Introduction

Significant recent developments in air displacement plethysmography (ADP) have led to the general availability of a commercial product, the BOD POD Body Composition System (Life Measurement, Inc., Concord, CA, USA), for measuring body volume (BV). Major obstacles have apparently been overcome in this system, which previously prevented attempts to supercede traditional hydrostatic weighing (HW) techniques by ADP.^1,2 The BOD POD protocol offers several practical benefits, in that it is quick and easy, noninvasive, and is potentially more acceptable than HW to a wide range of subjects (most notably children, the elderly, sick and disabled persons). However, the accuracy and precision of ADP need to be fully quantified if this technique is to be used routinely as a valid alternative to HW.

Various studies have assessed the accuracy of body composition estimates obtained with ADP against established techniques, such as HW, in both adults^{3,4,5,6,7,8,9,10,11,12} and children.^5,6,10,13,14 However, although reported estimates of mean differences (biases) between HW and ADP are generally relatively small, there are inconsistencies in the observed direction of these biases and wide variation for individuals, as shown by a large range for the limits of agreement between methods (for review, see Fields¹⁵). Clearly, the differences concerning methodological factors in measuring BV will be the main cause of any discrepancies in body composition estimates (eg body fat) between HW and ADP, since such estimates are based on the same two-component model (2-CM) in both techniques.¹⁶

Inconsistencies in findings among various apparently similar studies could be due to differences between laboratories using apparently the same HW techniques, but where certain aspects of their practical application may differ. For example, different methods have been applied to assess the volume of air in the lungs that is present at the exact moment that body weight (BWt) underwater is recorded: some laboratories estimated what the residual lung volume would be at this time using an isolated procedure before measuring BWt underwater (eg Nunez et al¹⁰), whereas others measured the lung volume at this exact moment by using gas dilution techniques (eg Dewit⁶) because, otherwise, it is difficult to ensure a consistent volume of air in the lungs during this comparatively arduous procedure.

Such practical inconsistencies do not create the same concerns with the BOD POD system as all instruments are standardized, with each seemingly manufactured to the same specifications and with an identical protocol recommended. However, it is important to understand the fundamental principles behind the BOD POD as there are some potential misconceptions with the manner in which it operates, which have a bearing on the distinction made in this study between precision and accuracy, and why the decision was made to apply predicted instead of measured lung volume to obtain estimates of actual BV and body composition.

The measurement of uncorrected (raw) BV is obtained by the difference between the volumes of air in the ADP chamber when empty and with a subject present. In contrast to HW, in which lung air is integral to BWt and BV underwater, in the BOD POD, lung air is continuous with chamber air and, therefore, does not constitute an integral part of the uncorrected BV estimate. However, it is necessary to take into account the isothermal effects of air in the lungs, and the surface area artifact, on the measurement of chamber air when a subject is present, in order to make appropriate corrections and provide an actual BV.² In principle, as isothermal gas is more compressible than the adiabatic chamber air; the chamber air is overestimated by the influence of lung air and the surface area artifact. Consequently, uncorrected BV of the subject is an underestimate of actual BV (as it is calculated by difference). These influences are not necessarily consistent between measurements because of different factors, such as temperature gradients, posture and so on, and so the correction factors applied to lung volume and skin surface area are estimates themselves.

Furthermore, another practical limitation of this ADP technique is that lung volume cannot be measured at the exact time of the uncorrected BV measurement. Therefore, whether measured or predicted, the value chosen for lung volume is only an estimate of the influence of lung gases on BV measurement at that particular time. In addition, the same value is applied by the BOD POD software to both measurements within a test procedure when estimating actual BV, so that the extent of precision remains unaffected as there is no additional variability introduced from this source. Also, there appears to be no real advantage with accuracy if measured lung volume is used instead of predicted values,¹⁷ especially as any error is reduced to only 40% of any discrepancy (a factor of 0.4 is assumed for the correction needed to account for isothermal lung gases²). Corrections for the influence of lung volume and surface area artifact affect the accuracy of BV measurements, whereas variations in the prevailing conditions in the BOD POD during a measurement will affect their precision.

Certain investigators have used repeated measurements to demonstrate that individual BOD POD systems have reasonable precision.^18,19 However, the extent of between-laboratory variation in ADP is not yet known. Reproducibility among different laboratories is an important consideration that may influence comparative studies of ADP against HW, in addition to large-scale, multicenter research conducted using ADP.

Therefore, the aims of this study were (a) to establish and compare within-laboratory precision (repeatability) in BV measurements from two BOD POD ADP instruments, located at two different laboratories in relatively close proximity; (b) to determine between-laboratory reproducibility in BV and so ascertain whether or not results from these two ADP systems could be used interchangeably, if required; (c) to identify potential sources of any within- and between-laboratory variation when using ADP; and (d) to interpret this variability in BV in terms of the precision of body composition estimates.

Methods

Details of the heterogeneous sample of 30 healthy adult subjects (16 females and 14 males) who volunteered for the study are shown in Table 1. In addition, one male (age, 28 y; weight, 62.98 kg; height, 1.78 m) and one female (age, 22 y; weight, 56.29 kg; height, 1.62 m) subject volunteered independently for 10 repeat measurements using each ADP instrument. Written informed consent was obtained. Ethical approval for the study was granted by the London Metropolitan University Ethics Committee and the Institute of Child Health Ethical Advisory Board.

Table 1 - Subject characteristics (mean and standard deviation (SD)).

Full table

Each subject was assessed on the same day using ADP systems at two different laboratories in relatively close proximity: laboratory 1, London Metropolitan University; laboratory 2, Institute of Child Health, London. Subjects were asked not to eat for 2 h prior to the first measurement, and during the testing period all subjects refrained from food or drink. Minimal amounts of water were allowed to be sipped in the event of dry throat caused by traveling between laboratories. Subjects were required to void the bladder immediately before the first set of measurements with the intention that the next voiding would not be necessary until the second set of measurements had been completed. To avoid the possibility of introducing a systematic error, the initial measurements on each individual were not routinely taken at either laboratory. Instead, half of the initial measurements were performed first at laboratory 1 and the other half at laboratory 2. In order to eliminate interobserver variability and identify the extent of variation attributable to BV measurement alone, a single trained observer performed all measurements at both the laboratories.

The journey between laboratories involved a short underground train journey and a 5–10 min walk either side, in total taking not more than 40 min. In order to account for potential within-individual variability caused by this traveling (ie insensible losses and substrate metabolism), two subsamples, five subjects each, were created within the study sample. Immediately after finishing at the second laboratory, individuals allocated randomly to each of these two subsamples returned to the first laboratory, in which their initial assessments had been performed, for a third set of measurements: the first subgroup starting and finishing at laboratory 1, and the second subgroup starting and finishing at laboratory 2. The average time taken to perform duplicate measurements in subjects at both laboratories, including the traveling in between, was 2 h (SD, 34 min).

Ambient temperature and atmospheric pressure were recorded during each session at both laboratories in an attempt to identify any environmental influence on within- and between-laboratory variation.

Air displacement plethysmography

The BOD POD ADP system was prepared strictly according to the manufacturer's instructions prior to each measurement session; this preparation and the whole measurement procedure were identical in each setting, the details of which are provided elsewhere.^1,2,19 For clarity, in this particular study, an ADP test procedure consisted of two uncorrected BV measurements and, if these two values were within a predetermined limit of 150 mls (a value set by the manufacturer), their mean was used by the instrument software for subsequent calculations of corrected BV, body density (BD) and body composition (see below and Dempster and Aitkens²). The values for the uncorrected BV measurements appeared transiently on the computer screen, but were not otherwise available, and were recorded for subsequent assessment of precision.

Figure 1 is a simple schematic of the suggested process for obtaining, with greater certainty, an assessment of BV using ADP. Such routine use of repeated procedures has previously been recommended, as an additional precautionary measure to minimize the occurrence of rogue values.¹⁸ According to this recommendation, a third test procedure would be conducted if the two ADP test procedure values differed by more than a predetermined value for BD of 0.007 kg/l, corresponding to a value of about 2% body fat (this is based on the previous finding that 95% of repeated measurements differed by less than 0.007 kg/l ¹⁸). However, for this study, all test procedure values were recorded and used in subsequent calculations of precision.

Figure 1.

Schematic representing the assessment of body volume as recommended^*† when using the BODPOD body composition system air displacement plethysmograph.

Full figure and legend (83K)

Two reference cylinders of accurately known fixed volumes (nominal volume approximately 20 and 50 l), and one male and one female volunteer were subjected to 10 successive repeated test procedures conducted at both laboratories, with the exact protocol reproduced in each. The 30 individuals of the main study group were measured in duplicate only at both laboratories. The procedure for the cylinders was identical to that for the human measurements, but without requiring corrections for lung volume and surface area artifacts applied to obtain actual BV.

The standing height was measured to the nearest 0.1 cm using a wall-mounted stadiometer. Repeated test procedures, with each of the four measurements of uncorrected BV lasting approximately 40 s, were conducted with the volunteers wearing their own minimal clothing (ie lycra swimsuit for the women, and lycra trunks or tight-fitting shorts for the men), the attire being identical at both laboratories. Subjects were weighed to the nearest 0.01 kg, on the weighing scale integral to the ADP system, and BV was assessed with the subjects sitting in the BOD POD wearing a standard tight-fitting acrylic swim cap.

Uncorrected BV was adjusted for predicted lung volume, instead of measured lung volume, to eliminate potential variability in lung volume measurement, as documented previously,^18,19 which would have confounded the estimation of precision of actual BV measurement (see Introduction). The BOD POD system software used actual BV, after correction for lung volume and body surface area artifact, along with BWt in air to calculate BD (BD=BWt/actual BV kg/l), from which the fraction of body fat (%BF) was calculated using the equation of Siri¹⁶ based on the assumed densities of 0.9 kg/l for fat and 1.1 kg/l for fat-free mass (FFM):

Fat (kg) and FFM (kg) were also calculated.

Statistical analysis

Elements of precision within each laboratory (repeatability) and between the two laboratories (reproducibility) were elucidated according to British Standard 5497, part 1.²⁰ Although the sample group was not randomly selected, so not necessarily demonstrating a normal distribution, it was assumed that any differences between measurements were normally distributed. The mean, standard deviation (SD) and coefficient of variation (CV=100SD/overall mean) were calculated for each series of 10 repeats for the reference cylinders and for the two human subjects. Bias (mean difference) and SD of the differences (SD_d=[d²/n], where d is the difference between measurements and n is the number of repeat measurements) were obtained between measured and known volumes of the reference cylinders for the ADP systems at each laboratory (precision of a single measurement may be estimated from the precision of repeated measurements (eg SD_d) divided by 2). Bias and SD_d were obtained between the two ADP instruments for the cylinders and for the two volunteers, in whom body composition estimates were also compared. Agreement between laboratories for measured BV and derived body composition for the 30 volunteers was assessed using bias and 95% limits of agreement (bias2SD_d), according to the method of Bland and Altman.²¹ Paired sample t-tests were performed to examine the significance of any differences within- and between-laboratory (a P-value of <0.05 was taken as significant). The strengths of relationships were assessed using Pearson's correlation coefficient.

In order to allow appropriate comparisons with other reports, which might include estimates of precision expressed in different ways, the technical error of measurement (TEM) may be calculated from the information presented: TEM=[d²/2n] (ie SD_d/2), representing the standard error of a single measurement, such that 95% of estimates would be expected to fall within the true value 2 TEM (eg Withers et al²²), and is the absolute equivalent of the measurement error in relative terms given by 2 CV (calculated as above). The coefficient of repeatability (CR) may also be calculated: CR=2 SD_d, representing values within which the difference between two repeats would be expected to fall in 95% of cases.²⁰

The contribution of biological (b) and methodological (m) variabilities to total (t) measured variability was assessed using:

whereby V represents SD, all in the same units, with V_m and V_b known to be uncorrelated.¹⁸ The biological contribution to the total variability has two components. One of these biological components concerns the technical aspects of the procedure, that is how repeatable are factors such as the positioning of a subject in the ADP chamber, the tidal lung volume, skin temperature and lung gas temperature, etc, which may have small effects on precision. This within-individual variability is an additional consideration to the purely technical precision observed when, for example, the reference cylinders are assessed. This aspect of precision is integrated into the methodological variability to enable quantification of, and to distinguish it from, the second biological component that concerns the much larger inter-individual variability, which might be observed in a sample in cross-sectional or in longitudinal studies of individuals with changing body composition, for example due to weight-loss regimens or clinical interventions. Such a distinction has been made possible in this study.

Results

Within-laboratory variability (repeatability)

The precision for the 10 repeated test procedures on the two reference cylinders of known volumes at both centers (Table 2) represents technical precision only, that is no involvement of biological factors. The precision for the 10 repeated procedures in the two independent volunteers at both centers (Table 3) showed greater variability at laboratory 2. The major contribution to the total methodological precision for BV in these two subjects was clearly the within-individual biological variability. Apart from the male subject at laboratory 1, with only about 45 or 55% (depending on which value for precision is used) of the still relatively low variability contributed by biological factors, 78–95% or more of the total variability was due to within-individual biological influences, regardless of which cylinder data were used for these estimations. As expected, because of the nature of the calculations, the imprecision for body fat estimates were clearly larger than for the uncorrected BV measurements (Table 3 and Discussion).

Table 2 - Within-laboratory precision (repeatability) for repeated (n=10) test procedures (mean of two test measurements) for reference cylinders at two laboratories (laboratory 1, London Metropolitan University; Laboratory 2, Institute of Child Health, London).

Full table

Table 3 - Within-laboratory precision (repeatability) for repeated (n=10) test procedures (mean of two test measurements) for human volunteers at two laboratories (laboratory 1, London Metropolitan University; laboratory 2, Institute of Child Health, London).

Full table

Repeatability for BWt, BV and derived body composition estimates based on two test procedures on the 30 volunteers is shown in Table 4. There was some variability associated with BWt that, although much less than that for BV, influenced estimates of body composition from BD due to the nature of the calculations (above and Discussion). Following estimation of body composition indices from BD, an outlier at laboratory 2, which had a major effect on variability, became more evident than when BV was scrutinized in isolation. Values obtained with this outlier removed are presented in Table 4 in parentheses to enable evaluation of precision with and without the inclusion of such rogue values.

Table 4 - Within-laboratory precision for measured BWts, BVs and derived body composition using repeated test procedures for 30 volunteers (16 females, 14 males) at each of the two laboratories (laboratory 1, London Metropolitan University; laboratory 2, Institute of Child Health, London).

Full table

Between-laboratory variability (reproducibility)

The reproducibility of measurements between laboratories for the reference cylinders is shown in Table 5. The bias is clearly of more material importance than the variability, as in only one case of the small cylinder at laboratory 2 do the 95% limits of agreement include zero bias between the reference and measured values. In addition to the significant underestimation of cylinder volume at both laboratories (Table 5), there was also a significant difference in volume (P<0.05) between them for both the small cylinder (bias, -0.06 l; SD_d, 0.05 l; 95% limits of agreement, -0.16 to 0.04 l) and the larger cylinder (bias, -0.03 l; SD_d, 0.04 l; 95% limits of agreement, -0.04 to 0.11 l). Depending on the planned use of ADP, the biases between instruments may be of less importance than the difference between the reference cylinder volume and the measured value because the 95% limits of agreement incorporate zero bias, even if the absolute measurement is in error for both. Using the large cylinder as an example and assuming a BWt of 50 kg, a 0.1 l bias would translate into a difference in body fat of 1.0% of BWt. Of course, the measured values for the human subjects were not compared to any reference in the same way, as the absolute value for their BV could not be known for certain. However, there was a significant difference (P<0.01) between laboratories for estimated uncorrected BV for both the male (bias, -0.21 l; SD_d, 0.14 l; 95% limits of agreement, -0.50 to 0.08 l) and the female (bias, -0.29 l; SD_d, 0.14 l; 95% limits of agreement, -0.57 to 0.01 l) independent volunteers.

Table 5 - Accuracy and between-laboratory precision (reproducibility) for repeated test procedures (n=10) between laboratory 1 (London Metropolitan University) and laboratory 2 (Institute of Child Health, London) for reference cylinders assessed using bias and 95% limits of agreement²¹.

Full table

Table 6 presents the reproducibility for BWt, BV and derived body composition estimates between laboratories for the 30 volunteers. Figure 2 is a graphical representation of the BV data. None of the biases between laboratories for any of these measures of body composition reached significance, neither were such differences governed by sex, BWt, height or fatness. Although log transformation of the data was possibly indicated from the apparent increased spread of points with increased BV in Figure 2, no additional information was forthcoming with such transformation. Comparison of the total variability (SD_d) for BV presented in Table 6 compared with the purely methodological equivalent obtained with the reference cylinders (SD_d between laboratories, as above) indicated that the within-individual biological component of the methodological precision accounted for between 97 and 99% of the total variability between laboratories.

Figure 2.

Agreement between laboratories, showing bias (mean difference) and 95% limits of agreement (2SD), for body volume (BV; liters) measured using the BOD POD body composition system at two different laboratories (difference between measurements against their mean.²¹

Full figure and legend (19K)

Table 6 - Agreement between laboratory 1 (London Metropolitan University) and laboratory 2 (Institute of Child Health, London) for BWts, BVs and derived body composition based on repeated test procedures for 30 volunteers (16 females, 14 males) assessed using bias and 95% limits of agreement²¹.

Full table

Environmental conditions

Although there was a significant (P<0.01) difference in temperature between laboratories (laboratory 1: mean, 23.27°C; 95% confidence intervals (CI), 22.73–23.81°C; laboratory 2: mean, 22.55°; 95% CI, 22.37–22.73°C), and the difference was related to the absolute temperature on the day, this did not account for the bias in body composition estimates observed in this study, as there was no indication of any relationship between the change in temperature between laboratories and the observed difference in estimates. Atmospheric pressure differences between laboratories were negligible (mean, 0.7 mmHg), with no relationship between these and body composition differences.

At each laboratory, the small values for the precision of repeated BV procedures (CV, 0.8%) on the subjects returning to the first site for additional measurements, to test the possible effects of travel, etc, were similar to that given in Table 4, with no obvious outliers (ie each was internally consistent). Comparison of the initial repeated test procedures at each laboratory with the repeated test procedures on return to that same laboratory indicated a significant (P<0.05) bias in BV of 0.43 l and 95% limits of agreement of -0.58 to 1.43 l. In part, this was due to one obvious outlier, the removal of which reduced this bias to 0.31 l with 95% limits of agreement of -0.43 to 1.06 l. Interestingly, and to reiterate that stated above, this outlier was internally consistent at both visits. Further examination of the data indicated that this significant bias originated almost entirely at laboratory 1, for which the bias (significance, P<0.05) was 0.64 l with 95% limits of agreement of -0.29 to 1.57 l with the same outlier included; the bias was 0.46 l with 95% limits of agreement of -0.02 to 0.93 l with it excluded from the analysis. This is in contrast to the consistency observed for the initial and return visit measurements at laboratory 2, in which there was a small (not significant) bias of 0.18 l and 95% limits of agreement of -0.62 to 0.97 l.

These differences in BV assessments were mirrored by the change in temperature observed between the initial and return measurements, but not by any pressure changes, which were negligible. The difference in temperature for both laboratories combined (initial visit: mean, 22.66°C; 95% CI, 22.45–22.87°C; return visit: mean, 23.59°C; 95% CI, 23.19–23.99°C) was significant (P<0.01). However, when each laboratory was analyzed individually, only the bias for laboratory 1 (initial visit: mean, 22.62°C; 95% CI, 22.41–22.83°C; return visit: mean, 24.18°C; 95% CI, 23.97–24.39°C) proved to be significant (P<0.001), as the bias for laboratory 2 (initial visit: mean, 22.70°C; 95% CI, 22.31–22.61°C; return visit: mean, 23.00°C; 95% CI, 22.69–23.39°C) did not reach significance. Furthermore, in contrast to the evident lack of relationship between the differences in BV and the differences in temperature between laboratory 1 and laboratory 2 for the 30 subjects, the changes in temperature between the initial measurements and those of the return visit were significantly related to the differences in BV (r=0.63; P<0.05).

Discussion

The BOD POD body composition system is one of many new and increasingly sophisticated techniques available for the determination of body composition in vivo. However, assessments of BV provided by this improved ADP instrumentation are still only estimates, even if they should prove to have similar or better levels of accuracy and precision than existing methods, such as HW (for review, see Fields et al¹⁵). Therefore, it is important to understand the probable extent and underlying causes of inaccuracy and imprecision when evaluating estimates provided by ADP systems. The precision associated with individual ADP systems has been assessed previously,^18,19 leading to recommendations for duplicating the test procedure in order to reduce or eliminate the possibility of rogue results.¹⁸ However, this is the first study to investigate the extent of reproducibility of BV measurement between two independent ADP systems located in different laboratories. Certain possible underlying causes of variation between instruments have also been identified, and the effects of such variation in terms of body composition estimates have been demonstrated.

The estimation of precision for the BOD POD ADP system is influenced by its operational procedure; that is, the values for BV and derived body composition, either stored in the accompanying computer software or provided as hard copy, are calculated from the means of the two (or the best two of three) uncorrected BV measurements that constitute a test procedure (see schema, Figure 1). Owing to this, the apparent precision of the mean values is usually found to be better than the precision of individual measurements. The individual test measurement figures for uncorrected BV appear transiently on the computer screen, and can be recorded at that moment, but are not retrievable afterwards with current software. As the precision of these repeated measurements has been reported previously,¹⁸ this study has focused on the variability of the results obtained from each whole test procedure, evaluating and reporting within-laboratory precision, or repeatability, and reproducibility between laboratories.²⁰

The use of reference cylinders of known volumes enabled estimation of the technical precision of each instrument without the confounding effects of within-individual variability. The relatively low values obtained for precision (Table 2) confirmed previous findings.^2,19 However, a major concern identified here is that there was a bias observed between the instruments at the two laboratories for these reference cylinders (results), as well as both instruments substantially underestimating the reference value (Table 5). These findings persisted to a large extent when human subjects were repeatedly assessed at both laboratories so that, as expected, the imprecision was greater (Tables 3 and 4) due to inconsistencies in within-individual biological factors such as subject positioning and movement in the ADP chamber, breathing patterns, skin and lung temperature and clothing, albeit minimal. Ambient conditions such as temperature and airflow may also influence these factors (see below). Although there was also a substantial bias between laboratories for the 10 repeated procedures on the two human volunteers (results), as for the cylinders (above), this was not evident for the 30 subjects in whom procedures were performed in duplicate only (Table 6). For the cylinders, the bias may have been due to the only variable not really controlled for, which was environmental temperature (atmospheric pressure was fairly constant). In the two human subjects, the explanation for this may originate in part from the number of measurements obtained and the time over which they were taken, leading to greater differences overall in the 10 repeats because of possible changes in BWt following greater body water and food fluxes, and temperature variations. However, no obvious trends in BWt were evident.

Although we did attempt to minimize the effects of traveling, temperature differences and within-individual biological influences between laboratories, by performing all measurements on the same day with minimal time delay, it was difficult to schedule all subjects in the morning and it was considered unethical to instruct them to go without any food or drink for such long periods of time (especially for the volunteers measured 10 times at both laboratories, including the traveling in between). The standardization of conditions for comparisons to be made between subjects was not a primary concern of this study, provided that repeated BV measurements for each subject were obtained under identical conditions. Therefore, subjects studied in the mornings were required to fast overnight and those studied later in the day were not permitted to eat for at least 2 h before the first measurement. None of the subjects were allowed to eat between measurements, and only minimal water was allowed to be sipped to counteract dry throat due to traveling. All subjects voided their bladders before the first measurement and were encouraged not to void between measurements. Only two subjects voided during the course of the measurements reproduced at the two laboratories and two others before returning to the first laboratory for repeated procedures. The error from this source was less than 0.6% in all four cases and, although contributing negligibly to the variability in mean BV values overall, BV estimates were adjusted accordingly. Any postprandial effects on precision (eg production of intestinal gas only, as the subjects were not allowed to consume food between visits) for the 30 subjects measured at both laboratories were minimized by randomly assigning the initial venue for each subject (this did not occur for the two independent volunteers undergoing multiple repeat measurements, and this may have contributed more to variations in BV estimates, despite no evidence of changing BWt).

Apart from the large reference cylinder volume measured at laboratory 2, for which the internal precision was similar to that for laboratory 1 (Table 2), there was generally a greater variability of repeated measurements observed at laboratory 2 (Tables 2, 3 and 4). However, the precision was still reassuringly good, with the largest CV being only 0.24% for volume (Table 2). This difference may be inherent in the two instruments, or it may be due to slight changes in other factors, such as environmental temperature and pressure. Although these factors were not responsible for the bias between instruments, temperature differences between the first and second visits to the sites of the initial measurements were significantly related to differences in BV. This suggests that ambient conditions may affect the workings of the BOD POD, even though temperature and pressure are accounted for in the calculations of BV, based on the principles of the Gas Laws (Poisson's and Boyle's Laws). Indeed, the impacts of temperature, pressure and humidity have all been previously implicated in the variability of ADP measurements.^23,24 There are differences in the placement of the instruments in the two laboratories, which may exacerbate the effects of ambient conditions, thus creating variability in measurements. At laboratory 1, the BOD POD is situated in a small room within a much larger room and with only walls for separation between them, that is, no ceiling, so that ambient conditions are consistent and interdependent during periods of change. For example, the room heating is activated in the early morning, but takes some time to reach equilibrium. This could affect either the subject's lung and skin temperature or the operating characteristics of the BOD POD, or both, so that during the day the difference in temperature could explain the bias in BV when the subjects returned for a second set of measurements. The effects of opening and shutting any doors in close proximity, or even elsewhere in the building depending on the type of environmental control utilized, may be dampened because of the large room volume. In contrast, at laboratory 2 the BOD POD is situated in a small room isolated from others, but where opening and closing of doors in the vicinity may affect pressure and temperature very much more quickly. In addition, there is an air conditioning system in this room that quickly establishes and then maintains a relatively constant ambient temperature. The variable influence of door usage was noticed in a previous setting, whereby opening the door to the assessment room coincidentally with ADP measurement had no major impact, whereas opening either of the hallway or external doors almost always compromised the integrity of the measurement. Unfortunately, as this phenomenon was never fully defined scientifically, it remains a cautionary note only. Therefore, the slightly greater imprecision in laboratory 2 could be due to greater fluctuations in ambient conditions caused by the more rapidly changing local factors, the influences of which have less impact at laboratory 1. However, the more constant temperature at laboratory 2 appeared to ensure that there was no bias when subjects returned there for the additional set of measurements.

At both laboratories, the precision of BV measured by ADP was comparable to previously presented values.^{4,6,11,18,19,25} These values are also probably dependent to some extent on prevailing ambient conditions and variations in the performance of individual instruments, although the latter should be minimized as each instrument is of a standard manufactured by a single company. Accordingly, a variable level of precision (although even the worst case was still very good) has been observed in a number of different subjects undergoing repeated measurements in other studies (JCK Wells and JE Williams, unpublished). Nevertheless, the results obtained in this study and others confirm the consistently equal or better precision of the ADP system compared to that reported for HW,^6,22,26,27 which is typically 2.0%.²⁸ Furthermore, the between-laboratory bias reported here corresponded to a measurement error, which is less than that for HW^29,30 and is more comparable to the reported precision of HW in a single laboratory.^{6,22,26,27,28} The better precision of ADP combined with the ease of application are two important factors suggesting that ADP is a suitable alternative to HW, especially if its accuracy can be established unequivocally.

To reiterate, it is of concern that biases were observed in this study between the ADP instruments at both laboratories and the known volumes for the reference cylinders, but estimating the volumes of cylinders is not the primary aim of the BOD POD and, as such, and along with the factors discussed (above), should be used as a guide only. Furthermore, such comparative findings may depend on day-to-day prevailing conditions in each laboratory, which may influence both the size and direction of any bias. Importantly, the bias between the two instruments for BV of this sample of 30 subjects was not significant, implying that they can be used interchangeably for groups of subjects with similar characteristics. However, the larger 95% limits of agreement indicate that greater consideration may be needed for the proposed assessment of individuals with different ADP instruments. Clearly, the use of a single ADP instrument is preferred when assessing individuals on a longitudinal basis, especially in view of the reasonably good precision within each instrument confirmed by this study.

The precision of BV and BWt estimates, established in terms of both repeatability and reproducibility in this study, translates into evidently less precise estimates of BD and concomitant body composition indices (Table 6). The nature of the calculation generates this increased level of imprecision as, although an error in BD might appear to be small compared to the overall BD, the true error is relatively much larger over a limited theoretical range for BD of only 0.9 (assumed density of pure fat) to 1.1 kg/l (assumed mean density of FFM), the values on which these densitometric methods were founded for calculating the proportion of body fat.^16,31

The primary purpose of the study was to assess precision and not accuracy, which would have required a different study design and has been variously reported on many previous occasions (for review, see Fields¹⁵). As reasoned in the introduction, it is important to appreciate that the precision of BV (uncorrected and actual) measurement is not influenced by the figures chosen to represent lung volume, but the accuracy of actual BV and body composition estimates may be affected considerably by whether these figures, measured or predicted, are valid estimates of lung volume at the exact time that the BV measurement is obtained. Only after the uncorrected BV has been obtained does the instrumentation software attempt to adjust this measurement for the confounding and uncertain effects of warmer and more humid air in the lungs and air close to the skin, in order to provide the best possible estimate of actual BV. Air of inconsistent temperature gradients and comprising molecules of different sizes causes artifacts in volume measurement through inconsistent changes in the compressibility of chamber air. Estimates of lung volume and skin surface area are used to provide approximate quantification for these artifacts through established correction factors, which are generalized estimates only because of inconsistencies in breathing patterns, assumptions of mean lung volume, relative concentrations of diatomic and triatomic gases, posture, etc, and are ubiquitously applied to the uncorrected BV to provide an actual BV. Estimates of the lung volume correction are not subtracted from the uncorrected BV estimates, as they are with HW, a common misconception (eg Weyers et al ³²). In contrast, artifacts caused by the greater compressibility of isothermal air compartments within the otherwise adiabatic conditions of the BOD POD provide estimates of uncorrected BV that are actually lower than the true value, and a correction based on lung volume is, therefore, added for estimating actual BV. Therefore, instructing the subjects to comply with as near-normal tidal breathing patterns as possible and consistent posture should minimize any imprecision and inaccuracy from this source, but it is extremely unlikely to eliminate them altogether. Indeed, although measurement of lung volume has been advocated,¹⁷ this particular recommendation appears to be based on inconsequential differences between predicted and measured values without knowledge of lung volume at the exact time of measurement of BV. Furthermore, it has been shown that repeats of lung volume assessments may differ by as much as 0.5 l,^18,19 a difference that translates into only 0.2 l in actual BV. Also, in these particular studies,^18,19 no systematic difference was detected between measured and predicted lung volumes, with a small 95% CI for the estimation of body fat (0.3%). Assessment of the accuracy of actual BV measurements, to include the effect of lung volume assessment, and body composition estimates derived from these would require a different study design altogether, in which BOD POD measurements were validated against a reference method of established accuracy.

The body composition values presented in Table 6 were derived using the 2-CM of analysis,¹⁶ integral to the BOD POD software. BV can also be used in other multi-CMs of body composition (eg Fuller et al,³³ and Wells et al³⁴). Along with actual BV, measured total body water is incorporated in the 3-CM^33,34 and total body water and bone mineral content are integrated into the terms in the 4-CM,^33,34 effectively removing the need for the more major assumptions of constant relationships within the healthy body. Therefore, any imprecision in BV and BWt has less effect in these multi-CMs than it does in the 2-CM. For example, a bias of 0.03 kg for BWt combined with a bias of -0.05 l for BV (rounded figures taken from Table 6) applied to the mean values in this table (BWt, 70.30 kg; BV, 66.89 l), and assuming zero error in both total body water and bone mineral content, leads to a relative error in the 2-CM for body fat (as %BWt) of -0.55% (equivalent to an absolute error in body fat of -0.38 kg; FFM, 0.41 kg), in the 3-CM for body fat as %BWt of -0.23% (body fat, -0.15 kg; FFM, 0.18 kg) and in the 4-CM for body fat as %BWt -0.29% (body fat, -0.20 kg; FFM, 0.23 kg).

The data presented in this report enable the usefulness of the BOD POD to be evaluated for a number of purposes from cross-sectional studies of groups of interest to longitudinal studies following the effects of intervention, and may indicate whether or not the use of multiple instruments is justified. For example, a recent report in which ADP was used³² identified a mean decrease in fat mass, due to a weight-loss regimen, of 3.5 (SD 2.5) kg. This value is substantially greater than the measurement variability of either a single BOD POD instrument (Table 4) or between the two instruments in the two different locations (Table 6), so that the integrity of such changes can be viewed with confidence.

Conclusion

The different aspects of precision associated with ADP were observed to be reasonably good both within and between laboratories studied here. Such precision, along with its speed and ease of measurement, confirms that this ADP system could be used in preference to HW, making a valuable contribution in the long term to the field of body composition. However, although the bias between the two instruments was not significant, implying that they can be used interchangeably for groups of subjects, the relatively large 95% limits of agreement indicate greater concern for assessing individuals with different instruments. Furthermore, certain factors in the measurement protocol, the setting of the BOD POD and ambient conditions need to be carefully controlled in order to minimize potential measurement variability. Certain precautionary measures can be taken to achieve this, such as placing the BOD POD in a room where sudden pressure fluctuations are avoidable, temperature is kept constant and airflow is controlled. Consistency in subject preparation and compliance with standard instructions regarding breathing patterns and posture are also recommended. This study has also confirmed the value of using two complete test procedures in order to eliminate or at least minimize the occurrence of rogue values.

References

1. McCrory MA, Gomez TD, Bernauer EM, Mole PA. Evaluation of a new air displacement plethysmograph for measuring human body composition. Med Sci Sports Exerc 1995; 27: 1686–1691. | PubMed | ISI | ChemPort |
2. Dempster P, Aitkens S. A new air displacement method for the determination of human body composition. Med Sci Sports Exerc 1995; 27: 1692–1697. | PubMed | ISI | ChemPort |
3. Biaggi RR, Vollman MW, Nies MA, Brener CE, Flakoll PJ, Levenhagen DK, Sun M, Karabulut Z, Chen KY. Comparison of air-displacement plethysmography with hydrostatic weighing and bioelectrical impedance analysis for the assessment of body composition in healthy adults. Am J Clin Nutr 1999; 69: 898–903. | PubMed |
4. Collins MA, Millard-Stafford ML, Sparling PB, Snow TK, Rosskopf LB, Webb SA, Omer J. Evaluation of the BOD POD for assessing body fat in collegiate football players. Med Sci Sports Exerc 1999; 31: 1350–1356. | Article | PubMed | ISI | ChemPort |
5. Demerath EW, Guo SS, Chumlea WC, Towne B, Roche AF, Siervogel RM. Comparison of percent body fat estimates using air displacement plethysmography and hydrodensitometry in adults and children. Int J Obes Relat Metab Disord 2002; 26: 389–397. | Article | PubMed | ChemPort |
6. Dewit O, Fuller NJ, Fewtrell MS, Elia M, Wells JCK. Whole body air displacement plethysmography compared with hydrodensitometry for body composition analysis. Arch Dis Child 2000; 82: 159–164. | Article | PubMed | ChemPort |
7. Fields DA, Wilson GD, Gladden LB, Hunter GR, Pascoe DD, Goran MI. Comparison of the BOD POD with the four-compartment model in adult females. Med Sci Sports Exerc 2001; 33: 1605–1610. | Article | PubMed | ISI | ChemPort |
8. Levenhagen DK, Borel MB, Welch DC, Piasecki JH, Piasecki DP, Chen KY, Flakoll PJ. A comparison of air displacement plethysmography with three other techniques to determine body fat in healthy adults. JPEN 1999; 23: 293–299.
9. Millard-Stafford ML, Collins MA, Evans EM, Snow TK, Cureton KJ, Rosskopf LB. Use of air displacement plethysmography for estimating body fat in a four-component model. Med Sci Sports Exerc 2001; 33: 1311–1317. | Article | PubMed | ISI | ChemPort |
10. Nunez C, Kovera AJ, Pietrobelli A, Heshka S, Horlick M, Kehayias JJ, Wang Z, Heymsfield SB. Body composition in children and adults by air displacement plethysmography. Eur J Clin Nutr 1999; 53: 382–387. | Article | PubMed | ISI | ChemPort |
11. Sardinha LB, Lohman TG, Teixeira PJ, Guedes DP, Going SB. Comparison of air displacement plethysmography with dual-energy X-ray absorptiometry and 3 field methods for estimating body composition in middle-aged men. Am J Clin Nutr 1998; 68: 786–793. | PubMed | ISI | ChemPort |
12. Wagner DR, Heyward VH, Gibson AL. Validation of air displacement plethysmography for assessing body composition. Med Sci Sports Exerc 2000; 32: 1339–1344. | Article | PubMed | ISI | ChemPort |
13. Fields DA, Goran MI. Body composition techniques and the four-compartment model in children. J Appl Physiol 2000; 89: 613–620. | PubMed | ISI | ChemPort |
14. Lockner DW, Heyward VH, Baumgartner RN, Jenkins KA. Comparison of air displacement plethysmography, hydrodensitometry, and dual-energy X-ray absorptiometry for assessing body composition of children 10 to 18 years of age. Ann NY Acad Sci 2000; 904: 72–78. | PubMed |
15. Fields DA, Goran MI, McCrory MA. Body-composition assessment via air-displacement plethysmography in adults and children: a review. Am J Clin Nutr 2002; 75: 453–467. | PubMed | ISI | ChemPort |
16. Siri WE. Body composition from fluid spaces and density: analysis of methods. In: Brozek J, Henschel A (eds) Techniques for measuring body composition. National Academy of Sciences, National Research Council: Washington, DC; 1961. pp 223–244.
17. McCrory MA, Mole PA, Gomez TD, Dewey KG, Bernauer EM. Body-composition by air-displacement plethysmography by using predicted and measured thoracic gas volumes. J Appl Physiol 1998; 84: 1475–1479. | PubMed | ChemPort |
18. Wells JCK, Fuller NJ. Precision of measurement and body size in whole-body air-displacement plethysmography. Int J Obes Relat Metab Disord 2001; 25: 1161–1167. | Article | PubMed | ChemPort |
19. Collins AL, McCarthy HD. Evaluation of factors determining the precision of body composition measurements by air-displacement plethysmography. Eur J Clin Nutr 2003; 57: 770–776. | Article | PubMed | ChemPort |
20. British Standards Institution. Precision of test methods. Part 1: Guide for the determination and reproducibility for a standard test method (BS5497, part 1). BSI: London; 1979.
21. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; I: 307–310.
22. Withers RT, LaForgia J, Heymsfield SB. Critical appraisal of the estimation of body composition via two-, three-, and four-compartment models. Am J Hum Biol 1999; 11: 175–185. | Article | PubMed |
23. Gnaedinger RH, Reineke EP, Pearson AM, Van Huss WD, Wessel JA, Montoye HJ. Determination of body density by air displacement, helium dilution, and underwater weighing. Ann NY Acad Sci 1963; 110: 96–108. | PubMed |
24. Gundlach BL, Visscher GJW. The plethysmometric measurement of total body volume. Hum Biol 1986; 58: 783–799. | PubMed |
25. Vescovi JD, Zimmerman SL, Miller WC, Hildebrandt L, Hammer RL, Fernall B. Evaluation of the BOD POD for estimating percentage body fat in a heterogeneous group of adult humans. Eur J Appl Physiol 2001; 85: 326–332. | PubMed |
26. Durnin JVGA, Taylor A. Replicability of measurements of density of the human body as determined by underwater weighing. J Appl Physiol 1960; 15: 142–144. | PubMed |
27. Mendez J, Lukaski HC. Variability of body density in ambulatory subjects measured at different days. Am J Clin Nutr 1981; 34: 78–81. | PubMed |
28. Going SB. Densitometry. In: Roche AF, Heymsfield SB, Lohman TG (eds). Human body composition. Human Kinetics Publishers: Champaign, IL; 1996. pp 3–21.
29. Deurenberg P, Westerterp KR, Velthuis-Te Wierik EJM. Between-laboratory comparison of densitometry and bio-electrical impedance measurements. Br J Nutr 1994; 71: 309–316. | PubMed |
30. Lohman TG. Advances in body composition assessment. Current issues in exercise science, Monograph number 3. Human Kinetic Publishers: Champaign, IL; 1992.
31. Brozek J, Grande F, Anderson JT, Keys A. Densitometric analysis of body composition: revision of some quantitative assumptions. Ann NY Acad Sci 1963; 110: 113–140. | Article | PubMed | ChemPort |
32. Weyers AM, Mazzetti SA, Love DM, Gomez AL, Kraemer WJ, Volek JS. Comparison of methods for assessing body composition changes during weight loss. Med Sci Sports Exerc 2002; 34: 497–502. | Article | PubMed |
33. Fuller NJ, Jebb SA, Laskey MA, Coward WA, Elia M. Four-component model for the assessment of body composition in humans: comparison with alternative methods, and evaluation of the density and hydration of fat-free mass. Clin Sci 1992; 82: 687–693. | PubMed | ISI | ChemPort |
34. Wells JCK, Fuller NJ, Dewit O, Fewtrell MS, Elia M, Cole TJ. Four-component model of body composition in children: density and hydration of fat-free mass and comparison with simpler models. Am J Clin Nutr 1999; 69: 904–912. | PubMed | ISI | ChemPort |

Acknowledgements

We thank Deborah Ridout of the Centre for Paediatric Epidemiology and Biostatistics, Institute of Child Health, 30 Guilford Street, London WC1N 1EH, for expert statistical advice.