## Abstract

Because of the minimal demand for cooperation by the subject, the forced oscillation technique is increasingly employed in routine lung function testing. However, comprehensive and device-independent values of respiratory impedance at baseline and after bronchodilation have not been established for healthy adults.

The aim of this multicentre study was to collect impedance data from 4 to 26 Hz in healthy Caucasian subjects between 18 and 80 years of age. Five different devices were employed to assess baseline values and the bronchodilator response.

Altogether, 368 subjects were examined. Despite adjustment for anthropometry, the impedance spectra differed in frequency dependence between the centres, and hence could not be pooled. However, resistance at all frequencies except 20 and 25 Hz, and the low-frequency (≤14 Hz) values of reactance did not exhibit a centre dependence. The regression equations for resistance reflected a greater height dependence in males and a greater weight dependence in both males and females than those published previously. Bronchodilation resulted in a statistically significant decrease (11%) in resistance and a 95th percentile equal to a 32% decrease in resistance at low frequency.

We conclude that rigorous calibration procedures should be developed to ensure data compatibility. Furthermore, new reference equations based on different setups are recommended to replace those established with a single device.

## Abstract

**New prediction equations for respiratory impedance in adults have been developed based on multiple devices** http://ow.ly/p9t4e

## Introduction

Measurement of the mechanical properties of the respiratory system using the forced oscillation technique (FOT) is increasingly employed in lung function laboratories. The main advantage of the FOT is ease of application: the oscillatory signal is superimposed on spontaneous breathing and, hence, no special breathing manoeuvres are required.

This minimal demand for cooperation has made the FOT especially attractive in paediatric lung function testing, where a number of studies have been performed to establish normative data using various FOT devices [1, 2], the results proving largely coherent. In contrast, there have been only a few reports on normative respiratory impedance (*Z*_{rs}) data in healthy adults, and some involved only relatively young or only elderly subjects [3, 4], or only a small number of oscillation frequencies were investigated [5]. Moreover, the criteria for selection of the subjects were not always reported or the sample population was limited to a rather narrow specific subgroup of subjects [3, 6]. Equally important, most such normative studies were conducted in the same laboratory, using the same forced oscillation setup, the Oscillaire, developed by Làndsér and co-workers [3, 6–8], which is no longer available.

Although international guidelines have been developed concerning use of the FOT in clinical practice [1], appropriate reference data derived from a healthy population must be available prior to clinical use, preferably obtained by using equipment and measurement procedures similar to those employed in the clinical setting. Additionally, data on the bronchodilator response in healthy subjects should be collected for the appropriate evaluation of reversibility tests in patients.

The goal of the present study was therefore to collect baseline values of *Z*_{rs} and the bronchodilator response with different FOT devices, and to develop reference ranges of respiratory resistance and reactance as a function of frequency (*f*) for healthy subjects aged 18–80 years. With this aim, three commercial FOT devices and two custom-made FOT setups were used in this five-centre (C1–C5) study for the measurement of *Z*_{rs}.

## Methods

### Subjects

The healthy, nonsmoking Caucasian adults (≥18 years of age) included in the study had no history of pulmonary or cardiac disease, and no current wheezing, breathlessness, cough, phlegm production, hyperresponsive airways or recent respiratory tract infection. Ex-smokers with a smoking history of >10 pack-years were excluded. Each centre was requested to include at least five subjects of each sex per decade of age, with a total number of ≥60 subjects per centre. Except for Perth (Australia), where the invited subjects had previously taken part in the Busselton Health Studies [9], all the centres recruited subjects from among the local hospital or university staff and their family members. All subjects gave their written informed consent and the protocol was approved by the local ethical committees.

### Measurements

#### Impedance measurements

*Z*_{rs} was measured using two home-made setups in Antwerp, Belgium (C1) and Szeged, Hungary (C3), while ROS Oscilink (SensorMedics, Bilthoven, the Netherlands), i2M (Chess mT, Ghent, Belgium) and IOS (Jaeger, Würzburg, Germany) devices were used in Enschede, the Netherlands (C2), Perth, Australia (C4) and Maastricht, The Netherlands (C5). The measurements were made in accordance with recent European Respiratory Society (ERS) guidelines [1]. The averages and standard deviations of three to five technically acceptable *Z*_{rs} measurements from each subject were retained for further analysis. Specifications on the oscillatory signal type, frequency content and recording time of the different setups are summarised in table 1.

#### Spirometry

The subjects performed repeated spirometric manoeuvres that met the American Thoracic Society standards [10] in C4 or the ERS standards [11] in C1–C3 and C5.

*Z*_{rs} data were collected prior to spirometry. The measurements of the baseline values of *Z*_{rs} and forced expiration were repeated after 15 min to assess short-term repeatability. Subsequently, 400 μg of salbutamol was administered through a spacer and the lung function measurements were repeated 15 min thereafter.

### Reference impedance

A high-resistance (8.9 hPa·s·L^{−1}) reference impedance was circulated among the centres to check the accuracy of the different FOT setups.

The different FOT devices and calibration methods, the spirometers, and the measurements of the reference impedance are described in the online supplementary material.

### Statistical methods

#### Anthropometric data

The intercentre comparisons of anthropometric and spirometric data were performed with one-way ANOVA with Fisher’s *post hoc* least significant difference analysis.

#### Impedance data

The intercentre comparisons of the *Z*_{rs} spectra were made with repeated-measures ANOVA. Because of the skewed distributions, resistance (*R*_{rs}) and reactance (*X*_{rs}) of the respiratory system were analysed after logarithmic transformations: ln(*R*_{rs}) and ln(4–*X*_{rs}), respectively. To take into account the different sets of oscillation frequencies included in the different FOT setups and to model the covariance structures of the simultaneously measured data, mixed-model ANOVA [12] was employed. Prediction equations were derived from multiple linear regression models.

From the short-term variability of the lung function data, the repeatability measure defined as the within-subject within-occasion variability was assessed in 302 subjects (82% of the total population; 150 male and 152 female). This variability was expressed as the coefficient of repeatability, defined as twice the standard deviation of the difference of the two baseline measurements made 15 min apart [2]. The coefficient of repeatability defines the limits within which 95% of the differences between two measurements will lie if the bias is zero. Coefficient of repeatability was used to estimate the number of responders to the bronchodilation administration. Since the largest changes in *Z*_{rs} occur in asthmatic patients at low frequency [1], this part of the analysis was restricted to *R*_{rs} and *X*_{rs} at low frequency in order to compare with the change in forced expiratory volume in 1 s (FEV_{1}). The 95th percentile of the bronchodilator response in our population was also calculated.

## Results

A total of 368 subjects (88% lifetime nonsmokers) were included in the study; the ex-smokers had an average smoking history of 4.4 pack-years. The ex-smokers and never-smokers, corrected for anthropometry, were not different in terms of spirometry results and *Z*_{rs} data. The anthropometric characteristics of the studied subjects are listed in table 2; the only significant intercentre differences were that the subjects from C2 were younger, taller and lighter, and the subjects from C4 were older than those from the other centres.

As expected from the anthropometric data, higher absolute spirometric values were observed in the subjects from C2, but they were not significantly different when the results were expressed relative to the predicted value (table 2). In contrast, the C4 subjects displayed significantly lower spirometric volumes than those from all other centres. The ratios of FEV_{1}/vital capacity (VC) were slightly lower for the C5 subjects than for those from the other centres.

### Quality control of the FOT equipment

The impedance of the reference device was assessed with the different FOT setups and proved to be within 5% and 5° of the expected magnitude and phase shift, respectively, at all frequencies for all devices except that of C4, where a deviation of >5% from the expected magnitude was observed at frequencies ≥10 Hz (online supplementary fig. S1).

### Centre dependence of *Z*_{rs} data

Mean values of *R*_{rs} and *X*_{rs} at the different centres after adjustment for the covariates height, age and weight are plotted in figure 1; some systematic intercentre differences are apparent in the frequency dependences of *R*_{rs} and *X*_{rs}. As described in detail in the online supplementary material, mixed-model analysis was applied to choose the autoregressive covariance structure with which to model the correlation between spectra. Mixed models with the interaction centre × frequency and the covariate combinations height–age–weight and height–age–body mass index (BMI) were addressed. Since the latter combination did not improve the model performance, height, age and weight were retained for the subsequent analysis. The dependences of the *R*_{rs} and the *X*_{rs} data on sex, frequency and centre were analysed on ln(*R*_{rs}) and ln(4–*X*_{rs}). Linear interpolation of *Z*_{rs} values between some frequency values was necessary to facilitate the combination of data from the different centres. The frequency dependence of *Z*_{rs} was significantly different in the various centres, thus the full set of *Z*_{rs} spectra from all the centres could not be combined; however, data pooling was possible for *R*_{rs} (with the exceptions of 20 and 25 Hz in the females and 25 Hz in the males) and for *X*_{rs} up to 14 Hz in both sexes. Exclusion of the data from C5 eliminated the significant differences in *R*_{rs}, but not in *X*_{rs}. The C5 data were retained for further analysis.

### Prediction equations

Values of single-frequency *R*_{rs} and *X*_{rs}, the resonant frequency (*f*_{res}), the reactance curve area (*A*_{X}) [14] and the mean *R*_{rs} (*R*_{mean}) were included in the predictions on height, age and weight. The coefficients of the prediction equations and the residual standard deviations for ln(*R*_{rs}) and ln(4–*X*_{rs}) are listed in tables 3 and 4, respectively. Although age did not prove to be a significant contributor to the *R*_{rs} values at high frequency (>12 Hz) for the females, it was not omitted from the model for reasons of consistency of the predicted *R*_{rs} as a function of frequency. However, age did not contribute significantly to the prediction of the *X*_{rs} values for the males at any frequency, and was therefore omitted from this particular prediction. Unlike the frequency dependence of *R*_{rs}, the *R*_{mean} values were not statistically different between the centres, and the *A*_{X} calculated either between 4 Hz and *f*_{res} (*A*_{X4}) or between 5 Hz and *f*_{res} (*A*_{X5}) were also centre-independent. Although there was a centre dependence in *f*_{res}, the differences between the centres were small (the largest difference was <2 Hz) and considered clinically irrelevant. Calculation examples for the predicted median values and upper/lower limits of the normal values are given in the online supplementary material.

### Baseline variability of *Z*_{rs}

The variability in *R*_{rs}, expressed in both absolute and relative terms, did not differ between the centres or between the sexes. Furthermore, no systematic differences were found between the two baseline measurements in *R*_{rs} and in *X*_{rs}. The coefficient of repeatability values of *R*_{rs} at low frequency (n = 302) and that of FEV_{1} (n = 179) are reported in table 5. Bland–Altman analysis [15] revealed that the absolute difference between the repeated baseline measurements was related to the magnitude of the baseline value of *R*_{rs}. The relative difference between the two baseline measurements, however, was independent of the magnitude of *R*_{rs} (fig. 2).

### Bronchodilator response

There was no centre dependence of the bronchodilator effect as assessed with FOT. Bronchodilation induced a significant decrease in *R*_{rs} at all frequencies (mean decrease 11%) and an increase in *X*_{rs} at 4, 5 and 6 Hz and FEV_{1} (p<0.001, mixed-model ANOVA), whereas vital capacity did not change. The changes in *R*_{rs} were more pronounced than those in *X*_{rs} (fig. 3).

The relative coefficient of repeatability was used to identify responders to bronchodilation. *R*_{rs} at 5 Hz identified 36% of the healthy subjects as positive responders to salbutamol and 3% of the subjects as contraresponders. By contrast, FEV_{1} identified 17% of the studied subjects as positive responders and 1% as contraresponders (online supplementary table S1). Only 28 (8%) subjects were positive responders according to both methods. *X*_{rs} at low frequency was markedly less sensitive for detection of the bronchodilator response compared to *R*_{rs} at low frequency (online supplementary table S1).

In order to help define a positive response in asthmatic patients, the 95th percentile of the bronchodilator response in healthy adults was established. The 95th percentiles for the absolute and relative changes in *R*_{rs} and *X*_{rs} at low frequencies due to bronchodilation are reported in table 5. For example, the 95th percentile response in *R*_{rs} at 4 Hz was 1.41 hPa·s·L^{−1}, corresponding to a relative decrease of 33% (table 5).

## Discussion

As far as we are aware, this is the first study in which baseline and post-bronchodilator *Z*_{rs} data were collected from healthy adult subjects at multiple centres. The *R*_{mean} values found in our study are close to the published data [3, 4, 6, 8]. In accordance with earlier findings, females displayed larger values of *R*_{rs} than those of males. The impedance data of our healthy adult subjects were described by variables similar to those used previously in prediction equations: sex, height, age and weight.

### Study population

The age and sex distributions of the population were well controlled; males and females contributed virtually equally, and the participants of each sex were evenly distributed between 18 and 80 years of age. All subjects included participated in spirometry. The FEV_{1} and VC values of the subject population were close to the expected values (table 2). The baseline spirometry data from the individual centres were essentially similar, with one exception: the C4 subjects furnished values which were significantly lower than those at the other centres, but still within the expected range [13].

Obesity was not an exclusion criterion in our study. The average BMI was 25.3 and 26.1 kg·m^{−2} for females and males, respectively; values close to those reported by the World Health Organization for adults [16]. *R*_{rs} is expected to rise and *X*_{rs} to fall with the level of obesity [17, 18], but BMI did not seem to be a stronger predictor than weight in our study.

### Comparison with earlier predictions

Our predicted values for *R*_{mean} are compared with previous predictions in figure 4. The current equations predict higher values of *R*_{mean} than most of the earlier predictions [4, 5, 6, 8, 19] for short and young males, and lower values for tall males, the predicted *R*_{mean} therefore exhibiting a larger height dependence, whereas lower values of *R*_{mean} for young females were found, but with a similar height dependence, compared to most previously published equations.

A comparison of the present and previous predictions for *R*_{rs} and *X*_{rs} at individual frequencies is difficult, since in most of the literature reports *Z*_{rs} data were described by polynomials of frequency [3, 6, 8, 20] or linear regression determining the frequency dependence of *R*_{rs} and its extrapolated zero-frequency value [21]. A similar comparison of *R*_{rs} as a function of weight with earlier predictions is presented in online supplementary figure S6.

The present study allows the prediction of *R*_{rs} between 4 and 26 Hz and *X*_{rs} at ≤14 Hz on the basis of data pooled from the five centres. *Z*_{rs} data, and especially those measured at low frequency, are sensitive to bronchial obstruction and constriction [1], and the newly derived prediction equations adequately cover these frequency ranges. Measured data can be considered abnormal if the values of *R*_{rs}, *A*_{X} or *f*_{res} is greater or *X*_{rs} is lower than the corresponding 95% confidence limit, *i.e.* the reference value ±1.64×residual standard deviation (tables 3 and 4, and the online supplementary material).

### Equipment dependence

Whereas all manufacturers stress the uniqueness of their devices, their users expect all FOT setups to comply with the ERS guidelines [1] and that the results obtained with the various setups should be comparable. Thus, it was an underlying assumption of the current study that the *Z*_{rs} data measured in the different centres could be pooled. To test this assumption, a reference impedance was first sent to each of the centres in order to test the accuracy of the equipment. While all setups recovered the reference value more or less accurately, it should be noted that routine calibration devices usually have much lower impedance values (∼2 hPa·s·L^{−1}), which allow only a less rigorous testing.

The possibility of a device dependence in the *Z*_{rs} data was tested next. The statistical analysis demonstrated that the frequency dependence of the *R*_{rs} data on these healthy subjects differed in the various centres (fig. 1). However, despite anthropometry adjustment, there still might have been differences between the studied populations. This possible effect cannot be distinguished from the impact of device differences. The device at C5 was found to provide outlying *R*_{rs} data at 20 and 25 Hz, and their *R*_{rs} data therefore exhibited a more marked negative frequency dependence than those from the other centres. Omission of C5 data from the regression changed the coefficients in the prediction equations of *R*_{rs} to a negligible extent (data not shown). However, it is important to point out that the interpretation of the negative frequency dependence of the *R*_{rs} data in terms of the function of the small airways (*R*_{5}–*R*_{20}) and that of the large airways (*R*_{20}) has been associated in the literature with this particular device [22, 23]. The present finding, that the frequency dependence of *R*_{rs} is device-dependent, seriously challenges this simplistic interpretation.

The *X*_{rs} data at frequencies above *f*_{res} were also centre dependent, although there was no single outlying centre or device; the centres rather differed pairwise (online supplementary fig. S1). The most probable explanation for this deviating *X*_{rs} at high frequency is a difference in the inertance of the connecting tubes/mouthpiece. These differences in construction and the correction procedures, which are not always transparent for the users, have less influence on the *X*_{rs} data at low frequencies.

A too-low common mode rejection ratio of the flowmeter may be an important determinant of a distorted measurement of the flow at high frequency [24]. Unfortunately, the common mode rejection ratio of the commercial devices was not specified and, in most cases, could not easily be checked by the users. However, all setups accurately measured the high reference resistance at all frequencies (online supplementary fig S1). Nevertheless, the frequency dependence of *R*_{rs} of healthy subjects after adjustment for anthropometry was shown to be centre, and thus device, dependent (fig. 1). This discrepancy might then result from differences in instrumentation and signal/data processing procedures, whose details are unavailable for the commercial devices. Perhaps most importantly, we implemented a calibration check procedure without pressure and flow fluctuations similar to those from spontaneous breathing, and hence the possibly different nonlinear performances of the FOT setups were not tested.

Previous studies have suggested that the frequency dependence of *R*_{rs} is a markedly sensitive index that permits differentiation between health and disease and an assessment of the disease severity [7, 25–27]. However, as noted above, our results indicate that this index is device dependent. We therefore suggest that the optimum way to describe abnormalities in *Z*_{rs} is to look for an increase in *R*_{rs} and a decrease in *X*_{rs} at low frequency only. *A*_{X} and *f*res are additional indices *via* which to characterise abnormal *X*_{rs}.

### Baseline variability and bronchodilator response

Use of the coefficients of repeatability as threshold values to define a positive (or negative) response to bronchodilation implies that, by definition, 5% of the studied subjects are expected to be marked as positive or negative responders if the intervention has no effect. The number of positive responders was much larger than 5%, indicating that the bronchomotor tone changed in a significant proportion of our healthy adults. *R*_{rs} at low frequency detected twice as many positive responders than FEV_{1} did (36% *versus* 17%). The number of contra-responders with either index was <5%, but *R*_{rs} detected more contra-responders than FEV_{1} did (online supplementary table S1). Being more sensitive than FEV_{1} for detection of a change in bronchomotor tone in the clinical setting, *R*_{rs} has been proposed for broncho-challenge testing, but not for reversibility testing [1].

Perhaps the most important result of the bronchodilation part of our study is the 95th percentile of the response in our group of healthy adult subjects, which was equal to ∼32% for *R*_{rs} at low frequency. The bronchodilator response of healthy adults is largely unknown. We are aware of only one study in which subjects screened in an allergy and asthma clinic were assigned to a healthy control group [23]. They exhibited an average decrease of 15% in *R*_{5} after bronchodilation, a value close to the decrease of 11% in the present study. The 95th percentile of the response in our group of healthy adults is slightly lower than the values previously reported for healthy preschool children: 42% [28] and 43% [29].

### Conclusions

In summary, our multicentre, multidevice study on normative respiratory impedance data for adult subjects revealed that FOT devices yield systematic differences in *R*_{rs} at 20 and 25 Hz and in *X*_{rs} at frequencies >14 Hz. This indicates that the standardisation rules of the *Z*_{rs} measurements should be complemented with more rigorous calibration procedures in order to ensure data compatibility at all frequencies. However, for the lower-frequency *Z*_{rs} measures, the present prediction equations allow the comparison and interpretation of adult impedance data relating to Caucasians worldwide, and our data also provide information on the baseline variability of *Z*_{rs} and the bronchodilator response in healthy adults. The use of our baseline *Z*_{rs} data and the bronchodilator response should be evaluated prospectively to establish their ability to aid in the diagnosis and management of adult patients with asthma or chronic obstructive pulmonary disease.

## Footnotes

For editorial comments see page 1435.

This article has supplementary material available from www.erj.ersjournals.com

Conflict of interest: None declared.

- Received August 10, 2012.
- Accepted February 6, 2013.

- ©ERS 2013