerbations, defined as two or more exacerbations within the previous year, as the dependent variable. Sex, BMI and ACQ score were entered into the model as independent variables, as well as one of the impedance or impedance entropy parameters. The most favourable model, chosen on the basis of the greatest predictive value in this dataset (table S4), was that incorporating SampEn R5–R20. The parameters of this model, which correctly classified 74.2% of asthma patients as having frequent or infrequent exacerbations, are shown in table 7. SampEn R5–R20 was the only variable significantly associated with frequent exacerbations in this model (p=0.016), with an odds ratio of 3.23 for every 0.1 increase in SampEn R5–R20.
DISCUSSION
We have shown for the first time that increased heterogeneity of airway constriction and closure (R5–R20 and AX), and increased fluctuation of these biomarkers over time (SampEn) are associated with an exacerbation-prone phenotype in patients with severe asthma. This association appears to be strongest with SampEn R5–R20, a putative marker of time-varying fluctuation of heterogeneous airway constriction. We have established that impedance time series entropy measurements are closely correlated with raw impedance values, suggesting that the raw values alone may provide useful prognostic information, but that time series analysis could be of additional value. Our data suggest that airway closure and narrowing (in particular, heterogeneous time fluctuation behaviour) identifies patients at the greatest risk of exacerbations. Indeed, a 0.1 increase in sample entropy of heterogeneous airway constriction (R5–R20) was associated with a 3.2-fold increase in the risk of an exacerbation. It has previously been shown that minimal heterogeneity of airway constriction can result in catastrophic shifts in ventilation to particular lung regions [24]. We speculate that the fluctuations in R and X we observed in patients with severe asthma represent such step-wise shifts in airway patency, occurring over a time scale of seconds, and that patients who exhibit such fluctuations may be constantly on the cusp of asthma exacerbations. In such patients, small perturbations in smooth muscle tone, for instance caused by minor allergen exposure or a viral upper respiratory tract infection, may result in life-threatening airway constriction and closure.
Significance of the impedance parameters R20, R5–R20 and AX
It has previously been shown that in severe asthma, the general level of airway resistance is raised, and that there is an additional increase in R at low oscillation frequencies [25]. This frequency dependence of R is thought to be due to heterogeneous airway constriction, since mathematical modelling indicates that even severe homogeneous constriction would produce an elevated baseline R but not frequency dependence [25]. Furthermore, an image-functional modelling approach has suggested that the ventilation defects and frequency-dependence of R and elastance seen in asthma can only be explained by small airway constriction, or a combination of large and small airway constriction [26]. In this study, we used R20 to represent general airway resistance and R5–R20 to represent frequency-dependence of resistance, a marker of heterogeneous airway constriction. AX is thought to be a marker of airway closure, since such closure results in peripheral airway capacitive properties not being measured, thus increasing the effective elastance of the respiratory system [22].
Impedance and impedance entropy parameters in healthy subjects and patients with asthma
In line with previous studies [27, 28], we found significant increases in R5–R20 and AX, and to a lesser extent R20, in patients with severe asthma compared with healthy controls, suggesting that the baseline level of airway narrowing, and the accompanying heterogeneity of narrowing and closure, are increased in patients with severe asthma. We were unable to demonstrate a difference in these parameters between patients with severe asthma established on maintenance oral corticosteroids (GINA treatment step 5) and patients on high-dose inhaled corticosteroids (GINA treatment step 4), suggesting that these processes may be resistant to corticosteroid therapy.
The SampEn parameters were higher at baseline in the asthma groups than in the control group, although this only reached statistical significance for the GINA 5 versus control comparison. Close correlations were seen between impedance measurements and their respective SampEn parameters, as shown in table S2 and figure S4, suggesting that entropy of airway impedance is strongly associated with baseline airway calibre.
Determinants of the exacerbation-prone asthma phenotype
In our cohort of patients with asthma, we found that those who had suffered two or more exacerbations in the previous year were significantly more likely to be female, and had a significantly greater BMI and ACQ score. Obesity [29, 30] and reduced asthma control [29] have previously been associated with asthma exacerbations, although a recent large observational study of asthma outcomes [29] did not find female sex to be a significant predictor of having had a recent severe exacerbation. Interestingly, we found that markers of eosinophilic airway inflammation were no higher in patients with frequent compared with infrequent exacerbations, suggesting that the excess exacerbations in the former group were due to factors other than uncontrolled eosinophilic airway inflammation. Of note, all patients in this study were being treated with high-dose inhaled or long-term oral corticosteroids, which would be expected to suppress eosinophilic airway inflammation [31].
Previous studies have shown that fixed airflow obstruction is a risk factor for frequent exacerbations [32]. Although post-bronchodilator FEV1 was numerically lower in our patients with frequent exacerbations than those without frequent exacerbations (78.2 versus 85.8% pred), this result did not reach statistical significance. In contrast, all three of the impedance parameters at baseline, namely R5–R20, AX and R20, were significantly higher in those with frequent exacerbations compared with those without. Very similar results were obtained for the SampEn of impedance parameters. However, as shown in table S1, post-bronchodilator values of the parameters were much less discriminatory, suggesting that pre-bronchodilator values may be more valuable as predictors of clinical outcome. This is concordant with the results of Shi et al. [33], who found that pre-bronchodilator impedance parameters could more effectively discriminate between children with well-controlled and poorly-controlled asthma than post-bronchodilator values.
A logistic regression model (table 6]) showed that SampEn R5–R20 was independently associated with frequent exacerbations. However, substitution of this with any one of the other impedance or impedance entropy parameters resulted in similar model performance (table S4). This suggests that calculation of the SampEn of impedance time series may provide only a small amount of additional information over and above the mean values of the impedance parameters themselves.
Limitations of the study
We have shown that increased airway impedance parameters are associated with frequent asthma exacerbations. However, our conclusions are based on retrospective data, and thus prospective studies are required in order to validate the predictive capacity of the impedance measurements. Nevertheless, our results suggest that airway impedance measurements represent a marker of exacerbation risk that may be used in clinical practice or as an outcome measure in clinical trials.
In addition, further work is required to interpret the structural basis of the impedance parameters derived from impulse oscillometry (R20, R5–R20 and AX). Inverse modelling approaches [21, 22] and ex vivo airway models may provide a deeper understanding.
Finally, in this study we chose to examine the sample entropy of impedance time series as a marker of the temporal fluctuation of airway calibre. However, a number of other techniques exist to analyse time series data, including de-trended fluctuation analysis, a measure of long-range scaling [5, 6], and dynamic systems analysis [34], and it is possible that one or more of these alternative techniques may provide greater predictive power. Future studies should investigate this possibility and further refine these novel fluctuation biomarkers against clinically important patient-centred outcomes.
Footnotes
This article has supplementary material available from www.erj.ersjournals.com
Statement of Interest
Statements of interest for S. Siddiqui and C. E. Brightling can be found at www.erj.ersjournals.com/site/misc/statements.xhtml
- Received December 30, 2011.
- Accepted February 13, 2012.
- ©ERS 2012
REFERENCES
