A Realistic Validation Study of a New Nitrogen Multiple-Breath Washout System

Background For reliable assessment of ventilation inhomogeneity, multiple-breath washout (MBW) systems should be realistically validated. We describe a new lung model for in vitro validation under physiological conditions and the assessment of a new nitrogen (N2)MBW system. Methods The N2MBW setup indirectly measures the N2 fraction (FN2) from main-stream carbon dioxide (CO2) and side-stream oxygen (O2) signals: FN2 = 1−FO2−FCO2−FArgon. For in vitro N2MBW, a double chamber plastic lung model was filled with water, heated to 37°C, and ventilated at various lung volumes, respiratory rates, and FCO2. In vivo N2MBW was undertaken in triplets on two occasions in 30 healthy adults. Primary N2MBW outcome was functional residual capacity (FRC). We assessed in vitro error (√[difference]2) between measured and model FRC (100–4174 mL), and error between tests of in vivo FRC, lung clearance index (LCI), and normalized phase III slope indices (Sacin and Scond). Results The model generated 145 FRCs under BTPS conditions and various breathing patterns. Mean (SD) error was 2.3 (1.7)%. In 500 to 4174 mL FRCs, 121 (98%) of FRCs were within 5%. In 100 to 400 mL FRCs, the error was better than 7%. In vivo FRC error between tests was 10.1 (8.2)%. LCI was the most reproducible ventilation inhomogeneity index. Conclusion The lung model generates lung volumes under the conditions encountered during clinical MBW testing and enables realistic validation of MBW systems. The new N2MBW system reliably measures lung volumes and delivers reproducible LCI values.


Introduction
Multiple-breath inert gas washout (MBW) tests are used to assess uniformity of ventilation distribution in infants, children, and adults [1]. Both global indices of ventilation inhomogeneity such as the lung clearance index (LCI) and specific indices of nonuniform gas mixing in the conducting and acinar airway zones (S cond and S acin ) are frequently reported [2][3][4][5][6]. The accuracy of these indices relies directly or indirectly on correctly measured functional residual capacity (FRC). Therefore current guidelines for lung function testing recommend in vitro FRC measurements over the full range of lung volumes in question at various respiratory frequencies [7,8].
FRC measurement validation is commonly assessed using calibration syringes to simulate FRC and tidal breathing washout [9]. However, in vivo accuracy of FRC measurements depends on the individual performance under physiological conditions of the gas and flow sensors used, ignored by the syringe. Given the increasing interest in MBW as a clinical outcome measure there is an urgent need for improved validation models which incorporate conditions encountered during clinical testing (i.e. variations in temperature, pressure and humidity over the breath cycle).
This report describes (i) the development and utility of a lung model incorporating simulated body temperature, pressure and water vapor saturation (BTPS) conditions, and (ii) the in vitro and in vivo performance of a commercial N 2 MBW system (Exhalyzer DH, Eco Medics AG, Duernten, Switzerland). Primary outcome was the accuracy and precision with which FRC can be generated and measured in vitro and the reproducibility of FRC in vivo in healthy adults. Secondary outcomes were the reproducibility of ventilation inhomogeneity indices and their correlation in healthy adults.

Nitrogen multiple-breath washout setup
We used an unmodified open-circuit N 2 MBW hardware and software package (Exhalyzer DH and SpirowareH 3.1, Eco Medics AG) for all recordings, and an in-house customized software based on TestPoint TM (Capital Equipment Corp, Billerica, MA, USA) for off-line data processing and analyses. Flow was measured using a mainstream ultrasonic flowmeter and used to derive tidal volumes [10]. Gas concentrations were measured by a side-stream laser O 2 sensor (Oxigraf, Inc, Mountain View, CA, USA) and a main-stream infra-red CO 2 sensor (CapnostatH 5, Respironics Novametrix LLC, Wallingford, CT, USA). In this device, F N2 is measured indirectly based on Dalton's law of partial pressures: F N2 = 12F O2 2F CO2 2F Argon (F Argon = F N2 * 0.00934/0.78084). The F Argon (0.00934) is treated as a fixed proportion of the F N2 assuming similar washout during N 2 MBW. Daily two-point calibration and verification of the flow and O 2 sensors, and zero calibration of the CO 2 sensor were performed. The O 2 sensor has a slower 10-90% response time (140 ms) than the CO 2 sensor (55 ms). To align their signals, a speeding algorithm was applied to the O 2 signal reducing its response to approximately 110 ms [11,12]. Gas signals were synchronized to the flow signal using the re-inspired post-capillary dead space to produce a step response in CO 2 and O 2 . The gas signal vectors were time shifted to the point in time when the post-capillary dead space had been inhaled such that a 50% change in gas signal deflection then occurred. This was repeated over a minimum of ten washout breaths and median ''delay times'' for CO 2 (50 ms) and O 2 (565 ms) were used for signal alignment. Quality of superimposition of the inverted O 2 signal on the CO 2 signal was assessed visually.
Lung (model) resident F N2 was washed out using 100% O 2 applied via open circuit at either 200 mL/s for 100-400 mL FRCs or at 1000 mL/s for 500-4200 mL FRCs and in vivo, respectively ( Figure 1). These bypass flows were chosen to exceed maximum tidal inspiratory flows and to minimize rebreathing of CO 2 or N 2 . For respective lung volumes (Table 1) we used post-capillary dead space reducers (infant set 1, preschool set 2, adult set 3) and hygienic inserts (Spirette) provided by the manufacturer (Eco Medics AG). These reduced equipment related post-capillary dead space (volume between CO 2 /O 2 sampling point and bypass) to 1.5 mL, 16 mL, and 26.9 mL, respectively, as measured by water displacement. Bacteria filters (air eco slimline, Vickers Ind Est, LA, UK) had 30 mL dead space and were used for preschool and adult sets. Accordingly, pre-capillary dead space (volume between lung compartment top and CO 2 /O 2 sampling point) was 37.9 mL in the large model and 3 mL in the small model, respectively. Apparatus resistance was measured by a pressure transducer (Timeter RT200, Allied Healthcare Products Inc., MO, USA). We applied pure O 2 and increased flows stepwise between 0-200 mL/ s for the infant set, 0-500 mL/s for the preschool set, and 0-770 mL/s for the adult set with bacteria filters in place. Maximum resistances were 0.03, 0.06, and 0.19 kPa/L*s, respectively, and complied with the recommendations of previous standards [8,13].

The lung model
The framework of the lung model was constructed from acrylic glass (Soloplex, Tidaholm, Sweden) and consisted of two rigid chambers: An inner chamber divided into two communicating compartments (via their lower aspect) termed the lung and the ventilation compartment, and an outer chamber ( Figure 1). Two different size models were constructed to allow for both infant (100 to 400 mL) and children/adult (500 to 4200 mL) lung model volumes. The inner chamber was filled with distilled water until the desired FRC was achieved, measured as the end-expiratory water level using a transparent vertical tape measure fixed to the lung compartment. FRC volume was determined geometrically from known dimensions: one millimeter corresponded to 18.7 mL in the large and to 4.8 mL in the small lung model. The lung model was placed inside a water tank which served to heat the water in the inner chamber to 37uC, monitored using a thermostat. To simulate various breathing patterns, a bi-level positive airway pressure ventilator (Vivo 30, Breas Medical AB, Mölnlycke, Sweden) for the large lung model, or a 100 mL calibration syringe (Hans Rudolph Inc, Shawnee, KS, USA) for the small lung model was connected to the top of the ventilation compartment and exerted hydraulic pressure transmitted through to the lung compartment. During ventilation, temperature and humidity were measured every third FRC trial inside the lung

In vitro study
Static signal linearity of the N 2 MBW system was assessed over the full range of O 2 , CO 2 , and N 2 fractions encountered during N 2 MBW testing and performed at 37uC with 100% relative humidity: By increasing F O2 stepwise (60 steps), F O2 ranged from 15-100%, F CO2 from 0-6%, and F N2 from 0-79%. The reference for measured gas concentrations was a respiratory mass spectrometer (AMIS 2000; Innovision A/S, Odense, Denmark).
In vitro assessments of the N 2 MBW system performance were undertaken as triplicate FRC measurements (n = 150) across 50 different nominal FRCs (100 to 4200 mL), over two days. Ventilator pressures in the large model or syringe stroke volumes in the small volume were chosen to achieve physiological tidal volumes (V T ), V T over FRC ratios (V T /FRC), and respiratory rates (RR) for each FRC setting (Table 1). To assess the possible influence of different F CO2 on FRC measurement accuracy, the lung compartment was washed-in with either ambient air (0.05% CO 2 ) or CO 2 enriched air (n = 45, CarboairH, Aiolos Medical, Karlstad, Sweden) prior to N 2 MBW. CarboairH contains 5% CO 2 and balance air.

In vivo study
Thirty-two healthy adults performed N 2 MBW on two test occasions within a three week period. All subjects had a standardized interview on respiratory health. Inclusion criteria were adults aged 19 to 70 years with a smoking history less than five pack-years, no history of acute or chronic airway disease, and no on-going medication potentially affecting lung function. N 2 MBW was performed in the sitting position using a nose clip with the dead space reducer (set 3), hygienic insert, and bacterial filter in place. N 2 MBW was done in triplets with between-test intervals exceeding the washout time. The subjects were instructed to breathe regularly with relaxed expirations. The washout phase was terminated once end-tidal F N2 was less than 1/40 th of the starting F N2 for at least three breaths.

Ethics statement
The in vivo study was approved by the Ethics Committee of the University of Gothenburg Sweden. We obtained informed consent from all participants involved in the study.
Nitrogen multiple-breath washout outcomes FRC was calculated as net expired N 2 volume (expired N 2 volume minus re-inspired N 2 volume) divided by the difference of F N2 at start of MBW minus F N2 at end of MBW. Pre-capillary dead space was subtracted from FRC such that the reported FRC corresponds to the volume of the inner compartment of the lung model or FRC at the airway opening in the in vivo recordings. Indices of ventilation inhomogeneity were calculated from in vivo N 2 MBW trials. Because LCI, S acin , and S cond relate inversely to ventilation efficiency, their values increase with increasing ventilation inhomogeneity. LCI was calculated as cumulative expired gas volume (CEV) required to reduce F N2 to 1/40 th of the starting F N2 , divided by FRC. Two phase III slope (S III ) indices, S cond and S acin , were calculated. Automated S III fitting over 50-95% of expired volume was performed for each breath and manually adjusted to exclude phase II or IV from the linear regression fit. S III was normalized (Sn III ) by dividing S III by the mean F N2 over S III and multiplying S III with V T , and averaged per breath from the three N 2 MBWs [14]. Mean Sn III per breath was then plotted against the corresponding mean lung volume turnover (TO = CEV/FRC) for each breath. S cond is defined as the rate of Sn III increase between lung volume turnovers 1.5 and 6.0. S acin is defined as the first breath Sn III value minus the convection-dependent inhomogeneity contribution to this value [15].

Statistics
Linearity of sensors compared to mass spectrometry was assessed using uni-variable linear regression and respective signal offsets (linear model intercept) and gains (linear model slope) with their 95% confidence intervals (CI) were reported. Intra-test variability of both model and measured FRCs as generated by the model and measured by the N 2 MBW setup, respectively, was calculated as intra-test coefficient of variation (CV = SD/ mean*100). Accuracy of FRC measurements was expressed as (i) absolute (mL) difference (difference between measured FRC minus model FRC), (ii) relative (%) difference (difference*100 divided by model FRC), and as (iii) relative error (square-root of the squared relative difference). Acceptable upper limit was 5% error according to current infant lung function standards [7]. We reported coefficient of repeatability (CR) calculated as 1.96*SD of differences between paired measurements [16]. The CR estimates the 95% range of technical variability due to measurement error in vitro or the technical and physiological between-test variability in vivo. We assessed Bland Altman plots [16], and the relation of error with FRC, breathing pattern, different lung model gases, and between tests using uni-and multi-variable linear regression models and paired t-tests. P-values,0.05 were considered statistically significant and all analyses were done using Stata TM (Stata Statistical Software: Release 11. College Station, TX: StataCorp LP).

Lung model performance
The lung model successfully generated 146 (97%) out of 150 FRCs (range 100-4174 mL) under BTPS conditions. Mean (SD) of temperature and humidity were 35.9 (1.3)uC and 97.9 (1.0)%, respectively, obtained from 18 measurements. Following water refilling, four gas temperature drops of 3uC were observed and respective FRC recordings were subsequently excluded. Realistic breathing patterns were applied with V T between 30 and 877 mL, V T /FRC between 0.14 and 0.55, and RR between 10 and 35 min 21 . The intra-test variability of FRCs generated by the lung models was low overall, CV mean (SD) was 0.07 (0.31)%, and lower in the large model compared to the small model: CV mean (SD) was 0.03 (0.15)% and 0.23 (0.64)%, p = 0.108, respectively. The potential parallax error of reading water levels using the tape measure was estimated as one mm corresponding to 18.7 mL and 4.8 mL in the large and small lung model, respectively. Relating these volumes to the nominal FRC, the relative mean (range) parallax error was 1.2 (0.4-3.7)% and 2.9 (1.2-4.8)% in the large lung and small model, respectively.
Errors in FRC measurements in lung models containing either ambient or CO 2 enriched air were similar. Comparing the mean (SD) error using ambient air (n = 102) vs.  (Table 3). In vivo reproducibility for LCI was better than for FRC, and markedly better than for S acin and S cond (Figures 4, 5). LCI was correlated with S acin (Pearson r = 0.61, p,0.001) but not with S cond (r = 0.17, p = 0.377). S acin was correlated with S cond (r = 0.44,

Summary
This is the first study demonstrating that an available N 2 MBW system accurately and reliably measures FRC over the wide range of lung volumes and different breathing patterns encountered from early childhood to adulthood. This study also shows that BTPS conditions can be implemented into lung model studies to benchtest MBW equipment under realistic conditions. In vitro FRC can be measured with only 5% error over FRC volumes between 600 and 4200 mL with the N 2 MBW system. In FRCs below 600 mL, the error is better than 7%. N 2 MBW measurements in healthy adults show that LCI is more reproducible than FRC, S acin , and S cond . Variability of FRC measured in vivo is higher than in vitro suggesting substantial physiological fluctuation of FRC in humans [18].

Comparison to previous lung model studies
Data on quality of signals and their integration under realistic conditions are essential for appropriate evaluation of MBW setups [19]. As an initial step, linearity of O 2 , CO 2 , and N 2 signals was confirmed over the full measurement range in heated humid conditions encountered during clinical N 2 MBW testing. We improved a lung model described by Brunner et al. [20], which was originally ventilated manually and the water contained within the model was not heated. In addition to the incorporation of BTPS conditions, physiological breathing patterns were applied. These conditions recreate the stress the system is exposed to during clinical testing across wide age ranges. Previous in vitro lung models lacked one or more important physiological aspects [21][22][23][24][25].
The current N 2 MBW system precisely measures lung volumes between 600 and 4200 mL. In this volume range approximating late preschool age to large sized adults, the in vitro measurement error was below 5% in all FRCs (Figure 3). Importantly, this N 2 MBW system is not susceptible to different F CO2 . Indirect N 2 measurement may be susceptible to drift in F CO2 as the current N 2 signal or, e.g. molar mass, are sum signals of gas fractions including the F CO2 [26].
This study is able to disentangle technical from physiological aspects contributing to FRC variability. Comparing the in vitro CR of FRC (4.6%) with in vivo (24.9%), inherent physiological variability of FRC may explain 20.3% (or relatively 82%) of between-test variability in adults. Validation studies exclusively performed in vivo do not allow differentiation between technical and physiological variability [24,27]. In infants, the FRC error between tests may exceed 20% [28,29].

Ventilation inhomogeneity indices
The current N 2 MBW setup can be easily applied to measure ventilation inhomogeneity indices in vivo. These measures correlate as shown previously [15,30]. LCI, in particular, had high repeatability and reproducibility in healthy adults. The specific indices of non-uniform gas mixing in the conducting and acinar airway zones (S cond and S acin ) were more variable. Between-test reproducibility of LCI (CR = 9.5%) was markedly higher than for S acin (49.2%) and S cond (41.4%). In a recent study [31], comparable between-test reproducibility of LCI, S acin , and S cond was reported in healthy adults performing a sulfur hexafluoride MBW protocol with tidal breathing restricted to one liter V T . The small technical error of FRC observed in vitro strongly suggests higher physiological variability of these Sn III indices compared to volume indices (FRC, LCI). Sn III depends on breathing pattern  Table 3. Nitrogen multiple-breath washout outcomes and variability.
In vitro nitrogen multiple-breath washout outcomes  [32], which was not restricted in this study. Interestingly, LCI is even more reproducible than FRC as also shown in healthy children [33]. This may reflect the fact that LCI is a robust volume ratio (CEV/FRC), cancelling the effect of physiological changes in lung volumes.

Limitations
While the large lung model was suited for the MBW bench test, the small model may require adjustments. Automated ventilation and a laser sensor for determining nominal FRC could improve the infant lung model. Parallax error in the large model was small enough to avoid over-or underestimation of nominal FRCs (Figure 3).
Within the small technical measurement error of FRC, RR and tidal flow were correlated with error. Whether this association was causal leading to signal asynchrony or rather a surrogate for flowdependent temperature and humidity fluctuations hampering adequate BTPS correction remains to be determined [10]. We hypothesize that the apparatus dead space volumes behave as systems taking up and delivering heat energy and moisture nonlinearly over breath cycles. Reduced dead space and dynamic BTPS and synchronization algorithms may further decrease measurement error.
The current and previous models did not allow assessment of ''tissue'' N 2 contribution to the lung inherent F N2 . This contribution is difficult to estimate and account for, as available data are limited [34]. In adults with more advanced lung disease, duration of N 2 MBW and bias from tissue N 2 may increase, and intervals between tests should be adapted accordingly [8].

Implications
Availability of realistic validation protocols will aid the transition of MBW from promising research tool into routine clinical use. This is the first validation of an available N 2 MBW setup using  physiological in vitro test conditions. Many other MBW setups remain to be validated and are either customized, expensive, or have been taken off the market [2,3,9,35]. N 2 MBW is also advantageous to using foreign tracer gases because 100% O 2 is economic, readily available in all hospitals, and non-polluting.
The between-test variability of ventilation inhomogeneity indices reported here sheds further light on their natural variability over time. The CR aids estimating clinically relevant change in lung function outcomes for future intervention studies. Given a hypothetical intervention in the current study, a decrease of LCI of 0.68 units (9.5%) would constitute a physiological change with 95% probability. Further longitudinal studies in disease groups, such as cystic fibrosis where variability is increased, are required [36][37][38]. In infants and preschool-aged children, safety of N 2 MBW and variability of N 2 MBW indices need to be determined.

Conclusion
The performance of MBW setups can be realistically assessed using a new in vitro lung model producing BTPS conditions and representative breathing patterns. This lung model protocol is suitable for validation of other MBW systems. The new N 2 MBW system accurately and precisely measures lung volumes between 600 and 4200 mL. Future work may extend this into smaller lung volumes and younger age groups. The LCI represents the most reproducible N 2 MBW in vivo outcome as compared to FRC, S cond , and S acin . This study may aid the transition of this important type of lung function test from being a promising research tool to becoming a routine method in clinical practice.