Respiratory inductance plethysmography (RIP) is purported to be a valid method for measurements of ventilation derived from chest wall motion (12,29). The physiological validity of this approach has been justified by the findings of Konno and Mead (17). The close relationship between changes in rib cage (RC) and abdomen (AB) and changes in the corresponding anterior–posterior diameter illustrates the authors’ essential postulate. Since the 1980s, RIP became an applicable monitoring device to observe ventilation, primarily in clinical settings (12,29). It has been stated that once valid calibration is applied, RIP is able to detect 89% of all breaths accurately within a ±10% error margin under resting conditions and various postures (29). The advantage of RIP in contrast to methods requiring mouthpieces, nose clips, or facemask is obvious. The influence on natural breathing patterns caused by diagnostic apparatus itself is well known and documented (2,5,10,21,28). A less restrictive device might be useful for performance assessment and monitoring training in the field.
Although the application of various methods for RIP calibration (3,8,10,17,23,25) has never been properly validated under exercise conditions, some studies suggest an apparently acceptable accuracy during treadmill running and bicycling (6,7,9,14,16,22,30). Applied gain factors for RIP calibration in these studies have been determined before the actual experiment (a priori). However, the majority (6,9,14,16,30) did not consider changes in thoracoabdominal contribution during the relevant exercise. Hence, applied gain factors would have been invalid as the calibration was based on a subset of quasi-constant breaths (23). Other investigators (7,22) applied least square regression (LSQ) for calibration (8) on an initial subset of variable breaths. Nevertheless, the chosen subsets were still too homogeneous so that in turn, calculated gain factors have been a result of statistical overfitting and thus invalid for any application. The validity of determined gain factors itself has also never been examined.
To validate calibration procedures of RIP for subsequent measurements, it is necessary to know how accurate RIP possibly can be. From an algebraic perspective, calibration via LSQ provides the most reliable gain factors as long as high interbreath variability in terms of tidal volume and partial contribution (24,26) is ensured. These kinds of data are only available after the actual observation (a posteriori). The main purpose of this study was to provide a rationale for future validations of a priori calibration methods of RIP. Therefore, the validity of a posteriori–adjusted gain factors and accuracy in resultant breath-by-breath RIP data recorded under conditions of standing still, treadmill running and multiple postexercise recovery periods has been examined in comparison with a flowmeter (FM).
Before the experiment, minimum sample sizes of 185 and 126 subjects for resting and peak tidal volumes, respectively (alpha = 0.05, power = 0.9), were calculated via power analysis for equivalence statistics (15). The upper and the lower limits of equivalence for resting and peak tidal volumes have been set on ±10% according to previous findings (8). In total, 98 male and 88 female healthy subjects were finally recruited from public homepage-based advertisement and intern mailing lists. The acquisition and data collection occurred within an 8-month period (April–November). All subjects completed questionnaires of their own health status. Only subjects without any indication of cardiovascular or lung diseases as well as therapeutic medication were included. In addition, anthropometrical data have been collected to ensure heterogeneity in terms of body height, body weight, sex, and age. All subjects stated written informed consent with the experimental protocol, which was approved by the ethics commission of Albert Ludwigs University of Freiburg.
All subjects were experienced in treadmill running and familiarized with the experimental protocol. An initial phase of standing still for 5 min was followed by a standardized incremental running test on treadmill (Quasar, H/P/Cosmos Sports and Medical GmbH, Nussdorf-Traunstein, Germany) in upright posture. The slope of the treadmill was set on 1% incline for all subjects. Starting speed was 6 km·h−1. Every 3 min, the speed was increased in 2 km·h−1 steps until voluntary exhaustion. After each step, the test was interrupted for 30 s to collect blood lactate samples. This has been performed to collect useful data for additional hypotheses, which are not further discussed here. After the incremental test, subjects were instructed to stand still for an additional 10 min of recovery. All subjects were instructed to avoid physical exercise the day before testing. Neither diet nor time of day was controlled in this study.
Ventilatory timing and volume were measured by an FM fixed on the subject’s face mask, which was connected to a stationary system (Oxycon Pro Care Fusion, San Diego, CA) and calibrated automatically using a two-point flow calibration. Accuracies of ±2% and ±3% for volumes of 0–10 L and flow rates of 0–15 L·s−1, respectively, were documented by the manufacturer. Before the incremental test, individual inspiratory vital capacity (IVC) was measured by the FM according to ATS recommendations (1).
Respiratory Inductance Plethysmography (RIP)
The integrated RIP device of the LifeShirt™ garment (VivoMetrics, Ventura, CA) is formed by two parallel elastic bands with embedded insulated sinusoidal wires surrounding the RC and AB. The RC band is placed between the second and the forth intercoastal area and the AB band, respectively, at the umbilical level (Fig. 1). A low-voltage alternating electrical current is passed through the insulated sinusoidal wires. An expansion in the relevant body chamber leads to a change in the magnetic field around the wires, which in turn alters the inductance of the embedded wire leading to proportional voltage changes (28). On the basis of a 50-Hz sampling frequency, time series data from both bands were recorded on an incorporated SD card.
The VivoSense™ software (Vivonoetics, San Diego, CA) was used to decrypt and process the recorded data. Several steps of automated data processing in the VivoSense™ software occurred initially (see Figure, Supplemental Digital Content 1, http://links.lww.com/MSS/A361, which illustrates the modus operandi of data collection and analysis). A Butterworth low-pass digital filter (cutoff frequency of 1.4 Hz) was applied by the software to remove signal values resulting from nonrespirational causes. Noncalibrated units of AB- and RC-raw signals were multiplied by initial (not final) calculated gain factors so that further derived tidal volumes of the AB + RC sum signal equal, on average, 400 mL over all breaths in the individual session. This step was performed automatically by the software to initiate further breath detection. Finally, we normalized the initial gain factors of AB and RC band to “1” in all subjects. Breath detection was performed by setting the onset of inspiration and expiration on a minimum flow rate of 50 mL·s−1 to further derive ventilatory timing. The quantities in signal amplitudes within the onsets represent the corresponding inspiratory and expiratory tidal volumes, which have been calculated after final RIP gain adjustment.
FM–RIP Data Synchronization
Accurate breath-by-breath method comparison requires synchronous data streams of RIP and FM. A VBA macroroutine in Microsoft Excel (Microsoft Office 2010 Microsoft Corporation, 2003) was used to systematically synchronize both data streams in all subjects. When executing the routine, the parameter breath cycle time (tTOT) in both data streams was compared within a window of 15 breaths. The window moved breath-by-breath through the complete data set and stopped automatically when differences between methods exceeded limits of ±10% (in relation to the corresponding tTOT value). Occasionally, RIP provided incomparable values caused by dividing one valid breath in two or combining two valid breaths in one. In these cases, the complete respective subsequent RIP data set was shifted manually by one breath up or down to get as many comparable and consecutive breaths as possible. However, the relative portion of incomparable breaths was in average less than 5% in each individual. In some subjects, data synchronization was less successful within resting periods at the beginning and end of each test when low ventilatory flow was apparent and also during phases when breathing rate was nearly constant due to coupling to step rate. In such cases, the initial window was manually widened up to maximal 50 breaths to maximize the correlation in the overall trend. As it was intended to get accurate gain factors, residual differences in tTOT beyond three SD as well as incomparable breaths were excluded for accurate gain calculation but not for the actual method comparison. Final gain adjustment of RIP and method comparison as described in the next section was performed by using the Matlab package (MathWorks Inc., Natick, MA).
RIP Gain Calculation and Gain Adjustment
According to the two-degree-of-freedom model (17), any given tidal volume (VT) measured by FM can mathematically be estimated as
represent the noncalibrated values of partial VT in both bands and gainAB/RC are the corresponding gain factors. LSQ can be applied on the two-degrees-of-freedom model and calculates the most accurate gain factors for a known subset of noncalibrated RIP data (8,29). Any valid calibration of RIP consists of an initial qualitative calibration in terms of proportional gain weighting within both bands as well as a quantitative gain scaling (23). Proportional gain weighting within the AB and the RC band is mathematically expressed as k-ratio (gainAB/gainRC) and is important to ensure valid measurement of thoracoabdominal contribution (23). The gain scaling further values the raw signals in absolute quantities. Statistical computation of LSQ considers both proportional gain weighting and quantitative gain scaling. Thoracoabdominal contribution was calculated by VTIN-RC/VTIN.
We used normalized (for details, see previous section) breath-by-breath RIP data to calculate the unknown gainRC and gainAB via LSQ by solving equation  in all comparable breaths, respectively, for the complete individual data set as well as for separate data subsets. As it was suspected that during expiration the relaxation of the garment might be delayed and further affect measured expiratory tidal volume (VTEX) of RIP, inspiratory tidal volume (VTIN) was chosen as dependent criterion variable in LSQ. Three data subsets referring to the corresponding conditions (standing still, incremental test, and recovery) have been categorized. Additional subsets of exercise (including breaths only recorded under running conditions) and nonexercise data (including breaths recorded under conditions of standing still and recovery) as well as low flow (<23% peak V˙I) and high flow data (>38% peak V˙I) during nonexercising were defined. Flow percentages have been set to get 150 breaths, respectively, for each subset and subject. Finally, gain adjustment of the complete data set and data subsets was performed by applying calculated gains directly on the same corresponding data set or subset which was used for the actual gain calculation. This has been performed to examine the validity of calculated gain factors and to compare accuracy in further calculated tidal volumes when statistical overfitting is regulated.
To specify the pool of subjects, anthropometrical data in terms of age, body weight and height, IVC, abdominal and RC circumference, resting and peak values for breathing rate (fR), VTIN, inspiratory ventilation (V˙I) were described by maximum, minimum, means, and SD. As a rationale for valid RIP calibration as described by others (17) could not be provided, we examined systematic and non systematic differences of separately calculated gainRC and gainAB between four different data subsets across individuals, respectively, using Wilcoxon signed ranked test and equivalence statistics (15). Individual calculated gains of the exercise data subset were compared pairwise with the gain factors of nonexercise data to investigate the influence of body movement. Similarly, calculated gains based solely on values of low flow and high flow data during nonexercising were compared with evaluated dependencies on flow rate. Levels of significance were set on (P < 0.05). When nonsystematic differences between calculated gains of data subsets have been examined, limits of equivalence (LoE) were set on ±10% to minimize possible miscalculation in further derived tidal volumes less than ±10% (8). Percent error levels within and outside LoE were presented as relative differences of low flow–high flow in relation to low flow data and nonexercise–exercise in relation to nonexercise data.
After RIP gain adjustment by using gain factors of the complete data set, coefficients of determination (R2), bias, and SEE between methods were calculated based on breath-by-breath data in derived parameters as fR, tTOT, partial inspiratory (tIN) and expiratory (tEX) time, VTIN, VTEX, V˙I, and expiratory ventilation (V˙E) for all conditions of standing still, treadmill running, and recovery. Common techniques to investigate agreement between independent methods as described by Bland and Altman (4) could not be applied because VTIN has been used as a dependent criterion variable for gain calculation and further gain adjustment (20). Because of the high amount of compared breaths, even a single plot of differences across all data would hide any trends of individual error distribution and possible under- or overestimations. Therefore, to prove the reliability of RIP in measuring fR and VTIN, statistical hypothesis testing for equivalence (15) was performed (confirming null hypothesis). After gain adjustment of the complete data set and defined data subsets, the relative amount (in percent) in accurate estimated values of VTIN within predefined LoE has been examined and compared. According to previous findings (8), LoE for VTIN and fR were initially set on ±10% to estimate not less than 93% of all breaths within the predefined margin. Percent error levels within and outside LoE are presented as relative differences of FM-RIP in relation to FM data.
Subjects and incremental test data.
Within ±1.96 SD body height, body weight, age, IVC, RC, and abdominal circumference ranges of 35.8 cm, 43.6 kg, 32.7 yr, 5.02 L, 27.3 cm, and 28.4 cm, respectively, were observed (Table 1). Male subjects differed significantly (P < 0.0001) in all parameters when compared with females. Mean VTIN during quiet breathing under resting conditions was 766 ± 226 mL and increased up to 2380 ± 582 mL at individual peak speed during the incremental test. Changes in thoracoabdominal contribution VTIN-RC/VTIN (%RCIN) changed significantly (P < 0.0001) from 0.55 ± 0.13 to 0.66 ± 0.11, respectively, between low ventilatory flow at resting conditions and peak flow during running in all subjects (Table 2).
Validity in calculated gain factors.
Wilcoxon signed ranked test showed no significant systematic differences in calculated gainRC and gainAB when nonexercise versus exercise data are compared. In contrast, gainAB determined from high flow data was significantly higher (16.5%, P < 0.0001) than the corresponding gainAB from low flow data. Thus, k-ratio was also significantly higher (29.0%, P = 0.007) in high flow data. However, all gains and k-ratios showed no equivalence between the compared conditions (Fig. 2). Therefore, in further method comparison, individual RIP gain adjustment was also performed for the complete data set to get the most reliable gainRC and gainAB for all breaths across all conditions.
Accuracy in measuring ventilatory timing.
Between 403 and 1473 single breaths (mean ± SD = 944 ± 200) per subject measured by RIP were compared with FM. After synchronization tTOT and fR, the mean R2 was 0.96 ± 0.04 in all subjects. Respectively, under conditions of standing still, treadmill running, and recovery, 82%, 96%, and 92% of all breaths were accurate detected within ±10% LoE. Within ±20% LoE, the amount increased slightly up to 86%, 98%, and 94%, respectively, for the described conditions. RIP was more inaccurate in tEX and tIN (R2 = 0.89 ± 0.09, and R2 = 0.80 ± 0.14).
Accuracy in measuring ventilatory volume.
In regard to VTIN, R2 between methods ranged within 0.76 and 0.98 (mean ± SD = 0.91 ± 0.05) after applying the calculated gain factors of the complete data set (Table 3). Respectively, in the compared conditions, on average, 54%, 68%, and 59% of all VTIN values were estimated within ±10% LoE. Within ±20% LoE, the amount increased up to 78%, 97%, and 88% in contrast to 88%, 97%, and 91% when either gain factors of the complete data set or separate data subsets were used. Differences in VTIN between methods were inversely correlated to VTIN measured by FM (r = −0.11 to −0.67, P < 0.0001, power = 0.95). When individually plotting the differences of VTIN between methods against VTIN measured by FM, the slope in the line of regression was slightly but significantly correlated to individual AB (r = 0.32 P < 0.0001) and RC (r = 0.29 P < 0.0001) circumference. This tendency for underestimation can visually be seen in the left shift of the error distribution in Figure 3(A).
The main purpose of this study was to examine the validity of a posteriori–adjusted gain factors and accuracy in resultant breath-by-breath RIP data recorded under exercise conditions to provide a rationale (“best possible fit”) for future validations of a priori calibration procedures. Ventilatory timing and volume were compared with FM for a total of 172,104 single breaths in a heterogenic sample of 186 subjects.
Validity in calculated gain factors.
As no systematic differences between calculated gains of nonexercise and exercise data were observed, further derived tidal volumes seem to be not affected by body movement during running. It has already been shown that calculated gains are posture dependent and can differ up to ±40% when changing from supine to upright posture or vice versa. (31). Especially during physical exercise, apparent differences between calculated gains in data subsets of low and of high ventilatory flow (Fig. 2) are also due to changes in the compartmental contribution. Systematic but also nonsystematic differences of more than ±100% between the compared gains illustrate the risk (overfitting) of statistical optimization. Caretti et al. (7) observed mean differences in calculated gains of 33.9% ± 28.2% and 47.6% ± 34.5%, respectively, for gainAB and gainRC in homogeneous data subsets at resting conditions before and after exercise. Although LSQ provides the best way to fit statistical identity, the likelihood for inclusion of a function of model error may lead to invalid calibration, as already pointed out by others (11). Valid determination of gains via LSQ requires high interbreath variability in terms of tidal volume and partial contribution (24,26) to minimize statistical overfitting. Our gain calculation (of the complete data set) was based on an average of 906 ± 201 breaths per subject characterized by high interbreath variability. Because, individually, RIP in our data tended to underestimate VTIN, statistical overfitting might be reduced. Further work is required to substantiate confidence in respect of reproducibility in individual determined gains.
Linear models are consensually used for RIP calibration (8,17,23,25). An advantage of nonlinear models that apply exponential gain factors has shown negligible improvements (27). Models including an additional measurement of axial displacements might be able to increase the accuracy slightly, especially when changes in body posture occur (19). In our data, RIP tended to underestimate higher tidal volumes slightly especially in subjects, indicating a larger RC or abdominal circumference. A larger circumference in both compartments encircles larger cross-sectional areas so that axial displacement in either cranial or caudal direction may be accompanied by an enlarged volume displacement not measured by the current RIP configuration. Therefore, we would support the inclusion of additional degrees of freedom or at least the application of size-dependent correction models considering the thoracoabdominal dimension. To improve the practicability of RIP for ambulant measurements under exercise conditions, valid and easy performable procedures for a priori calibration considering thoracoabdominal changes need to be developed.
Accuracy in measuring ventilatory timing.
Under running conditions, RIP was consistently accurate when measuring tTOT and fR in all subjects. This is in agreement with previous observations (9,16,30). The more inaccurate measurements in tIN and tEX are basically due to the device settings in breath detection. Although FM sets the onset of inspiration and expiration depending on passages of zero ventilatory flow, the breath onset in RIP occurred when passing a minimum flow threshold of 50 mL·s−1. This could also be the reason for the lower amount of accurate detected breaths during periods of low ventilatory flow at the beginning and the end of each test trial. Different configurations in breath detection affect the calculation of VTIN and fR in both methods and in turn their comparison. A slight delay in the onset of inspiration or expiration inevitably leads to higher differences in tIN or tEX when RIP and FM are compared. However, in general, the sum of tIN and tEX should cancel the error so that tTOT and fR are calculated nearly accurate.
Accuracy in measuring ventilatory volume.
The initial predefined LoE were applied based on the previous findings of Chadha et al. (8). These authors documented that 93% of varying tidal volumes in standing and supine position were within ±10% accuracy limits after gain adjustment via LSQ. Our data exceed these limits. This might be due to statistical overfitting caused by low interbreath variability in the data subset, which has been chosen for gain adjustment by Chadha et al. (8). Similar to our observed conditions, Sackner et al. (22) reported amounts of 75.3%, 61.6%, and 51.3% accurate estimated breaths within ±10% limits. On the basis of breath-by-breath data, Brüllmann et al. (6) provided relative errors of ±20% within ±2 SD limits of agreement under conditions of walking. Slightly lower limits of agreement (±17%) were found during treadmill running (9). However, the results of these studies (6,9,22) are not comparable as the validity of applied gain factors is questionable. Other observations show lower mean error or better limits of agreement (13,16,18,30), which is rather due to the comparison of pooled mean values.
Figure 3B illustrates the trend and the margin of accuracy in RIP within ±5% and ±20% LoE under exercise conditions. However, even after the best possible (statistical) gain adjustment, on average, more than a quarter of all derived tidal volumes exceeded ±10% LoE. Therefore, we additionally analyzed low-pass-filtered means of RIP raw signals in terms of quantitative shifts, which can occur when the location of the RIP band changes. Although relative differences between individual means across all data and means of the condition-dependent data subsets in AB signal varied less than 5% in most subjects (2 SD), RC signal, respectively, varied up to 11%. This either might be due to unintended slippage or intended by the subjects’ adjustment for their own comfort. Any band slippage or adjustment will lead to changes in the RIP configuration. If, in a given RIP configuration, the encircled cross-sectional area changes, volume–motion proportion changes as well. This, in turn, leads to changes in the actual valid gain factors and may explain in general the limited accuracy of RIP within ±10% LoE under all conditions. Validations of a priori calibration procedures as well as any purposes for hardware reengineering should consider these limitations.
It is concluded that RIP in the case of the best possible gain adjustment provides valid measured tidal volumes confidentially within ±20% LoE compared with FM under exercise conditions, irrespective of analytical bias in the criterion method. In respect of relative changes, RIP shows higher confidence (R2 > 0.76), indicating the actual practical relevance. For example, metabolic performance assessment in field without using any restrictive or invasive devices might be possible for the very first time. Suppose ventilation recorded during a standardized ramp test is plotted against time, at least the point of respiratory compensation should be detectable. In addition to heart rate, the use of ventilation may support individual training guidance under aerobic conditions as it is less influenced by the cardiovascular drift. The appearance of inadequate ventilation after repeated bouts of high exercise intensities as it occurs in team sports should indicate the systemic fatigue, either caused by bicarbonate buffering or metabolic acidosis. This might be helpful for coaches and athletes when monitoring individual performance during training and competing. Future observations should be focused on the validity of the repeated use of uniquely determined gains and the development of easily performable a priori calibration procedures for application in the field.
This study was supported in part by Adidas Inc. (Portland, US); the Bavarian Ministry for Economic Affairs, Infrastructure, Transport and Technology; and the European fund for regional development. The authors state that the sponsors did not have an influence on the decision to submit the manuscript or on the content of the article.
No conflicts of interest, financial or otherwise, are declared by the authors.
The present study does not constitute endorsement by the American College of Sports Medicine.
1. American Thoracic Society. Standardization of spirometry, 1994 update. Am J Respir Crit Care Med
. 1995; 152 (3): 1107–36.
2. Askanazi J, Silverberg PA, Foster RJ, Hyman AI, Milic-Emili J, Kinney JM. Effects of respiratory apparatus on breathing pattern. J Appl Physiol
. 1980; 48 (4): 577–80.
3. Banzett RB, Mahan ST, Garner DM, Brughera A, Loring SH. A simple and reliable method to calibrate respiratory magnetometers and Respitrace. J Appl Physiol
. 1995; 79 (6): 2169–76.
4. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet
. 1986; 1 (8476): 307–10.
5. 5. Bloch KE, Barandun J, Sackner MA. Effect of mouthpiece breathing on cardiorespiratory response to intense exercise. Am J Respir Crit Care Med
. 1995; 151 (4): 1087–92.
6. Brüllmann G, Fritsch K, Thurnheer R, Bloch KE. Respiratory monitoring
by inductive plethysmography in unrestrained subjects using position sensor-adjusted calibration. Respiration
. 2010; 79 (2): 112–20.
7. Caretti DM, Pullen PV, Premo LA, Kuhlmann WD. Reliability of respiratory inductive plethysmography for measuring tidal volume during exercise. Am Ind Hyg Assoc J
. 1994; 55 (10): 918–23.
8. Chadha TS, Watson H, Birch S, et al. Validation of respiratory inductive plethysmography using different calibration procedures. Am Rev Respir Dis
. 1982; 125 (6): 644–9.
9. Clarenbach CF, Senn O, Brack T, Kohler M, Bloch KE. Monitoring
during exercise by a portable respiratory inductive plethysmograph. Chest
. 2005; 128 (3): 1282–90.
10. Gilbert R, Auchincloss JH, Brodsky J, Boden W. Changes in tidal volume, frequency, and ventilation
induced by their measurement. J Appl Physiol
. 1972; 33 (2): 252–4.
11. Gribbin HR. Using body surface movements to study breathing. J Med Eng Technol
. 1983; 7 (5): 217–23.
12. Grossman P. The LifeShirt: a multi-function ambulatory system monitoring
health, disease, and medical intervention in the real world. Stud Health Technol Inform
. 2004; 108: 133–41.
13. Grossman P, Spoerle M, Wilhelm FH. Reliability of respiratory tidal volume estimation by means of ambulatory inductive plethysmography. Biomed Sci Instrum
. 2006; 42: 193–8.
14. Grossman P, Wilhelm F, Brutsche M. Accuracy of ventilatory measurement employing ambulatory inductive plethysmography during tasks of everyday life. Biol Psychol
. 2010; 84 (1): 121–8.
15. Jones B, Jarvis P, Lewis JA, Ebbutt AF. Trials to assess equivalence: the importance of rigorous methods. BMJ
. 1996; 313 (7048): 36–9.
16. Kent L, O’Neill B, Davison G, Nevill A, Elborn JS, Bradley JM. Validity and reliability of cardiorespiratory measurements recorded by the LifeShirt during exercise tests. Respir Physiol Neurobiol
. 2009; 167 (2): 162–7.
17. Konno K, Mead J. Measurement of the separate volume changes of rib cage and abdomen during breathing. J Appl Physiol
. 1967; 22 (3): 407–22.
18. Lafortuna CL, Passerini L. A new instrument for the measurement of rib cage and abdomen circumference variation in respiration
at rest and during exercise. Eur J Appl Physiol Occup Physiol
. 1995; 71 (2-3): 259–65.
19. McCool FD, Kelly KB, Loring SH, Greaves IA, Mead J. Estimates of ventilation
from body surface measurements in unrestrained subjects. J Appl Physiol
. 1986; 61 (3): 1114–9.
20. O’Connor DP, Mahar MT, Laughlin MS, Jackson AS. The Bland–Altman method should not be used in regression cross-validation studies. Res Q Exerc Sport
. 2011; 82 (4): 610–6.
21. Perez W, Tobin MJ. Separation of factors responsible for change in breathing pattern induced by instrumentation. J Appl Physiol
. 1985; 59 (5): 1515–20.
22. Sackner JD, Nixon AJ, Davis B, Atkins N, Sackner MA. Non-invasive measurement of ventilation
during exercise using a respiratory inductive plethysmograph. Am Rev Respir Dis
. 1980; 122 (6): 867–71.
23. Sackner MA, Watson H, Belsito AS, et al. Calibration of respiratory inductive plethysmograph during natural breathing. J Appl Physiol
. 1989; 66 (1): 410–20.
24. Sartene R, Dartus C, Bernard JL, Mathieu M, Goldman MD. Comparison of thoracoabdominal calibration methods in normal human subjects. J Appl Physiol
. 1993; 75 (5): 2142–50.
25. Stagg D, Goldman M, Davis JN. Computer-aided measurement of breath volume and time components using magnetometers. J Appl Physiol
. 1978; 44 (4): 623–33.
26. Strömberg NO. Error analysis of a natural breathing calibration method for respiratory inductive plethysmography. Med Biol Eng Comput
. 2001; 39 (3): 310–4.
27. Watson HL, Poole DA, Sackner MA. Accuracy of respiratory inductive plethysmographic cross-sectional areas. J Appl Physiol
. 1988; 65 (1): 306–8.
28. Weissman C, Askanazi J, Milic-Emili J, Kinney JM. Effect of respiratory apparatus on respiration
. J Appl Physiol
. 1984; 57 (2): 475–80.
29. Wilhelm F, Roth W, Sackner M. The lifeShirt. An advanced system for ambulatory measurement of respiratory and cardiac function. Behav Modif
. 2003; 27 (5): 671–91.
30. Witt J, Fisher J, Guenette J, Cheong K, Wilson B, Sheel A. Measurement of exercise ventilation
by a portable respiratory inductive plethysmograph. Respir Physiol Neurobiol
. 2006; 154 (3): 389–95.
31. Zimmerman PV, Connellan SJ, Middleton HC, Tabona MV, Goldman MD, Pride N. Postural changes in rib cage and abdominal volume–motion coefficients and their effect on the calibration of a respiratory inductance plethysmograph. Am Rev Respir Dis
. 1983; 127 (2): 209–14.