A Macro‐quality Evaluation of DXA Variables Using Whole Dissection, Ashing, and Computer Tomography in Pigs

Abstract

Although body composition (BC) data acquisition and ad hoc analysis are both popular and important, selecting an appropriate method or technique for accurate and/or precise assessment of individuals and/or groups remains a challenging task within various sectors of public health. Since 1950s and 1960s, with the pioneer work of Keys and Brozek ((1)), Forbes et al. ((2)), Siri ((3)), Brozek et al. ((4)), Behnke ((5)), and Durnin and Rahaman ((6)), BC almost became a scientific discipline profiling itself with the development of many methods, techniques, and equipments. Popular approaches have been criticized over the years because they are subject to measurement errors and/or violation of basic assumptions underlying their use such as hydrodensitometry ((7),(8),(9),(10),(11)) or anthropometry, e.g., skinfolds ((12),(13),(14),(15),(16)) and the universally accepted new method of choice, the dual-energy X-ray absorptiometry (DXA) ((7),(17),(18),(19),(20),(21)). Curiously, after reviewing the literature of DXA application, one cannot avoid obtaining a very controversial impression of this new method. On the other hand, we find an important number of validation and application studies (too many to refer to all) that support the DXA technique as convenient, as the criterion for % fat for lean body mass (LBM), and as a criterion for bone mineral content (BMC) ((8),(10),(11),(22),(23)). A number of authors as mentioned in Provyn et al. ((21)) suggest DXA as the gold standard for validation of other techniques essential for the measurement of BC ((24),(25),(26)). In addition to the violation of basic assumptions as referred to earlier, one needs to repeat and underline that DXA, hydrodensitometry, anthropometry, air, gas, and water displacement methods, bioelectrical impedance are all indirect in vivo techniques for measuring BC. Validation or even cross-validation in between indirect methods cannot guarantee both accuracy and reality precision. Perfect correlations and low coefficients of variation allow for good predictions and assumptions only ((18),(21)). Possibly the greatest problems with accuracy/precision in DXA are found with fat and lean tissue estimates ((27)), with its projected areal bone density ((17),(20),(28)) and with the basic confusion between overall BC terminology, e.g., fat, adipose tissue (AT), fat-free mass (FFM), LBM, lean and AT free mass (ATFM), bone mineral density (BMD), surface and volume density, BMC, ash weight, actual mineral content, and BMC, with or without soft tissue (ST) covering ((12),(21),(29)). These issues give rise to concern, but the accuracy of absorptiometry can be affected by the choice of calibrating materials. As a consequence, both absolute and relative values can differ substantially between manufacturers, between instruments and the ad hoc software used ((10),(27)). Despite the multitude of DXA validation studies and despite the related controversy of its measuring quality, it is being reaffirmed that there have been comparatively few validation experiments of accuracy, and precision of either bone or BC measurements by cadaver and/or carcass analysis. More of these validations against direct values are necessary before we can be confident about the accuracy of absorptiometry ((27)). A review of the state of the art of carcass studies related to DXA ((28)) reveals validation attempts with rhesus monkeys ((30)), mice ((31),(32)), piglets ((9),(33),(34),(35),(36),(37)), pigs ((38),(39),(40)), pig hindlegs ((21)), chickens ((41)), and with dogs and cats ((42)). The majority of these validation studies were based on chemical analysis and only a few on direct dissection comparison. Almost all studies indicated perfect correlations for all variables with DXA, but approximately half of the results of the various variables were found to be significantly different (P < 0.001 and P < 0.05). In approximately a third of these studies, DXA was suggested to be valid and accurate for all its variables, whereas two studies indicated significant differences and/or erroneous data at all levels and for all variables. However, two important statements resulting from these studies are retained: (i) dissection and direct comparison combined with bone ashing are considered the most accurate and direct validation technique ((9)) and (ii) further research with direct dissection and ashing is needed ((27)), in particular, with focus on the influence of abdominal and thoracic organs associated with dispersed gas/air pockets and internal panniculus adiposus ((21)). Because BC measurements by DXA are increasingly used in clinical practice and because dissection is the best possible direct measure, no study has been giving clarity yet about the content and meaning of “lean” as produced by DXA, different intratissue combinations, e.g., skin, muscle, viscera, and bone will be related to the DXA-lean variable. The exact knowledge of what is the content of the meaning of “lean” as measured by DXA is mandatory. In order to crossvalidate this study, we will compare fan-beam data, with both dissection and computed tomography (CT) scanning data. Twelve 6- to 18-month-old “Belgian Native” pigs were prepared for human consumption and were acquired within 2-day intervals, immediately after electroshock slaughter (six females and six castrated males, mean weight ± s.d., 39.509 ± 4.335 kg). Special permission was obtained from the Belgian Directorate General of Public Health, Safety of the Food Chain and Environment, for the transport of the carcasses and for the nonremoval of abdominal and thoracic content, which is a normal procedure in consumption matters. The carcasses were exsanguinated and decapitated between the atlas and the occipital bone. To minimize further dissection error, front and hindlegs were disarticulated distal from humeri and femora, e.g., on elbow and knee levels, respectively. The mean weight ± s.d. of the remaining carcass plus viscera was 33.051 ± 3.324 kg (whole carcass weights being taken with a digital hang scale (KERN-HUS-150K50; Kern & Sohn GmBH, Balingen, Germany) accurate to 50 g). The composition of the carcasses was studied in the following order. A QDR 4500A upgraded to Discovery Hologic DXA device (Hologic, Waltham, MA) utilizes a constant X-ray source producing fan-beam dual-energy radiation with effective dose equivalents (EDE) of 5 µSv (e.g., to situate this low radiation in terms of example: a one-way transatlantic flight produces ±80 µSv EDE and a spinal radiograph ∼700 µSv EDE) ((27)). The estimations of fat and lean mass are based on extrapolation of the ratio of ST attenuation of two X-ray energies in nonbone-containing pixels. The two X-ray energies are produced by a tungsten stationary anode X-ray tube pulsed alternately as 70 kilovolts (peak) (kVp) and 140 kVp. The software (for Windows-XP version 12.4.3) performs calculations of the differential attenuations of the two photon energies and presents data for each carcass of percentage of fat, fat mass (g), lean mass (g), bone mineral mass (g), BMD (g/cm2), and total weight. According to the manufacturer, a coefficient of variation for human BMD of 0.5% can be expected during repeated measurements. To determine the reliability of DXA measurements, each pig carcass was scanned three times consecutively without (×2) and with (×1) repositioning. From these data, the coefficient of variation for the different tissue types was calculated. The DXA equipment was calibrated daily with a spine phantom (supplied by the manufacturers) to assess stability of the measurements, but also calibrated weekly using a step phantom to allow for correction of sources of error related to, e.g., skin thickness. Whole-body scans of the pigs were taken with a CT scanner (type Philips Brilliance BZC 16; Koninklijke Philips Electronics, Eindhoven, the Netherlands) using the following settings: 120 kVp, 200 mAs, pitch 0.641, slice collimation 64 × 0.625 mm, reconstructed slice width 0.75 mm and using the Brilliance V2.3.0.16060 software. Tissues (AT, ST, and bone (B)) were classified based on Hounsfield units (HU), and their respective volumes were calculated using a maximum likelihood Gaussian mixture estimator implemented in MATLAB (The MathWorks, Natick, MA). The following optimal classification scale was employed to determine each tissue: AT: −180..−7 HU; ST: −6..+142 HU and B: +143..+3010 HU ((43),(44)). Tissue volumes were multiplied by their reference densities with AT = 0.923 g/cm3, ST = 1.04 g/cm3, and B = 1.72 g/cm3 to obtain tissue weight estimates. After the DXA measurements, the carcasses were dissected into their various components as expressed on the tissue-level system: skin, muscle, AT, viscera, and bones ((45)). Muscle included tendon, blood vessels, and nerves belonging to the ad hoc muscle. The subcutaneous, intramuscular (mostly intratendon), and intravisceral ATs were combined as one tissue. Again, blood vessels and nerves within AT were attributed to AT. Bones were carefully scraped, ligaments were added with muscle tendons to muscle tissue, and cartilage remained part of the bone tissue. Seven expert prosectors and anatomists worked simultaneously and each dissected particle was collected under cling film and kept in color-labeled, continuously covered plastic containers (12 × 10 × 10 cm) of known weight in order to minimize or eliminate evaporation ((21),(46)). Full container mass was measured during the dissection by two researchers using Mettler-Toledo digital scales (Excellence XS precision balance model 40025; Mettler-Toledo GmBH, Greifensee, Switzerland) accurate to 0.01 g. Once a bone was fully prepared, the same procedure was followed but completed with its hydrostatic weight while placed in a wire cradle suspended to the same scale allowing for the volume-based bone density (g/cm3) calculation. After the dissection and multiple weighing procedures, samples of all tissues of ∼100–150 g (min–max) were deep-frozen. Small parts were cut off and weighed in recipients of known weight before lyophilization overnight. With dried samples, the water content was measured after storing into metal cells, and fat (lipids) extracted with technical hexane using a Dionex accelerated solvent extractor. After the hexane evaporation of the extraction, total (final) lipid content was determined (weighed). Part of the dissection protocol of the 12 porcine carcasses was the total defleshing of the skeleton, including the removal of extraosseous soft tendon and ligament tissue by scraping. Cartilage and intraosseous tissue (e.g., intervertebral discs) remained intact. The whole skeleton was diamond-cut into pieces in order to fit in the ashing furnace (type Nabertherm; Nabertherm, Lilienthal, Germany). After incineration, each sample was heated using a ramped temperature protocol of 2 h to 800 °C and ashed for 8 h, as determined by prior pilot work. Before weighing on the Mettler-Toledo precision scale (accurate to 0.01 g), the ash was cooled undercover and collected in a main container. The ashing of one full porcine skeleton took between 50 and 60 h. Data are reported as mean () ± s.d. Normality of all variables was verified with a Kolmogorov–Smirnov test and all DXA, CT, and dissection data were (matrix) compared with Pearson correlation coefficients, whereas differences were verified with one-way ANOVA repeated measures. Reliability and consistency of these results were verified with intraclass correlation coefficients, and Bland–Altman ((47)) plots were used to access agreement of the direct carcass dissection data with the indirect DXA and CT estimates. All statistical tests were performed using SPSS 16.0 for Windows (SPSS, Chicago, IL), and P values of <0.05 indicated significant differences. The purpose of this study is to compare directly and indirectly obtained data of masses and densities (e.g., of whole-body bone, adipose, and nonadipose tissue) using three different technique yield information on the ad hoc terminology used in the respective methodologies. Table 1 shows an overview of terminology used per technique as applied and that are assumed to measure the same values. Although the basic assumption of equality of outcome and despite the different terminology used, knowledge of the ad hoc mass and density names will create a better understanding of the respective data acquisitions (e.g., Table 2). Table 2 combines the data acquisition of all directly obtained measures and the complete set of indirect estimates made by DXA and CT. The purpose of this Table 2 is to evaluate the predictive quality of both DXA and CT, but also to evaluate precision and accuracy between direct and indirect values. For a good understanding and despite the significance of a correlation found, this study considers r ≥0.90 as a good, r ≥0.80 as a medium, and r ≥70 as an average (mediocre) indicator of prediction confirmed or rejected by the intraclass correlation coefficient. The ANOVA statistics are considered as an indicator of precision or accuracy. Significant differences are set at P < 0.05. If not significantly different with the dissection reference, one can assume an acceptable of measurement precision. A between DXA and CT between data only because DXA or CT is considered to be a reference in this Table 2 that for almost all ST including total a majority of good correlations two correlations and two average correlations prediction expressed in % to be for the CT. Despite the majority of good for prediction related to the dissection reference, we find significant differences in accuracy for total masses and and for all nonadipose ST and and for all for the there are of acceptable precision and with and of DXA and CT data acquisition with the dissection reference of the compared tissues and tissue are in 1 and 2 with 12 Bland–Altman ((47)) The the mean from and the variation between s.d. The intraclass correlation coefficient and the Bland–Altman plots the as in Table Bland–Altman plots adipose tissue and adipose tissue free mass by dissection to dual-energy X-ray absorptiometry (DXA) and computed tomography (CT) measures with assumed BMC, bone mineral Bland–Altman plots different nonadipose tissue by dissection to dual-energy X-ray absorptiometry (DXA) measures with assumed BMC, bone mineral bone mineral The dissection tissue masses were to into and (e.g., for skin, muscle, and For AT, was made for (e.g., and (e.g., AT. For each the water content and the fat (e.g., content was determined for the respective tissues and as % of the studied mass per tissue in Table the basic that the measurement of whole-body g or or nonadipose tissue and density with different techniques using different equipment not results on the same cannot be because of underlying or approaches of the techniques and/or equipment are different can be by the on one subject of which its % or was measured with different techniques on the same e.g., with an with bioelectrical with hydrostatic weighing including the the Siri and with whole-body According to the basic the of these measures of be the On the one with an with bioelectrical impedance with Siri and with of was In other a maximum of that the of whole-body % based on hydrostatic weighing obtained density and ad hoc with the Siri that was considered as the reference method. The reality is that one is measuring different approaches with the same terminology, e.g., % whole-body fat is as the of body tissues and be considered as a chemical of the is known Keys and Brozek The use of the terms and AT has and is to and all DXA validation studies, only a few have the meaning of its variables or as DXA fat and lean against chemical fat and In the in particular, the is used to the of of a body and with to the of the body that results from the of AT be as lipid that of such as the and free from AT and also such as of and of bone and a of other Because of the confusion terms that are used both and the terms and not be The used in a be by the and the will be by to the of AT in the ATs are masses by dissection and not lipid but also the of cells, such as water and and of the of the AT and tissue viscera and between e.g., the intramuscular AT. These are known 1950s and and were in by direct data acquisition Table 1 other e.g., for the nonadipose is an and in the of the AT DXA to measure lean or as to which be expected because to measure chemical In attempts to the the of was half a of the plus the essential fat that has from 2 to for the Because of its this also has to confusion in the literature and is used as a for In addition to and LBM, the of was as a for With the of the of AT the composition of the shows of its components and differences between and females body mass is used as a reference from to and/or to is to significant error because we are with two different If we at the mean of the respective variables in Table there cannot be that both DXA and CT are producing at all adipose and nonadipose In DXA and CT not into the water content and lipid content of both its adipose and nonadipose Small variation of tissue important differences of ad hoc estimates in CT, DXA, and other body fat is calculated on the assumption that of (e.g., lean or lean is water assumed of e.g., the ratio of total body water to was confirmed in by et al. However, this assumption is subject to that the for research on the tissue water content obtained by lyophilization in human tissue studies, one can two (i) a constant % of water in be by the tissue water content within and between the tissues that and (ii) water content in AT is e.g., from to in ((21)). is confirmed in this study on with % whole-body water content from to that the by DXA and CT cannot be (e.g., with between and for skin, between and for bone, but for Because no total tissue lipid was because technical sample lipid content is expressed as % of the measured sample From sample masses being for and lipid we that lipid content of tissues is related to its ad hoc content, but the are considered one an both in and lipid The that all tissue data (e.g., in skin AT, muscle, and but in and lipid content from the and the of the as and the associated As et al. ((9)) and Provyn et al. ((21)) were the accuracy of DXA with dissection in both studies the choice of using carcasses pigs without abdominal and thoracic or hindlegs to minimize various According to et al. with DXA this can no be on the not measuring the internal will the error because of an assumption of of and ad hoc lipid et al. in and in vivo studies allowing a review and of the of of and the of Provyn et al. ((21)). even a accurate including in a constant of The data in Table within a dissection model the of constant of In the assumed ad hoc of cannot be The the of LBM, or cannot be but it that nonadipose tissues the lean tissues of the or of the and of the and lipid content of nonadipose tissue, this study has not been to what the content is of the DXA nonadipose variables, e.g., “lean” and/or not what DXA is measuring under these ad hoc compared with muscle tissue, with muscle plus skin tissue, and with muscle plus skin plus viscera and in correlations values between and a good prediction but with significant its plus is not measuring (e.g., skin muscle viscera its r = but with a significant (P < a of precision and accuracy. to et al. but in agreement with et al. a good = with no significant of its ash weight. The impression is that DXA nonadipose values are expressed as values combined with chemical study cannot what the nonadipose of DXA is but it that all the DXA components and the CT bone components are subject not only to measurement error but also to terminology error and violation of basic is known many that density in its (g/cm3) can be considered as an and of BC. The is a and not a density, the density based on which of were studied in the In a pilot study using porcine hindlegs in which DXA BMD was compared with bone covered with muscle, AT, and skin tissue, and compared with bones only it was found that DXA BMD density with In this study under whole-body one a of of DXA but with a better e.g., r = for the whole-body against r = for the The work by shows BMD to be an and indicator of bone mineral and an erroneous of relative The of their to the in carcass of this study that the DXA cannot of bone densities and bone mass because of the of ST, e.g., and lipid content The majority of and understanding of the DXA quality a number of in vivo and in significant is In terms of DXA produces and values at all levels of its the adipose and nonadipose components of DXA the ad hoc lipid content, and the nonadipose variables not into the and lean of DXA not to tissue combinations, to chemical values. cannot be determined what DXA measures. mass of ash are the two variables with a reality and DXA and CT are based on a of within of and lipid content The that DXA and BC is to be and are for the of the prosectors during for the technical of for the DXA and for the CT The authors no of